Food comes A to B,
Floating on a magic plate,
Mobile order's here

This story begins as many do, with a fortuitous posting found on Facebook Marketplace. Sometime in August or September of 2025, a gentleman in the LA area posted a picture of 3x Pudubots, autonomous waiter robots from Pudu Robotics, asking if anyone knew how to revive dead ones, with the hopes that it was just some weird battery issue. I figured I knew batteries and robots somewhat, so I offered my services and ended up with 3x robots in my garage within the week. The wife was understanding but not necessarily happy. The guy from Marketplace apparently is a restaurant entrepreneur and bought these robots at a steep discount from a closed-down restaurant that he was rehabbing. I told him I'd give it the old college try for a few weeks and either give him back a functioning robot or a pile of (organized) ewaste with the more valuable and resellable components parceled out. In return, he said I could keep one of the bots if I could get them functional again.

I first came across Pudu Robotics back in 2019-2020 when I worked for UBTech. We were also developing autonomous waiter-type robots (many companies were at the time), and Pudu had one of the few mass-produced models deployed in real world environments. The early models have your basic SLAM and obstacle avoidance functionality in its base but seemed to primarily traverse with respect to ceiling-mounted stickers. Whenever I'd see these models in the wild, my first reaction would be to look upwards to find the fiducial stickers that would dot the ceiling. The more current Pudu models don't rely on stickers anymore, though I've seen less and less of them in the wild, so maybe there's a correlation. Naomi Wu filmed a fairly detailed tour of their factory and testing facility with some more details.

The trays can be set at different levels vertically and slide into adapters that can also adjust the tray's position forward and aft. I'm a bit surprised they're only held in by a couple of M3 bolts, most of which seem to only rely on simple split spring washers for tension.

The drivetrain is built around a simple differential drive with hub-motors, but also 4 caster wheels, with the rear pair spring-loaded relative to the drive motors. This ensures that the system doesn't get high-centered but still maintains as many points of contact on the ground as possible for consistent traction. For passive stability, I actually really prefer the drivetrain topology of the Fetch AMRs, which spring-loads the drive motors against the ground relative to statically fixed casters, but I guess then the drive preload wouldn't scale w/ respect to the payload. The front also appears to be reinforced with what I can only describe as mini-skid plates to reinforce the lower steel place and the plastic side enclosures.

You can see from the wear/tear and general grime buildup that the customer seems to have run this bot pretty aggressively. Does that mean it worked well? Or does it the customer had a really dirty restaurant? Not sure. To me, this also calls out the need for regular maintenance, something the typical restaurant customer probably wouldn't think too much about. Hell, it's probably not something a typical factory AGV/AMR customer would think about, as a factory floor is likely way more consistent and cleaner than your neighborhood restaurant (not throwing shade at restaurants).

Another crazy thing about the dirt/grime buildup: you can see through various shells/covers that the designers attempted to seal off the insides from the outer environment, but the gap around the drive motors still allowed dirt to enter the interior of the main lower body. In fact, when I popped open the outer covers, each bot had about a standard dustpan's worth of random junk and trash collected inside. Vacuum your bots' insides, people.

The outer plastic shells interlock somewhat with the metal base, and they pop out to expose a hodgepodge of mismatched sub-assemblies:

• The bottom steel plate serves as the primary structural component to which the battery pack and drivetrain subassemblies are attached
• A bent sheet metal substructure holds the audio subassembly
• A physical key serves as the main power-on relay near the charger, which relies on a simple barrel connector
• A combo of CNC aluminum brackets and 3d-printed mounts for the remaining RGBD and lidar sensors
• Custom CNC aluminum blocks hold up a central stainless steel cross-brace to which the primary side columns are bolted
• Aluminum bracing structure that could be simple bars if not for CNC cutouts to accommodate zip ties to hold down cabling
• Bent sheet-metal housing to protect the primary battery, and it's attached to the bottom base plate with obnoxiously long M4 standoffs
• Brushed aluminum extrusions serving as posts from the bottom mobile base to the compute structure and UI panels on top
• A COTS LED strip is taped to the interior of the outer plastic shell

It's quite a weird but clearly very functional mashup of engineering and design practices throughout the build for (to me) reasons unknown. Maybe the various subassemblies of this bot were copy/pasted from other prototypes. Maybe different groups/teams worked on each subassembly while woefully siloed from each other. Maybe quick/inefficient design changes to a formerly more cohesive and consistent design were made on the fly in a haphazard way as hot-fixes for new problems discovered during testing. Nothing in the system is anything I believe any engineer would say is categorically "wrong," and history has shown that this design got plenty of things functionally right, but I don't know that anyone would use this model as a reference for how to design these mobile bases optimally.

Here is what I assumed (going off of vibes and a similar design process we scrambled through at UBTech) happened:

• Designers started with a barebones, super simple two-wheel drive base with casters
• Suspension was over-designed to handle an ambitious worst-case system weight because no one really knew what the final bot would actually do
• Initial drive testing heavily utilized the lidar and top-mounted camera. Realsense RGBD cameras weren't added in until much later (hence the weird stack of disparate mounts)
• For the required stiffness of such a tall setup, it was decided that custom aluminum extrusions were necessary, and subsequent design changes were made to accommodate these extrusions
• The original frame was probably mostly 8020, but the primary cross-beam was swapped to steel for increased stiffness
• Someone decided audio would be important, so COTS parts were cobbled together on top of a bent sheet metal frame
• 3D-printed mounts were added to accommodate last-minute cosmetic changes to the outer shell
• Cabling started becoming a mess, and last minute modifications had to be made to the structure where adhesive pads and anchors wouldn't work

I'm sure plenty of other engineers have a different viewpoint, but I'm so fucking impressed this got released as a product and performed so well for so long. The insides were literally held together by zip ties and glue, but can we take a second to appreciate that those still held? There's no billet machined aluminum body frame, no custom modular harnesses, no consistent choice of connectors, standoffs and foam padding that did not match the reference images in any manual; and yet, it worked just fine.

The base is pretty awesome. It's a solid, proven frame with a simple, low-maintenance drivetrain and all the key sensors you would need for SLAM. I'm sure it's not the first robot base to repurpose hoverboard-style hubmotors and certainly not the last, but it's more than good enough for your typical robot mobility needs. Everything shown above could easily be driven by your own set of basic brushless ESCs and a cheap SBC.

As for the battery that I was tasked to fix this entire time, it was a discontinued 7s battery with a custom BMS board shrink-wrapped in plastic and covered with foam. It had separate charge and discharge connectors, but the BMS seemed to only be powered by the battery, isolated from the charging ports. This meant that once the batteries were discharged too low (which all of these were), I had to expose the direct leads to the battery to get them to charge at all. When I reached out to the distributor for a replacement battery, I was quoted 1-1.5k USD.

As for the rest of the bots' functionality, I found out that this Pudu model wouldn't function without the ceiling mounted fiducial markers, which not only used a proprietary pattern but also needed to be IR reflective. Wouldn't you know it, these markers aren't made anymore, and the company refused to disclose any additional details. I think the more recent models use just SLAM without markers, but I remain puzzled why integrators nowadays continue to shy away from this one-time setup cost that makes operation so much simpler.

If you stumble across one of these guys out of commission, it's likely to stay out of commision, unfortunately, but the structure is pretty neat if you're okay with replacing the battery and control boards with your own homebrew mix. After these disassembly shenanigans, my Facebook Marketplace benefactor let me keep one of the bots, so now I have a stripped-down base (seen above) that maybe I can remix in the future.

Fairy's magic smoke,
Contained in this meager box,
Chaos potential

tldr; If you enjoy putting sketchy components with explosion-potential inside of a sketchier DIY box and wiring everything through questionable boards bought for cheap off Aliexpress, then this is the intro blog post for you!

My last startup gig focused on the autonomous dismantling and testing of EOL electric vehicle batteries for potential re-use as second-life storage batteries. The idea was that batteries from old electric vehicles unsuitable for the road are probably still quite good for storage applications, so there should be a business case for repurposing those batteries instead of just disposing them or going through the costly process of recycling them. It wasn't a crazy revolutionary idea: Posh Robotics (now Posh Energy) was a YC startup that pitched that idea a year or so before us, Redwood Materials demonstrated a second-life BESS at their recycling plant many years ago, BigBattery in nearby Chatsworth had been running a seemingly profitable business on this concept for years, and B2U had been (and maybe still are) shoving old Nissan Leaf packs into storage containers for grid storage. Unfortunately, the startup died before we could get to building second-life batteries from salvaged cells (we had focused on autonomous dismantling and testing), but as part of our liquidation process, I ended up with a bunch of sketchy EV battery modules and cells in my house, so I (very slowly) built some off-grid power banks out of them.

A Warning on Battery Safety

I'd like to caution any reader who would want to attempt similar projects to read as much about battery safety as possible, especially if you intend to deal with high-voltage and high-current applications. At the very least, refer to ElectroBOOM's lovely video on electricity hazards. I think we take for granted how typical electronic devices are certified/tested to ensure that they don't blow up in our faces, and while I'm not saying that every battery cell you touch could end up being one of those hoverboard lipo fire disasters, the increased power density from these cells isn't something to take lightly. Shout-out to my former office ops manager K and my former battery engineer R for yelling at me to stay safe whenever I started poking battery packs and agreeing to hit me with a wooden stick if they ever saw me convulsing when touching a battery.

Here were some (better safe than sorry) practices I followed whenever I handled EV cells/modules:

• Wore rubber/electrician's gloves rated for >3000V whenever possible, especially during initial pack disassembly when I didn't know the state of the battery
• Tried to do anything I could one-handed so that in the worst case scenario, I hopefully wouldn't have the current run through my heart
• Taped over any terminals/busbars if I wasn't working on them
• Kept all loose fasteners in containers away from the modules/cells after removal
• Made sure to only cut cables when disconnected, and if not, only cut a single wire at a time to avoid shorts through the cutters
• Kept pack ventilated and near an open door for both quick ejections (if the pack was small) and quick escapes (if the pack was too big)
• Put a fire blanket over the battery pack when we weren't working on it
• Checked for any bulging cells and leaks at earliest opportunity

Second-life Battery Components

There are quite a few kits ([1][2]) already on the market and forums ([1][2]) to build second-life powerwalls or packs, mostly for off-grid or backup applications. I was not ambitious or brave enough to try for a powerwall-scale project, so I mostly spec'ed out handheld and suitcase-sized power banks with inverters that one may take camping or a construction/work site. The most basic set of components are pretty straightforward:

• salvaged cells/modules
• storage container or box for the pack [1][2]
• high-current busbars and connectors for the output (sometimes salvaged with the cells/modules)
• connectors and wires to cell level so that pack can be properly balanced (sometimes salvaged with the cells/modules)
• inverter matching the voltage range of the target pack [1][2][3]
• BMS matching the cell count and voltage range of the target pack [1][2][3]
• high-current power switch
• battery capacity gauge (usually just approximating capacity from voltage) [1][2]
• connector for charging port (I just used a standard barrel connector)
• ACDC charger for the target power bank voltage
• power cables matching the max expected current draw

Some nice-to-haves:

• zip-ties and zip-tie mounts with double-sided table to help with cable management
• kapton tape to isolate terminals and also help with cable management
• independent busbar blocks to help with final power output
• battery lugs for high current output
• screw terminals and crimps for early testing/validation
• anti-reverse diodes for (safer?) charging

I also ended up getting a hydraulic crimper to help with connecting/splicing the thicker gauge power cables. The internet as a whole seems mostly in agreement that crimping is more secure than soldering thick cable connections, especially if your home soldering setup is power-limited (as most likely are for these gauge wires) and would risk cold solder joints.

My Junk Powerbank Attempts

NiMH Packs

NiMH cells from old Toyota Prius hybrids were a really interesting use-case to start testing. From my time at the startup, these appeared to be the most readily available cells available on the salvage and reuse market. A lot of small mom-and-pop garages (at least in the LA area) where offering hybrid pack repair and replacement services for the Toyota Prius, and I bought many of our test packs from flip-flop wearing mechanics off of Facebook Marketplace.

The older NiMH chemistry for these cells (as opposed to NMC in most other EV packs) made battery management a bit difficult and limited pack configurations. Apparently it's not advised to use these cells in parallel configurations [1][2] due to how the cells behave as they approach full charge. Slow-charging is always recommended, and it looked like a proper charger should be measuring the pack temperature for safety purposes as it reaches full charge. The only cost-effective NiMH chargers I found were a set of USB-C boards from Eletechsup (resold through Aliexpress), and I could only use the 6S or 12S options due to the module design in the Prius (you don't have access to individual NiMH cells). I have no clue if these charging boards are safe or not, but I can attest that no cells bulged or expanded during the course of my random machinations.

Mechanically, the cells needed to be clamped. Despite their appearance (and really, the appearance of most EV cells), these are pouch cells (as opposed to cylindrical or prismatic), so they're prone to swelling if overcharged and not mechanically constrained. This setup is more common for second-life power banks that use Nissan Leaf cells, but I also used 2 aluminum plates and some threaded rod with locknuts to group the cells into a block. For busbars, I could re-use the default copper plates in the original pack.

For a typical module with 6 cells (nominal 7.2V) and 6.5Ah capacity, you basically get the performance of the battery to an average power drill. In one test pack, I 3D-printed some ends to cover the terminals and also house the charger board and some buck converters to use as an oversized phone power bank. In another test pack, I added a small 12V car inverter, but that tended to trigger an over-voltage alarm when the cells were fully-charged (12S NiMH gives 14.4V nominal but up to max 16.8V-ish). While I started with a 7-module pack configuration (50.4V nominal), I quickly abandoned that due to my lack of comfort and understanding with NiMH charging requirements as mentioned above.

One neat thing about the sketchy NiMH USB-C charger boards is that I got to use equally sketchy Aliexpress 5V solar panels to charge them. I did this mostly because I happened to have some unused flexible panels that were approximately the size of the NiMH module. If you go down this route, make sure there's either a diode or a solar charger sitting between the panel and the charger board, otherwise the panel will actually discharge the battery at night (I had no clue this was a thing). I'm happy to report that this approach successfully stole free energy from the sun without initiating any thermal rapid-disassembly events.

Ultimately, I never figured out a form factor that I liked, and the inability to comfortable configure a second-life pack with modules in parallel led me to just return/resell these modules back to the Facebook Marketplace ecosystem, for which there's still surprisingly active demand.

Rewired Standalone Toyota RAV4 Hybrid Module

This will sound weird: I've taken apart several post-2020 Toyota RAV4 Hybrid battery packs, I really like the form factor and layout of their cells, and as I'll get into, I built multiple power bank packs with these cells, but I still can't find any online documentation on these cells whatsoever. I think it's because most RAV4 Hybrids still use the NiMH cell chemistry, and the newest RAV4's are predominantly plug-in hybrids (PHEV), which use a totally different and much larger battery pack. Assuming a total 1.6-1.9kWh pack capacity, 70 cells, and a reported pack voltage of 259V, each cell has a capacity between 6.1Ah and 7.3Ah. The size difference gives you a pretty good visualization of the advances in power density between the older NiMH chemistry and lithium-ion.

To get these RAV4 cells out, you have to cut apart the aluminum frame that's clamping them together, so I wanted to first see if I could just re-purpose and re-wire a module as-is, re-using as much of the existing wiring harness as possible. Each module has 35 cells, so that gave me the option of 5S (18.5V) or 7S (25.9V) configurations unless I just wanted to leave some unused cells hanging out. I decided to go with the 7S configuration so that I could ideally use it for 24V devices somewhat safely. It's still an atypical voltage range, but it seemed more useful than 5S.

Rewiring the module from 35S1P to 7S5P was a little hairy. I removed all the jumper busbars between adjacent cells and then reinstalled them such that it was effectively 5 sets of 7S1P groupings. You can't actually do 7 sets of 1S5P groupings because of the way the terminals alternate to accommodate the original module layout.  I probed, re-probed, and re-probed the existing harness endpoints to determine which wires should be soldered together. I used to have a burnt pair of wire cutters to remind what happens when I probe incorrectly and accidentally short the wrong balance leads.  For charge balancing/monitoring, I used a JKBMS Smart BMS board with bluetooth, and the final output was wired to a beefy XT90 connector. This particular BMS had a separate on-off switch, so I didn't bother wiring a main battery switch, but given the awkward size of this thing, I'd probably only use this power a small, yet-to-be-designed, robotic cart of some sort in the future.

Small Case Powerbank from PHEV RAV4 Cells

For this power bank, I shoved a 3S2P configuration into a gun case from Harbor Freight, along with a 400W inverter. Unlike the NiMH case, 3S lithium-ion configurations pair pretty well with 12V subsystems without inadvertently triggering any low/high-voltage alarms. The case was sized very conveniently for the RAV4 cells, allowing me to reuse the standard plastic separators from the original module, which then meant I could re-use the busbars as well. I used a dremel to create access ports for both the inverter and the voltage indicator. The extra printed bracket around the inverter was meant to fill the remaining gap and keep some compressive force on the cells to prevent swelling. There just wasn't enough space for a 3S3P configuration, but this little cases provides about 150 Wh.

This pack used a much simpler 3S BMS board (HX-3S-FL25A) that is typically used for powered handheld drills. Despite the specs saying that the standby current draw is just 30uA, I found that leaving the battery switch on (but without anything other loads attached) drained the battery within about a week. Leaving the battery disconnected maintains the battery capacity as expected. Part of me wonders if that's an indication of poor cell health or a faulty BMS board, but at this point, I really don't know.

Medium Case Powerbank from PHEV RAV4 Cells

I lucked out again with this Bauer container box from Harbor Freight when I tried building a bigger version of the previous pack. The RAV4 cells also fit really nicely inside, and I was able to pack in a 6S6P (36 total cells, 22.2V nominal) configuration with a 600W inverter. Similar to the last build, I sized some 3d-printed adapters and spacers to keep everything as tightly constrained against the sides of the box as I could. I also tried to be less ass at wiring and tried to route the cabling around the hinge so that you can actually open the box all the way without disconnecting anything if you needed to do maintenance.

For electronics, I used the same power switch as the smaller power bank but opted for a 4-8s Daly BMS with bluetooth. The hardware was way sleeker and nicer compared to JKBMS but the app was a nightmare to use. I did however appreciate that all of these fancier BMS's have built-in temperature sensors to catch thermal runaway events. With a theoretical capacity of 850-ish Wh, this pack is more comparable to a typical camping power station from Ecoflow or Anker, just jankier and less reliable and with fewer ports, but I guess we're getting somewhere.

While I was building this, nearby folks listed some used solar panels for resale on Marketplace, so I of course fell down that rabbit hole for a bit too. I already had some questionable 5V Aliexpress panels, so I hooked those up first to a USBC-to-6S charger board (kind of crazy that these exist) and got a whopping 2W charge rate during sunny conditions after all the efficiency losses. The used solar panel reportedly was rated for 285W, and in combination with a PWM Renogy solar charger, I achieved a much more useful 160W charging rate. Unfortunately, I didn't keep that solar panel much longer, as I live in a condo and couldn't really justify the space to keep it around.

Rewired Hyundai Ioniq Powerbank

At my last startup, I really liked the Ioniq pack design, as well as their modules' prospects for reuse. The Koreans, much like the Germans, really know how to package their cells and design for modularity. No messy foam or trapped cells. All cable harnessing are neatly routed, and in this case, the 6s2p modules nestle nicely for both tight packaging and improved structure rigidity. Assuming 72.6kWh for the whole pack, running at 800-ishV, each module should have 2.42-ish kWh, which is pretty fantastic.

I remember being a little fast and haphazard when I disassembled the pack, so it wasn't too big of a surprise when one of the balance leads wasn't reading the voltage it should. Luckily, the modules are fairly easy to unscrew, and after I popped out the ends and did some probing, I found that one of the fuses had blown. Luckily, I conveniently still had access to screw terminals leading to each pouch, so I just replaced the breakout board with direct wire connections. Most of the pouches I've seen are spot-welded to the terminals, so unless you can work directly with the existing terminals or have your own spot-welder, you may be shit out of luck if it's damaged.

The test/build procedure was similar to that for the RAV4 module: I used some terminal blocks temporarily so that I could probe the balance wire pinout and double-check the BMS (another Daly BMS) operation. The end-result was a 12s2p configuration that I'd like combine with some 12V and 24V buck converters to power a future rover project.

A Solar and Inverter Side Quest

Like I mentioned, I started haphazardly trying to add solar as I was throwing together these packs, because hey, who doesn't like free energy? I don't have any novel insights to add (plenty of other folks have done a much better job of that), but there were some (now-obvious) things I learned along the way:

• You really need fairly large solar panels if you want to do any meaningful charging. Smaller portable panels are limited in voltage/power, will probably break far more easily, and have questionable specs.
• Whatever panel size you think you need for your application, you'll definitely need larger. Think about the panel size that you see in warehouse parking lots just to power a night light and security camera.
• Direct power output from a solar panel is very irregular, and an errant shadow is usually just waiting around the corner to disappoint you
• Cheaper PWM solar chargers ([1][2]) will tank the performance of your already cheap solar panels, and they're typically made for just a limited set (often just 12V and 24V) of pack voltages. However, it's hard to justify the cost of a more legit MPPT solar charger for DIY power banks. Apparently a DC-DC converter can sometimes be a more cost-effective choice.
• Solar panels need to output more than your pack's max voltage, and ideally its minimum VoC should exceed what you need to charge the pack. A "12V" solar panel would produce voltages in the 15-17V range before it gets regulated by a solar charger.
• Typical AC inverters also only work for limited voltage ranges (usually around 12V, 24V, and 48V), so don't expect a generic inverter work for any battery pack voltage you configure.
• Microinverters (well, with maybe one or two very expensive exceptions) don't work standalone, disconnected from the grid. They are not magical combo solar charger and inverter in one.

Thoughts and Notes on Economics of 2nd-life Packs

These thoughts are my own and don't reflect those of my former employer or collaborators or VCs

tldr; Aside from some grid storage cases, I don't see where second-life packs are viable unless OEMs drastically change how they design their packs

The original promise of second-life batteries relied on a couple of factors holding true:

• EV packs unsuitable for driving could still be useful for storage applications
• An EV pack may be damaged or bad as a whole but individual cells/modules could still be functional
• Batteries (due largely to their constituent materials) retain enough high value such that the cost of dismantling, testing, and refurbishing them remains sufficiently lower than producing a new battery. (I'd argue that the funding of behemoths like Redwood Materials gave this hypothesis credibility)

On the materials-side, I think the introduction of new lower-cost chemistries like LFP (LiFePO) was the primary killer of second-life ambitions. I'm going to ignore the effects of environmental policy shifts and changes in subsidies, since the political climate has been less than stable or rational the past decade, I have no clue how that game is played or measured, and I personally don't think those things should be considered when evaluating a market's viability. As with cars, there's not really any motivation for a customer to buy a used model when a brand new model performs better and isn't all that much more expensive. However, even if LFP batteries never gained traction, the second-life concept had a lot of challenges:

• Measuring battery cell health is not trivial. You generally want to run a cell through full discharge/charge cycles to determine its true capacity and performance. That typically gets done at the factory level during production. Reselling a literal lemon would open up a whole can of liability worms. I had and still have no clue as to the health of the cells I used in my DIY packs
• The EV pack form factor and configuration aren't great for other applications. The voltage (up to 800V nowadays) is much higher than what is safe or useful for grid storage applications, and the packing isn't as dense as it could be. As I showed in the RAV4 module example, you may need to do a good amount of rewiring to convert the module to a more conventional voltage range, assuming you want to keep the existing module structure and not break it down further. However, the more you break down the pack to cells, the less you can re-use the other components like harnessing and busbars and enclosures.  
• There are lots of different cell formats, so you can't easily mix and match across different EV packs. Most packs don't use a standard cell like the 18560 and even those that do have made their packs increasingly difficult to dismantle down to that level.
• The set of battery cells aren't the only components you'd need in a second-life battery pack. Any builder of second-life battery packs would also need to provide their own enclosure, BMS, harnessing, cooling (where necessary), connectors, etc.
• EV packs are heavy. Every problem has or is a logistics problem. The cost of shipping these packs (safely) can be substantial, both for the initial acquisition of the pack and then also for the shipping of the final product. Even if you're intent on re-packaging the cells, you'd still need to be responsible for shipping the weight of the rest of the original pack.
• Dismantling is hard, and while packs may be designed for assembly, no one cares all that much to design for disassembly. One thing more expensive than labor is skilled labor. As with obsolete electronics, it may very well be that shredding will end up being the preferred and most economical approach.

Successes in second-life battery storage have predominantly reused EV packs directly and without teardowns. This limits each battery storage solution to a specific model and requires someway to hack or bypass the existing pack BMS, but it also doesn't require the vendor to design/produce new auxiliary pack components. Still, this makes the most sense when the EOL packs are free or near-free, and also when they're still safe to run. Neither case should be presumed to be a common opportunity waiting to be had.

In the dark warehouse,
There's a buzz buzz in the air,
That bee high voltage

Unfortunately, this post is more retro than update. For most of 2024, I spent a lot of time tearing apart electric vehicle battery packs (somewhat safely). They're pretty cool, and since our work dealt with the full range of pack designs, from older hybrids to plug-in electric (PHEV) to full-electric, I felt I also got a decent view of the progressing design trends over the years. For posterity's sake, I wanted to document some of my observations here.

If you're looking for a technical deep dive into battery pack design, there are much better resources out there. In particular, I'd recommend the following channels on Youtube:

• Munro Live
• Weber Auto
• Gruber Motor Company
• PowerBastro

Maybe it's a sign of the times. Maybe it's just that video's a better medium for this sort of insight, but I found teardown videos from the sources above to be way more informative and helpful than any manual or research paper as I prepped for hands-on dismantling work.

All in all, the innards are pretty neat:

Pack Structure/Enclosure

Hybrids and smaller plug-in hybrids have batteries with primarily aluminum sheet-metal enclosures, whereas you'll see significantly beefier steel enclosures or even billet aluminum frames in larger full-EVs. While there's been more talk and proposals of structural battery packs, which are perhaps more prevalent for cell-to-pack architectures (more on that later), most packs appear to be designed independently of the overall car structure, and they either slot into an interior compartment or bolt onto the underbelly of the vehicle.

Smaller battery packs are not typically as well-protected against egress or the elements. The Toyota hybrid packs I worked on weren't sealed, and corrosion was a common defect on the interior of the pack, which did not bode well for the various busbars connecting the cells. It seemed like the covers were mostly to prevent technicians from shoving their hands into the cells and wiring. Some plug-in EV packs at least had gaskets and closed up egregious openings. Full-sized EVs took sealing a lot more seriously, and some (annoyingly) apply a fat bead adhesive around the outer boundary of the top lid to both minimize fastener account and really ensure nothings inadvertently gets inside. From what I saw, all top lids, regardless of pack size, were singular stamped metal panels, with rigidity coming almost entirely from embossing.

Speaking of fasteners, most of the battery packs, with the notable exception of Tesla, could be opened with a 10mm socket, which was greatly appreciated. Tesla, on the other hand, not only requires a custom pentalobe socket, but also several different socket and hex bit sizes, not to mention an abrasive cutter to get through the rest of the lid. I can't say that Tesla's packs have embarrassingly low accessibility and reparability in the interest of corporate profit, but I'm also saying I'm not ruling that out.

Modules/Cells

The typical battery pack has cell-to-module-to-pack structure, where the module is a grouping of cells, presumably to improve both assembly and overall structure. While some of the modules may wire their cells in parallel for increased capacity, all modules I saw were wired together in series to produce the pack's final output voltage. In the case of NiMH modules in the Toyota Prius packs, this apparently is especially important as parallel charging can be trickier than with Lithium-ion. Breaking up the pack into modules also makes it easier to disconnect the pack into discrete chunks with significantly smaller and safer voltage.

I worked primarily with older models, so most of the modules were comprised of pouch-type cells, which don't have a hard outer casing. This means that cells had to be clamped/packaged within a rigid outer casing. Many of these support structures are welded in place, so you may be able to get the cells out, but that's a one-way, one-time operation. Other cell types include prismatic and cylindrical, the latter of which the public is probably more familiar from consumer electronics.

Cell design has perhaps gotten more attention in recent years with more exotic/robust designs like the BYD Blade and Tesla's 4680, all of which aim to have less susceptibility to thermal runaway, higher energy capacity, better cooling, better structural rigidity/toughness. On that last point, it seems that the latest battery packs are shifting towards a module-less design, where a combination of the latest cells and foam provides the necessary structure. Stateside, this is apparent in teardowns of more recent Rivian and Tesla packs. While the foam reduces initial assembly complexity (and I assume cost), it also substantially complicates the repair and replacement of individual cells.

Safety/Contactors/Busbars

Orange you glad battery safety is a requirement? Across the board, all high voltage lines are terminals are marked super clearly with either orange covers or cable insulation. It's super obvious where the more dangerous areas in a pack are. For the overall pack, it should be noted that the big main contactor relay needs to be triggered for current to flow out of the pack, so a pack disconnected from the main body of the car shouldn't be energized at the main terminals. However, when you open the outer enclosure, all bets are off, so please be safe.

As I hadn't been too familiar with high-power applications prior to this gig, I didn't understand or appreciate how contactors worked, but simple switches don't quite work for the voltage and current levels in EV packs. Even the smallest packs have a precharge circuit to avoid high current surges and arcing. Some of the conductors in these switches are quite thicc (c is for charge), as they should be.

High voltage lines running between modules are fairly substantial as well. I saw some standed wire cables in older battery pack models, but most utilize chonky busbars: bent/twisted conductive bars. These are usually bolted down onto the module terminals. Being rigid, these have the added benefit of avoiding entanglement or snagging during assembly/disassembly. Most interestingly (to me), newer packs have replaced the copper in these busbars for aluminum, which I assume reduces cost and helps protect more against corrosion.

Wiring/Cabling

There's a shit ton more wires in a battery pack than I would've figured. With a whole bunch of cells, whether in series or parallel, keeping them balanced is critical to maximize longevity. Not only is there a connection to every single cell for voltage balancing, there are also intermittent sensors to prevent temperature overload and handle cooling. Smaller packs seem to route all wires to a central BMS (battery management system), which results in quite a substantial harness running across the entire pack. Larger packs are more likely to have dedicated BMS's per module, each of which then connected to a central computer.

Connectors for cabling are no joke. As you can imagine, they're designed to stay connected, not readily removed or swapped. All connectors have some sort of locking component, either compliant clips or latches. They're onerous enough that some companies make specialized maintenance pliers just to disconnect these connectors. As mentioned above, high voltage connectors are often bolted down directly to the modules.

Cells within a module are usually connected via spot-welded tabs, which means destructive dismantling if the cells are to be extracted and upcycled. Older and smaller packs may utilize discrete busbar plates between adjacent cells. It does however seem like screw terminals at the cell level are on there way out as they become more integrated via increased foam usage and fastener count reduction.

Heating/Cooling

Batteries work best within a certain temperature range, as they lose capacity/range on the lower end of the temperature range but also need to avoid excessively high temperatures when the vehicles utilizes high current. Hybrids can seemingly get away with simple forced air cooling, and there are usually some simple plastic duct covers that run along the length of the cell rows. Older full-electric models typically have channels with coolant/fluid running along the base of the pack, in contact with the modules/cells. The new Tesla models go as far as integrating cooling channels within the modules, interwoven between the cells. I should note here that draining the coolant out of the Tesla module channels was a total pain, and we got refused pickup by a shipping company the first time we didn't do it correctly. To assist in thermal transfer, there's often a paste of sorts underneath the modules.

Much of the attention in the battery space has been focused on updates to chemistry and composition, but there's substantial design work on the interior layout and structure as well. Unfortunately, while the product costs (per kWh) have continued to plummet, it may come at the cost of a less repairable and accessible pack, which may put consumers in a bind later on down the road.

Trigger a constant,
Reciprocate a pulse,
Pew-pause-pew-pause-pew

Kept getting ads for an electric, automatic water gun a while back on Temu/Aliexpress. For $10, it was hard to turn down if only for the lulz. Also kept getting yelled at by my dentist and hygienist for not flossing properly, so I got a cheap waterpick off of Amazon as well. I just took both apart, since they're pretty much the same, right?

Aliexpress Watergun

Surprisingly low-tech with wiring jankier than what I could've ever dreamed. Pulling the trigger like normal shoots out pulses of water, and pushing it the opposite direction allows the gun to siphone water through the barrel area and refill the internal tank. Electronics-wise, there's just a microswitch at either end of the sliding trigger, each one driving a separate motor subsystem: a pump that pulls in the water, and a reciprocating piston similar to what you'd find in airsoft guns to squirt water. No boards or any other circuitry, not even for charging. You actually need to unscrew/unclip an outer covering, detach the battery, and manually charge it via USB.

With switches, multiple motors, and a single battery, you'd expect at least a few split or spliced wires, and there were, but there was nary a piece of heat shrink or drop of solder in sight. The wires are just stripped and wrapped a couple of times around each other. This is some next-fucking-level tomfoolery. Or cost-saving genius. I still need some time to process.

Away from the exposed wiring, the tubing and sealing look pretty good. The entirety of the pump looked to be enclosed, and though the motor body of the output mechanism is exposed (presumably for cooling), the rest appear well protected against ingress with adhesives, seals, and grease. As long as only the business end of the water gun is submerged, I doubt water would get to anything it shouldn't, but it's still a bit unnerving that something battery powered plays so fast and loose with the internals. On the outside, there's a flimsy film and some minor grease covering each of the exposed screws, which I suppose is better than nothing.

Looking at the firing mechanism, I couldn't quite figure out how, if at all, the output line was primed, and if there were any other air holes elsewhere to help with the overall circulation. There's a heavy duty compression spring missing in the pictures above, as the gear train draws back the plunger until the partially-geared final spur gear runs out of travel. That action draws in water from the (presumably non-empty) reservoir. The spring should release energy much faster than the geared motor, which can reload the fluid much more slowly.

Waterpick

Compared to the discount electric water gun, the home waterpick is a lot of the same: a motorized pump draws water out of a refillable reservoir and then forces it through a much smaller orifice at a regular rate. Unsurprisingly, the assembly is much cleaner, miniaturized, and it doesn't elicit concern when I shove it in my mouth. No metal fasteners or anything else that could corrode are exposed on the outside in any way, seals and gaskets are found in any of the mechanical interfaces/separations, and the core fluid connections are tightened down via fasteners, not just reliant on press-fits.

The main outer body is a singular piece that appears to have been ultrasonic welded together out of smaller pieces. The internals sort of slide in and get sandwiched by the bottom base and top cap. There's a large number and selection of itty bitty fasteners. I'm not sure if this suggests a large number of 3rd party vendors for each of the subcomponents or poor DFM planning or something else. I've wondered before why manufacturers still go with a large number of small fasteners for mechanisms that aren't meant to ever be serviced, and now I wonder if it's because mechanical fasteners are still most reliable from an assembly workflow perspective. I assume finagling with adhesives, especially when dispensed via manual, handheld tooling, just invites more mess and inconsistency.

I found the little clip that holds onto the interchangeable nozzles to be pretty cool. Each nozzle has a little radial notch near the base, which gets clamped by a compliance c-clamp like wire around the output head at the top of the waterpick. The outer tool-release button wedges the metal wire open when depressed. I suppose it also helps that most nozzles are mostly press-fit onto the waterpick anyway even if the wire wasn't there.

Unlike the water gun, this waterpick was blessed with an actual electronics board with BMS to charge the battery and a way to toggle between different pulsing speeds. The soldering between the board and the motor terminals is part of the board's final attachment to the main assembly body.

The pump is a re-packaged/re-configured diaphragm pump that draws from the main tube and simply spurts the water out the nozzle while minimizing the overall packaging constraints. A bevel gear runs an eccentric cam driving the primary piston which runs a series of two one-way diaphragm valves to direct water in the appropriate direction. The piston itself interfaces with a thin stainless steel tube insert within the plastic housing. There's a tiny compression spring at the output that I assume is meant to preload the last diaphragm flap shut when not in use and avoid leaks, though I'm not totally sure.

The part geometries, especially the diaphragm housing, are non-trivial, though I wonder if other low-cost waterpicks (I didn't exactly buy a brand-name implement) share a common internal body structure, much like how basic 12V diaphragm pumps are all essentially identical. I wouldn't know without taking apart more of these from other vendors, but I'm curious just how else designers manage to keep the overall form factor minimal in reformatting the standard pump.

These wheels go around,
Not dry, not wet, let's say moist,
Grime mobility

A year or so ago, I bought a wet mop attachment for a Dyson vacuum while perusing Aliexpress late one night, as one does. In typical "I-didn't-read-properly" fashion, it was incompatible with our Dyson at home, and I was stuck with this random module on my shelf. A little more time (unfortunately) freed up in my schedule in the past few weeks, so I'm back to taking random stuff apart. There's a series of posts I want to do at some point on the number of add-ons and simple electromechanical devices we use in our day-to-day that's basically just a motor-on-a-stick, and how minor variations on that topic and produce so many different uses. I was a little surprised at some of the design choices made in this relatively simple attachment.

Any Dyson attachment needs to accommodate the vacuum hose and an optional, convenient 12V plug for power. I suppose this interface gets updated between models to move more product and confuse late-night Aliexpress browsers, but it's pretty nifty that we can have a swappable, motorized tool-end. The wet mop attachment is supposed to slowly drip water or cleaning solution in front of the dual rotating pads that funnels grime and dirt into the vacuum inlet port.

With these sort of mechanisms, I'm generally interested in how they deal with ingress and protecting the electronics and moving bits. For cheaper modules like this, I'm getting the sense that the answer is they really don't. There's not much in the way of seals around the rotational joints or even the main clamshell enclosure. To be fair, this thing's not meant to deal with significant amounts of water or get submerged. The water's only meant to help pick up dust and grime better than a totally dry vac. The grease lubricating the gears should be enough to keep the minor amount of moisture from corroding or damaging anything significantly.

The main pivot or swivel joint around the hose uses snap-fit plugs. The hose prevents any sort of rotary shaft that runs through the middle, and I guess two snap-fit pieces are significantly easier to install than anything w/ fasteners. I'm surprised this joint can hold up to use and abuse without bearings or bushings or at least something stronger than plastic, but maybe I shouldn't be, since the rotary range of motion should be fairly minimal, and you wouldn't expect any crazy side loads. The plugs were not great for disassembly, as the tabs were super-fragile and easily broke off as I tried to depress them w/ pliers while yanking them out.

Our increasing acceptance of accelerated obsolescence in the interest of cheaper products w/ "good-enough" performance has me constantly on the lookout for manufacturing cost-cutting and design compromises. Let's be real: I don't think anyone who buys this module is expecting to be able to take it apart and service it at home if necessary. They'd probably just junk it once it starts to struggle or if a critical component breaks. The water tank is irreversibly welded shut and traps the fastener holding in the air-release valve. The snap-fit plugs at the pivot joint further reinforces a disregard for future disassembly, and yet you see pairs of screws above for brackets or covers that retain the power wiring. I don't get why the wires couldn't be glued down in place, or why the plastic covers couldn't have been welded during assembly.

The transmission isn't anything too crazy: single DC motor with worm gear driving a pair of angled spur gears that then drive larger spur gears coupled to the spinning mop cover plates. The mop pads themselves affix to the bottom via velcro. I couldn't see anything clocking or restraining the motor aside from frictional force from the top-half of the clamshell, but with the degree of gear reduction and the non-backdriveability of the thing, I guess the motor isn't ever at risk of getting dislodged due to torque issues from jamming.

I do want to call out the nifty way that the bearings for the large gear and output plates are captured: a single snap-fit enclosure press-fit to the outer race of the bearing anchors to the main body. The assembly of the large gear sandwiches this piece and axially constrains the main output shafts.

The shaft for the mop pad plates seem to be either overmolded or glued to the plastic plate body. I wonder if that's to accommodate the basic 608 bearings used everywhere else. Intuitively, why wouldn't a larger bearing and then a plastic post have sufficed here?

tldr; I think I paid $30 for this thing. Don't do that. The Dyson probably wouldn't work that great as a wet vac anyhow.

Heed these online words,
Do not cross the Sodastreams,
Fizzy pop for all

I started drinking a lot of seltzer in the past few years after trying to cut back on soda and beer. My Costco membership made it really cost-effective to pick up a 35-pack of Kirkland sparkling water, but the number of cans always seemed super wasteful, and I got tired of shuffling them from the store to the car to home to the fridge, and so forth. A couple of previous roommates had Sodastream setups before, but I didn't like the amount of prep necessary for a relatively inconvenient container of fizzy water that would quickly go flat, and I could also never get it as fizzy as I liked. I wanted arbitrary amounts of cold water carbonated enough to make my mouth/throat hurt just a little. I'm told that's the real definition of the "spicy water" I like so much.

Being a frequent procrastinator on Facebook Marketplace and Craigslist, I was eager to leverage someone else's sudden or fleeting shame of hoarding for my own financial gain and snag a kegerator on the cheap. After all, I just needed a couple of tanks, some valves, and a mini-fridge, right? What could possibly go wrong?

A Summary of a Typical Kegerator Setup

A CO2 tank with regulator is connected to some reservoir (usually a keg) to pressurize it, which then forces some fluid (usually beer) to a dispenser. Setting the pressure and how much time the CO2 is connected to the reservoir determines the degree of carbonation. Not too fancy.

There are several Youtube videos/series on the topic, some of which go into far more depth than most would care to know:

• Everything I have learned about Unlimited Seltzer on Tap - Tyler's Garage
• Carbonation Playlist - Art of Drink
• DIY Home Seltzer "On-Tap" - Sean MacDonald
• Setting up a Home On-Tap Carbonation System - Dave Arnold

Sodastream users may be used to a screw-on plastic bottle as the reservoir, whereas homebrew setups probably use a Cornelius keg or similar variant. The nice thing about the latter is that they come in a range of various sizes and have quick-release connectors, as well as several add-ons for further modification beyond the basic configurations. All of this doesn't necessarily need to be stashed in a mini-fridge, but I didn't see the point of a kegerator build that doesn't dispense cold beverages.

Stuff I Messed Up in Setting Up My Kegerator

• The regulator comes with a nylon washer that MUST be placed in-line before screwing down the attachment. An old CO2 tank may have a damaged nylon washer, and without that washer properly in place, no amount of teflon tape is going to save you from gas leaks. I lost most of my first CO2 tank on the first night this way.
• Hose and barbed fitting sizes need to be selected such that they're all snug fits before you add hose clamps. Otherwise, leaks leaks leaks everywhere.
• Be careful about buying after-market corny keg quick-disconnect fittings. They look compatible, and the product listing probably says they're compatible, but the wrong ones will definitely leak. Be sure to have a spray bottle of soapy water on hand to test for leaks.
• CO2 tanks need to be tested before anyone will re-fill them, so if you're buying them secondhand, you'll likely need to swap them out. I read that paintball stores may sometimes swap out or refill tanks, but locally in south LA, I had better luck with welding supply and fire extinguisher stores.
• Didn't force-carbonate each batch aggressively enough, so then I'd have to go back and re-adjust the regulator pressure once the CO2 (slowly) diffused into the beverage.
• Didn't replace my faucet tap initially, so my first few gallons of seltzer were a bit mineral-heavy, to say the least

Does a Kegerator Actually Out-perform Sodastream?

Baseline Costs

The neighborhood Costco, as of mid-2024, sells a 35-pack of 12oz Kirkland-brand seltzer cans (3.28 total gallons) for $10, or approximately $3/gallon. There are more bougie and expensive brands, but I'll take the value pick for sake of this comparison.

A basic desktop Sodastream kit will cost around $90, with each CO2 tank refill costing about $15 or so. Sodastream lists the tank as sufficient for 60 liters (15.85 gallons) of seltzer, or approximately $1/gallon, though I guess your mileage will vary depending on how carbonated you like your drinks. Super rough math tells me that it'll take about 45 gallons, or 3 tank refills to recoup the cost of the Sodastream. Pretty good!

Hidden/Surprise Costs of Kegerator

A kegerator is going to be much more expensive, though the extra keg capacity plus the fridge plus the ability to carbonate other types of drinks may be completely no-brainers to some customers. Amazon lists a no-frills, basic kegerator for around $400, which comes with an empty CO2 tank but no keg. A typical 5 gallon Cornelius keg could run you an additional $100-120, and a tank refill/swap cost me $20-30, giving us a total of around $550 for a typical new startup cost.

I pieced my setup together from Facebook Marketplace and Amazon, for a total of around $300:

• Run-down, old/used kegerator w/ CO2 tank: $150
• Used 5-gallon Cornelius keg: $40
• CO2 tank swap: $30
• Replacement quick-release valves: $14
• Replacement CO2 tank regulator: 40
• Replacement faucet handle: $14
• Extra barbed fittings: $6
• Empty 5-gallon plastic jug for filling water: $10

If you have patience and filtered water at home, you could certainly fill yourself, but I found it easier to re-fill my 5 gallon tank from a water kiosk or purified water store (which I had no idea where so widespread and popular). That typically runs around $0.25/gallon. Compared to the Sodastream CO2 tank (which actually holds 410g or 0.9lb), a 5lb CO2 tank with $30 refill/swap cost results in about a $0.60/gallon rate. A properly handled 5lb CO2 tank could produce 80-90 gallons of seltzer. The shop that sold me my first CO2 tank gave me a 10-tank stamp card with a smirk, saying that she's never seen one of those cards redeemed ever. Surprisingly, I guess about 2 tank swaps should make up for my investment.

Procedural Differences

That said, my seltzer workflow has suddenly gotten more complex. As in, now I have a seltzer workflow to deal with. There's going to be some downtime when I run out of water or CO2. It's hard to tell when either will run out, though some DIYers have resorted to adding scales for monitoring usage. I just curse at the world and miserably resign myself to flat water (the horror) for a couple of days. Until I install a reverse osmosis setup or filter, I'll likely continue to lug my plastic jug to and from the nearby purified water store for refills.

Beyond refills, carbonation also takes time, since the gas needs to slowly diffuse through the exposed surface area of the liquid. There are a lot of suggested tips for forced carbonation, but ultimately it's a waiting game for the seltzer to reach the consistent and desired carbonation. That may not be something a customer is willing to deal with. There are both carbonation lids or dedicated carbonator devices to help expedite that process.

The build/debug process was pretty fun for me, and I thoroughly enjoy the dispensing aspect of the kegerator way more than the Sodastream, but be warned that it's not a set-it-and-forget-it system forever.

Dirty, gritty sludge,
Building little by little,
Progress is ruined

Around April or May of this year, I came back home to a flooded kitchen, and that kickstarted about three straight months of on-and-off frustration, complete with gerry-rigged pumps, greasy/stinky sludge water all over my condo, and many back-and-forth trips between the kitchen sink and my toilet to just get on with my day-to-day tasks. This is the story of that adventure, interwoven with some on-the-nose comparisons to what I think about the inner workings of teams and organizations, as I was also in the middle of another job search/switch in the same time period.

The Problem

The initial shock and disgust quickly gave way to the need to triage as quickly as possible. Out came the mop and bucket along with some frantic online searching for a local plumbing service. Beyond an initial cursory glance, I didn't put too much effort in trying to fully diagnose the problem. After mopping up the grime off the floor, I scooped out the dirty water from the sink and put down some old towels to catch any non-obvious leaks.

I just wanted the problem to go away, but without an obvious solution within reach, I settled for a slightly less terrible situation that I could bear with until I either learned more about the problem or found a more knowledgeable resource. When confronted with a new problem, I suppose a typical response can often be frantic and incomplete, but I much prefer that to nothing. That said, I'd find out later that my efforts the first night weren't exactly a net positive.

I have to admit that I had a lot of hope for the plumber we called the next day. I had the means, though not necessarily the practice, to outsource my problems away. That company had helped us prior with a shower pipe blockage, and I was looking forward to the guy pulling out the proper snaking inspection tool to unclog this set of pipes. He did indeed whip out a rather heavy-duty, drill-powered snake tool, and it very quickly churned out an alarming amount of black sludge from beneath my sink, but the pipes remained clogged, and after about 15 min of snaking, he actually suggested we find someone else, even going as far as to refund part of our payment. His reasoning was that if 10-15 min of snaking couldn't unclog it, the clog must be much further down the pipes, and without knowing the routing and structure of the pipes, it could cause damage, and we would be better off finding someone who could take apart portions of the wall to get a more direct look at the plumbing before continuing.

I've never had a contractor back out and partially refund his/her fee before, and this plumber could've easily stayed the allotted time to collect the full fee, but he clearly knew the limitations of his tools and time. I'm guessing he personally didn't think he had much of a chance at unclogging the pipe, and instead of leaving behind an unsatisfied customer, he could take a smaller loss through a partial refund, gain back some of his own time, and make me the customer feel somewhat better about shelling out money for a non-fix. During his short visit, we also discovered how my kitchen sink was directly tied to my neighbors', which made any sort of maintenance and repair even more difficult. As the next few months would prove, that plumber was much smarter than me.

I think people, especially engineers, have a tendency to be results and progress-driven, ever hopeful that some new development or insight will unlock a previously-unknown solution. That said, as any trite pontification on the sunk cost fallacy typically seems to converge, diminishing returns can easily become negative returns, and if your practice is well-established and tested, it also makes sense to set hard limits on how far to pursue the unknown before turning back. It's setting those hard limits without adequate knowledge that becomes challenging.

The failed plumber visit put me in a bind on two fronts: first off, I couldn't use my sink unless I re-routed to a waste bucket that I'd have to intermittently dump into my toilet; secondly, anything my neighbor did in his sink would ultimately end up on my side of the wall (and vice versa). I ended up getting a bunch of extra PVC pipes and re-did the entire bottom of my sink such that I had an extended pipe leading to a waste bucket that I could block off whenever I wasn't using the sink. This allowed me to at least get on with my life, though aside from waiting and hoping the clog would dissolve itself, I still didn't have a next step in terms of fully diagnosing and fixing the problem, as neither I nor my neighbor was interested in tearing apart our wall and pipes. In my neighbor's words, "We'll figure it out," which I guess was code for, "I'm going to ignore this problem. Good luck buddy."

A shared problem doesn't mean shared responsibility, and we shouldn't ever presume that a problem impacts all parties in the same way.

Without a viable alternative to tearing out the walls, I effectively started a timer of discontent that would continue accruing cost every day that the sink stayed clogged. Surely, if the problem persisted for some X number of days, it would've been worth it to just bite the bullet and find a wall-tearing, pipe-busting, pipe-replacing contractor. With that in mind, I set off on researching and trying out a whole bunch of DIY solutions. After all, I presumably was/am an engineer and should be able to figure out this bullshit.

Relevant expertise may not always be as relevant as you'd hope.

Two big unknowns remained in my path: I didn't know how far down the clog was in the pipes, and it could've very well been in a spot unreachable by anything I could reach through my sink; I also did not know how exactly the pipes connected between my neighbor's unit and mine. An early snaking attempt actually earned me a bemused tongue-lashing from my neighbor, who discovered my plumbing snake poking out of his kitchen sink. I'm fairly positive I tried every sort of mechanical probing or unclogging tool from Home Depot that was under $100, but to no avail. In the meantime, my imagination ran wild with what could've clogged the pipes, as neither myself nor my neighbor had any issues whatsoever in the 2+ years the two of us had been in our units. I found myself in the fuck-around-and-find-out part of the problem, except I wasn't finding out shit. Sometimes you're just ill-equipped to handle the problem.

The Grind

I eventually settled on using water pumps and flexible silicone tubing to pump out whatever sludge I could and then flush the pipes with hot soapy water to try and break down whatever was contributing to the clog. Through my various trials of shoving flexible probes down the piping, I had somewhat figured out that there was basically a simple T-connector between my and my neighbor's kitchen plumbing lines. While it was nearly impossible to route tubing down the central common pipe instead of my neighbor's sink pipe, it wasn't entirely impossible, and I made a series of strategic cuts to the end of the tubing so that it'd bend in advantageous ways to better facilitate it going down the correct path. Connecting the tube to a pump also allowed me to more easily determine when it had taken a wrong turn and start going upwards instead of down. Over time, I also started using metal beams as guides to further bias the tube's initial direction.

At times, I truly felt like I was trying to dig a tunnel with a wet noodle. Initially, I could only engage with the tubing at the opening underneath my kitchen sink. The direction of the tube as I fed more of it into the plumbing was largely determined by the interior geometry and curvature of the pipes, which I couldn't ever see. It would take on average at least 30 min of pseudo-random poking to get the tube down the correct pipe. However, once I got the tube moving down the central pipe, the suction from the pump and the continuous flow of fluid naturally pulled the tube further down the pipe, which was a neat physics quirk I didn't expect.

And thus, I settled in to a strange nearly-daily ritual:

• Come home
• Drain and dump out the liquid in the sink and pipes above the point of connection at the wall
• Undo all the pipes up to the wall and hope my neighbor doesn't run his sink while I'm setting up
• Connect the silicone tubing to the water pump
• Continue snaking the tubing through the pipes until it correctly goes down the central pipe
• Pump out all the grimey stillwater in the pipes
• Pump in clean, hot, soapy water
• Repeat pumping in/out until the water that's pumped out is clean/clear
• Be sad because the clog is still there
• Pull out the tubing and replace the plumbing fixtures so that it doesn't leak all over my kitchen floor
• Go to sleep

I ran into a couple of problems fairly quickly:

• The grit/grime that was pumped out would regularly clog the diaphragm pump, as they typically rely on a set of small flaps for check valves. In the beginning, I would regularly have to disassembly the pump entirely whenever it clogged, clean out the check valves, and then re-assemble the pump to continue. I eventually bought in-line filters and cleaned those out regularly instead
• After many days of use, the silicone tube would get softer due to the hot water I was pumping in, which meant that it would constrict more under heavy suction and not pull up water as quickly
• Suction pressure is limited and also went down as my pumps got dirtier, so there was a physical limit to the distance within the pipes I could pump out the dirty water.
• The hot soapy water may very well have been just further cleaning out the section of pipe prior to the clog, instead of eating away at the clog itself
• I had no guarantee that I was pumping out ALL of the dirty water before replacing it with soapy water, so perhaps I was merely replacing the top level water without influencing anything in the more effective proximity of the clog
• The water in the pipes was oily and gross. I washed and scrubbed my hands w/ dish soap sooooooo much and still had grime on them the next day

Despite all those issues, and despite the increasingly negative returns, I persisted, because what other option did I really have at my disposal? Unlike the plumber and his standards, I had none. Call it a combination of ignorance and arrogance. The process continued to make sense to me despite the lack of validating returns, and instead of giving up once positive feedback ran out, I elected to continue until I had a guarantee of a (literal) dead end. Oh, what a PhD does to a man...

Some Hope

I should add that a little under two weeks into this foray, I actually managed to unclog the pipes using this method. It was a magical moment, and I regained some normalcy in my life for three weeks or so until the pipes clogged yet again, for reasons unknown. Success here was perhaps a pretty terrible punishment. There really wasn't any concrete evidence that what I did had any effect on the clog, and yet at times I was trying to convince myself that I'd somehow invented a new, more effective way of fixing old plumbing.

More Grind

I would continue my personal vendetta against pipe grime for another four weeks or so after the clog came back with a vengeance. With my growing collection of water pumps (side note: macerator pumps are super cool) and tooling, I actually got pretty adept that getting the tubing shoved down towards the clog and slowly cleaning out the pipes. On a typical weeknight, you'd find me chilling on a rolling stool, slowly wrangling the tubing in and out of my kitchen plumbing while Spotify played in the background. You could say I achieved an acceptable routine in the interest of progress theatre: there really was no continuing concrete evidence that my work was effective, but I felt like it was, so I persisted and even worked to make my routine more efficient over time.

However, it became more apparent during this second plumbing period that my faith is my new process was a tad unfounded. It got easier/faster to clean out the dirty water, and yet I made no discernible progress on the clog. Not even a slow drain at the end of the day. A part of me became quite concerned that I had merely moved the clog further down the drain, and now it was even more difficult for me or anyone else to ever fix the problem.

I think people underestimate the fear, frustration, and confusion when an engineer/researcher can't reproduce a result. How many alternative contributing factors get blamed before the notion that the original premise was wrong even comes up? Perhaps it's human nature to latch onto even the smallest evidence that the daily toil has been progressive, and we miss the signals to move on and try something else.

I remember trying to evaluate my progress by subtle changes in how I felt the tube tug and pull under the pump's pressure, or the slight differences in the consistency and color of the gunk that was getting pulled up, or how fast I could get the waste water to run clear. All of this could only be seen at the output, far from the actual source of the problem: a true black box (or pipe) problem, if you will.    

Saving Grace(s)

I tend to gerry-rig a lot of bullshit professionally and at home, but I like to think that the thoroughness of my research is generally quite good. Going into my DIY efforts, I saw a lot of references to hydrojetting, where high pressure water can be used to eat through the clog, which we'd assume is made up of compacted food waste as opposed to a monolithic solid. Hydrojetting didn't initially seem like a viable choice as a DIY option, though I tried to simulate the high pressure bursts by pinching and releasing the tubing when I flushed the pipes. I went as far as to get a hydrojetting nozzle but was worried that I wouldn't be able to clamp it properly, which would risk losing it and worsening the clog. I also didn't want to risk unleashing a torrent of high-pressure in my neighbor's sink if I routed the tubing improperly.

When you're at the end of your rope (or tube), your range of options have a tendency to expand, along with a suddenly reckless risk tolerance.

Turns out, I didn't have to worry about clamping, as a standard hose clamp was more than sufficient to keep the nozzle connected to the tubing. The hose clamp also didn't make the assembly too large to freely move through the plumbing. The increased weight of the jet nozzle even made it much easier to snake the tubing down the central pipe instead of misrouting it towards my neighbor's side. It didn't take all that much jetting, even with a standard water pump supplying the positive pressure, to clear out whatever clog was there, though I ran through the whole process a couple of times just for the peace of mind.

I could've should've tried this solution earlier, but my sense of risk mitigation maintained this persistent fear that the possibility of making the situation worse couldn't possibly be worth it. I could've sprung for a professional (albeit expensive) hydrojetting service on the first day, but was dissuaded after hearing the first plumber's fears regarding whether it would work, so I didn't feel like throwing good money after bad. In retrospect, the threat of the worst-case penalty: completely ripping out my walls and possibly getting multiple neighbors involved in the process, led me to continue on the path that would cause the least harm, even when I had no conclusive feedback that the approach was having any tangible effect.

Post-Problem Insights

The first thing we immediately did after clearing out the pipes was to get those sink strainers to catch larger pieces of food. I had a disposal in-place, so it wasn't like I had been dumping junk directly down the sink, but at this point I was desperate to do all I could to prevent a clog from forming ever again. I'd also flush the pipes with hot water every night in hopes of further clearing out whatever residue was still left. In a short period of time, it became more apparent just how much food scraps my day-to-day cooking had been sending down the pipes. Should the pipes have been able to handle it? Most likely. Could the disposal have given a false sense of security and made the situation worse? Possibly. Either way, there was annoying but preventative care that could've been performed regularly to have avoided the problem entirely.  

As an aside, did you know that rice expands when it absorbs water?

I eat a lot of rice.

tldr; if you have the money, pay for the hydrojetting service. if you don't, make sure you have a sink strainer

Inflation is real,
Just can't say no to yum-yums,
Savings have flavor

A few things to note about the author, which may influence how seriously you take this post:

• more tinkerer than roboticist
• is not a great cook
• has limited taste (if at all)
• prefers Papa John's pizza to other pizzas (calorie-to-$ ratio is too high to ignore despite excessive bread-to-flavor ratio)

I've been tracking the use of robotics in food ever since the last few years of my PhD, mostly for fun. In fact, I remember my interviewers (and eventual managers) at JPL asking what I saw myself working in in a decade, and I answered restaurant robotics/automation. Probably not the right answer at that time. (It's a miracle I ever get offered jobs with the way I answer some things) However, I think the problem of preparing/assembling/cooking food is an elegant, complex, but still tractable problem for robotics, because:

• Cooking is semi-structured - There is a bound set of operations that would need to be done to a bound set of ingredients, which are typically packaged/sorted/contained in some repeatable way prior to reaching the chef/cook. The wide range of recipe complexity would (theoretically) allow us to more gradually solve technical challenges while still producing viable demonstrations/products. The job is also confined to a limited space where lighting and locations of tools/ingredients can be controlled.
• Companies have been streamlining/automating cooking via machinery/tools for ages now - It's perhaps more true in processed foods or fast-food restaurants, but many recipes have been reduced to as repeatable and simple of a procedure as possible, to ensure a consistent product to the consumer.
• There are already lots of tools/appliances - We're (hopefully) not cooking with only our bare hands. There are dispensers, serving utensils, mixers, blenders, molds, containers, and various other implements commonly used, and those can also (probably) be modified for use with robotics. I've never been a big fan of trying to get robotics to do anthropomorphic tasks in anthropomorphic (and bio-inspired) ways, when we as engineers can literally build whatever we want.
• Robots/automations are really good at timing - A big part of cooking (imo) is about adding enough heat to some subset of ingredients for some acceptable range of durations. As a person just trying to consume sufficient calories before running off to work in the morning, that can be difficult to track at times. For a robot, not really a challenge, and that seems like a fairly nifty build-in advantage we can leverage.

As I've spent more than a bit of time recently developing/deploying automation, I've grown a little more wary of the future prospects for robotic food startups in the near-term. I've also grown more appreciative of the machinery and techniques developed by the processed food industry over many past decades. Margins for restaurants are razor-thin, upfront costs for automation are high, and manipulation remains difficult. So, how well have the stalwarts in the industry done, where are we at now, and what can we look forward to in the near-term?

Types of Food Robotics Companies

I think it's most worthwhile to categorize food robotics companies by how they prepare the food, which to date hasn't shown too much variety and remains heavily limited by how we're able to physically handle the ingredients:

Beverages (Coffee, Alcohol, and Tea)

Machines for mixing quantities of fluids and dumping them into containers have been around for a long time, so it's not too surprising to see so many examples deployed in the wild, but perhaps their limited commercial success suggests shortcomings with the model or performance (or a blatant oversight on my part). Soda drink dispensers are everywhere, nearly every gas station and college cafeteria has some form of (not-so-great) coffee/cappuccino machine, and you're even likely to see slushie machines at your local robot combat tournament. This is to say that the process of making mixed drinks of many varieties have already been refined and mechanized fairly thoroughly. It would seem that the primary items left to automate would be the order-taking and the physical moving of the customer's container to the various dispensers and then finally to the customer.

For all the talk of how soda beverages are a high-margin business, it doesn't seem to stand on its own, unless you count vending machines (and maybe we should!). Sodas are typically an add-on to a meal or snack, and perhaps the returns are too meager for automating the placement of a paper cup beneath a valve and pushing a button for some time, especially when the pre-packaged alternative is indistinguishable in taste. Miso Robotics has recently introduced a concept for automatically prepping drinks for customers; and Cornelius, one of the primary vendors for restaurant beverage dispensers, has a similar desktop model as well. Neither approach puts the beverage in the customer's hand. It assists back-of-house and shaves off some time per order, but as with any new tech, also enforces new requirements for upkeep and debugging.

Likewise, you'd think mixed alcoholic drinks would be another highly profitable beverage to automate, and a few companies like Makr Shakr, and Tipsy Robot, have deployed industrial robot arms and a ceiling full of dispensers to make a wide range of mixed cocktails. A quick Google search also returns a long list of DIY robot bartender projects quite easily cobbled together from standard, readily-available 3D-printer parts ([1],[2],[3],[4]). There's no cute umbrella, lemon wedge, or sarcastic bartender discourse, but they're arguably faster and less error-prone than your typical bartender. I've actually visited the Tipsy Robot in Vegas, and I think it's pretty legit and slick: safety laser curtains set up to ensure customer safety, a fun robot performance to deliver my drink, and a high price per drink to remind me that I'm in Vegas. Maybe it's the latter, or maybe location/ambience matter a bit more than automated convenience, but the place was super empty everytime I visited or walked by. I also wonder if alcohol licensing requirements would prevent these sort of establishments from becoming fully automated and personnel-free.

High(er) end coffee perhaps offers the best standalone high-margin opportunity for automation, but I'd argue that the main technological advancement from the majority of robotic coffee kiosks, like CafeX, Crown Coffee, Cofe Plus, and many others, is automating the customer interface, independent of how the product itself is made. I will and should acknowledge that timing and scheduling of orders for hot beverages can be important to how well (and long) they are enjoyed, but it still stands that these kiosk concepts integrate a robotic arm and some flashy tablet displays to deliver a beverage made not all that differently than coffee vending machines from decades ago. How was there a single transition step between instant-coffee vending machines and pairing a robot arm with a Keurig?

To be fair, there are startups, like Rozum Cafe, Ross Digital, and Artly Coffee, using relatively complex robotic arm systems with tool changers and grippers to operate espresso machines in the traditional way. Instead of open-loop motions, there are fiducials or specialized fixtures to make sure the different espresso tooling is properly located and grasped. Instead of a no-contact procedure, various components need to be docked, picked, dropped, etc: a multitude of contact events need to occur reliably in sequence for the espresso to be made. In my opinion, the difference in task complexity is night and day. To me, this begs the question: is that extra complexity (and dramatically higher risk of failure) worth the (supposed) increase in product quality, especially if the majority of the clientele would hardly be able to tell the difference? After all, for all the talk among coffee enthusiasts of getting the perfect brewing profile through an ever more expensive progression of machines, I don't think that level of taste and obsession has reached the main populace. Although there are a multitude of machines built to prepare different forms of coffee with an increasing precision in terms of temperature and timing, I haven't found instances of coffee shops looking to automate the upskilling of their labor. For now, maybe the most economical coffee automation product is a conveyor moving cups beneath a Keurig after all.

A couple of companies (fewer than I thought), including Bobacino (via Future Pearl Labs), Grab Tea, and Koubei/Happy Lemon (via Taiwan Intelligent Robotics), are also targeting the boba tea application, which already utilizes a bunch of customized table-top appliances as part of a quasi-assembly-line layout not too different from your typical Starbucks. The Grab Tea setup looks very similar to the Makr Shakr solution for cocktails, down to the end-effector (maybe they're made by the same company), and all of the examples above employ some 6-dof arm to handle the cups, though I doubt they really need it. I wonder if it's additional complexity or lack of demand that's prevented this concept from gaining more traction. Ultimately, we're still dumping pre-made ingredients into a cup, but maybe the reliable dispensing of solid ingredients like boba pearls and pudding is still a tad beyond reach for the companies that have tackled this problem so far.

On the other end of the complexity spectrum in the automated beverage space, Blendid makes smoothies pretty much from scratch, using raw ingredients and your typical blender, without any (apparent) shortcuts. It even grips the blender containers with a standard robotic gripper as opposed to a custom end effector or tool-changer and dispenses the finished product via pouring instead of a spout or nozzle. Their magnetic half-moon slider to shift and stage the finished drinks around is also super slick (in my opinion). This concept could've gone with concentrates or pre-packaged pouches (ala Juicero but w/o the subscription nonsense) but I think the concept needs the fresh ingredients to distinguish the end product from a bottled beverage you snag out of a fridge.

Most of these beverage concepts pair some robotic arm with an array of liquid dispensers, boasting of a seemingly impressive range of flavor permutations (though who knows how many of these are worthwhile) and a quantifiable (but insignificant?) improvement in accuracy. I haven't found any published details on these systems' mean time to failure (not surprising), but many of these kiosks are deployed for real, not just set up for well-structured demos; and while regular restocking and cleaning is still necessary, these concepts can certainly run unattended. However, none of these seem particularly fast, and I think the cost of the arms/gantries makes the concept difficult to scale/adjust for the requirements of peak hours. As a result, I'd argue that each kiosk gives you the performance of a slower employee while offering a more limited menu. That's not to say there's no value with those limitations, and automated task scheduling could help mitigate the throughput issue with more optimal timing, but I think that even (arguably) the simplest food product doesn't have an easy path towards full automation to match the performance of your humble local cafe. I also think it's very telling that the large players like Starbucks, Dunkin', or Peet's haven't trialed these automated solutions at all.

Mixed Bowls

A couple of the first robotic concepts that got me interested in food robotics were Sereneti Kitchen (now dead) and Spyce Kitchen (now owned by Sweetgreen): systems that dumped ingredients into a heated container at the proper intervals and then mixed/stirred. I'm a big fan of one-pot recipes and stir-fry, and with regards to robotics, this cooking strategy seems like the next natural progression from beverage dispensing/mixing. The addition of ingredients can be haphazard and messy, and the cooking vessel naturally contains and cages the final product. Most of these concepts rely on some auger or pre-portioned containers of ingredients to dispense the proper amount. Some may discount the "automated-ness" of these concepts since many of the ingredients need to be manually prepped in a particular way, but I think you have to draw the line somewhere.

Chowbotics (acquired and then shuttered by Doordash) seemed(?) to have cornered the salad-making kiosk space in its heyday. A rotary carousel of paddle-wheel-based dispensers would dump a sequence of common salad ingredients and toppings into a bowl. No heating and the use of (mostly) shelf-stable ingredients seemed, in my opinion, to make this a viable and reliable deployment of a fully automated kiosk. The concept and business seemed to go pretty far prior to closing: they had a partnership with Aramark to set up these machines in large businesses (like hospitals and universities), and they went as far as to prepare instructional videos for service technicians. I wonder if the eventual failure was due to the larger customers implementing more WFH and remote work in the wake of the pandemic, or if the output product wasn't all that distinguishable from a pre-packaged and sealed equivalent sold for cheaper from a larger warehouse. The only other automated kiosk I found working on primarily serving cold salads is Bolk, out of France, and even they seemed to recognize the need to provide some hot offerings.

I'd divide the companies working on hot mixed bowls into two camps: those making tabletop appliances that adds ingredients to a heated pot at prescribed intervals and then mixes them, and those working on a more complete end-to-end model to eventually serve as a standalone restaurant. I'd argue that Spyce Kitchen started as a conglomerate of the former and was transitioning to the latter at the time of its acquisition. For me, the key differentiating insight here is that a high-throughput restaurant churning out bowls at scale (e.g. Chipotle, Junzi Kitchen, any ala carte Asian restaurant) isn't likely to cook individual bowls one at a time. They're primarily dumping pre-cooked and warmed ingredients into a bowl with minor topping modifications. There certainly continues to be an opportunity here to customize individual portions to a degree and with a precision unsustainable by manual means, but despite what 'secret menu' acolytes will argue, I think the masses do just fine with a limited menu.

I doubt that Sereneti Kitchen was the first to add an articulating stirrer above a pot, and Spyce probably wasn't the first to deploy a rotating, induction-heated bowl with built-in stirrer, but I think it's pretty insane that the automatic wok/cooker is a tested/proven restaurant appliance available in a range of form factors off of Alibaba, and the mechanization of that cooking motion was first implemented at least a decade ago. There's a semi-automated restaurant called Bowl & Bowl using an off-the-shelf fried-rice-cooker, complete with a mechanism that dumps separate ingredient batches into the pan in the prescribed sequence. Bots and Pots, out of Zagreb,  leverages a concept similar to the Sereneti Kitchen prototype in their kitchens for pasta dishes. Likewise, much like how Spyce Kitchen first started, we can now find the rotating, teflon-coated cooking tub w/ internal blade demoed at a multitude of restaurants on Youtube and Wechat, and you can find more advanced models, like the Wokie from Mukunda Foods, that also handle cleaning, cooking guidance, and adding oil.

More "advanced" efforts in the automated food bowl space mix in robotic arms to directly handle the movement of bowls or ingredient containers/dispensers between the storage, heating, and delivery areas. Karakuri has an expandable and modular workcell design with (what I'd call) swappable pods, each of which can be (theoretically) customized to dispense or cook a particular ingredient. It leverages a custom bowl carrier with an outer geometry making it easier for the robotic arm/gripper to hold onto. Aitme flips that script a bit and grabs the individual cooking vessels through a customized post instead. Fixed rollers rotate the inductively-heated bowls to generate the mixing motion. Beastro, via Kitchen Robotics, replaces the robotic arm with a customized gantry and chutes/slides for a more compact kiosk form factor; and they, like Chowbotics, are seemingly far along in their development to have started focusing on usability/deployment. Nommi, Mezli, RoboEatz, and Hyphen are other groups also developing alternative ways to move a bowl beneath various dispenser designs.

A slightly different subset of automated food-bowl companies incorporate freshly-cooked (and sometimes freshly mixed/extruded) noodles/pasta, which for the most part means the addition of perforated strainers dipped into a boiling vat of water and a flipping/dumping motion to transfer the noodles. Cala focuses on Italian pasta dishes while WangJieXie has deployed some standalone noodle soup stands to demonstrate their automatic noodle makers. P-Robo from Techmagic shuttles a cooking vessel back and forth between a noodle dispenser and a sous chef who adds the final garnishes and makes the dish more presentable.  I've often thought that hearty soups or thick pasta sauces make for ideal components in autonomously-cooked dishes, as they can be more easily dispensed and kept warm/homogeneously-mixed for longer periods of time while immediately upgrading a dish that would otherwise be indistinguishable from a vacuum sealed box out of a grocery store fridge. That said, maybe I'm severely underestimating the difficulty/cost of keeping food warm and fresh.

Now, this is pure speculation on my part, but as I alluded to earlier: while an all-in-one appliance may make a single dish/serving at a time, a restaurant is more likely to prep larger batches, keep them warm, and then parcel them out depending on customer orders. There's probably also limitations to what a teflon-coated, induction-heated cylinder can properly cook. I don't think Chipotle is transitioning to rotating drums to cook their barbacoa and carne asada, though I guess you could argue that a higher BTU cooking flame and a non-linear rotational trajectory could open possibilities beyond just evenly heating up food. The most recent videos from Mezli and Spyce seem to separate the cooking and dispensing steps, as well as hiding how the hot items are cooked. I would guess that it's more than just the parceling and portioning of ingredients that are still best handled by a careful human hand.

Pizza

Any attempt at an automated pizza dispenser will (and should) be overshadowed by the efforts and success in the industrial automation already developed for making frozen pizzas. Isn't a frozen pizza just a fresh pizza with an extra step? An entire brand's (DiGiorno) strategy relies on their reheated product being indistinguishable from (or better than) a restaurant equivalent. Likewise, pizza chains are already pretty well-oiled machines, with years of optimized procedures and supply chains. Your local pizza slinger is probably not spinning doughy discs into the air and stoking the fire on their wood-fire oven.

In my opinion, the overall strategy for automated pizza-making is fairly well-understood and tested: mix the dough, shape the dough, smear the sauce, sprinkle on the toppings, heat the pizza. The methodology doesn't seem all that much different from putting together a mixed bowl, except now we have to form the bowl before dispensing the toppings. In addition, toppings aren't as easily contained, and the spatial distribution matters significantly more. Another major challenge lies in miniaturizing the frozen-pizza production line and fitting all that in back of house, a shipping container (Hyper Robotics), or even a truck (Stellar Pizza).

It appears that most (if not all) of the players in this space prep the dough beforehand, whether it's a ball of dough that gets pressed/rolled flat by the machine or a round disc already in the cooking pan. I'd guess that it's unreasonable to mix the dough on demand from both a time and waste perspective without the torrid throughput of a major chain. The machines utilize the same sort of dispenser technology used in mixed bowls, but they can't dump the ingredients all at once, and the dispensation is typically coupled with some relative motion of the pizza pie itself to guarantee an even spread. Some implementations just dispense the sauce via a nozzle, while other ([1],[2]) use a physical massager of sorts. To compensate for the stickiness and compliance of the dough during transfer, systems either use a dedicated tray/plate per pizza or a series of mobile conveyors to avoiding dealing with friction. Pazzi (now defunct) used a nifty array of pins to lift up the pizza and still allow for a slotted spatula to easily go underneath the pizza without needing to get a robot to reliably and dynamically slide beneath the pizza crust without snagging.

The advent of specialized home pizza ovens and popularity of places like Blaze and Pieology would suggest to me that there's an appetite for customizability and freshness available without the wait, and a robot is seemingly well-suited to handle the increase in permutations. However, the failure of Zume just a few years ago, despite being backed by Softbank's coffers, and the more recent closure of Pazzi should compel us to ask if automation (at the customer-facing restaurant level) is worthwhile for this food product. If so much of the ingredients need to be prepped a certain way beforehand, and human assistance is still needed during operation (if only to supervise), how much can you really save on labor costs, and how much better could the food actually taste? As with my thoughts regarding Starbucks above, when the biggest player (Domino's) in the space isn't driving the automation efforts, the potential returns may not be as great as we think. (To be fair, Pizza Hut is driving the Hyper Robotics efforts)

An aside: the author grew up on microwaved (not even oven baked) Totino's pizza and enjoyed it very much

Burgers (and other meat in/on bread)

Like I mentioned, Creator (formerly Momentum Machines) first introduced me to food automation, and they spent many years developing their solution before ever revealing it to the public. Their commercial implementation was equal parts performative art piece and whimsical assembly line, handling everything from cutting of the bread and toppings to the cooking of the patties to the final burger assembly (sort of). I've always found it weird that they use a series of flipping paddle wheels to move the burger down the assembly line as opposed to a conveyor, though I guess that allows for multiple burgers to move down the line at different rates if necessary. The final output is really an open-face sandwich, with the burger container structured to help localize the bottom and top buns. That said, despite its temporary (or maybe permanent) closure due to Covid, I believe it's successfully proved out the concept for a modern-day end-to-end autonomous burger shop.

Surprisingly (to me), AMF (yes, the people behind bowling alleys) produced an autonomous burger-making system already back in the 1960s. This suggests that for constrained problems, people have been quite good at developing series of mechanisms to execute even highly complex tasks without the aid of the complex vision/manipulation robotic systems we take for granted today. An amalgation of conveyors, alignment features, and chutes are more than sufficient to put together a burger. Foodom (or Qianxi Robotics Catering Group) developed a similar collection of suction cups, gantries, dispensers, and conveyors to make simple burgers in China, as has RoFood, out of Russia, and RoboBurger scaled back the complexity to focus on a very simple hamburger with only condiment toppings. Like Creator, these efforts share a few limitations: the packaging (whether box or wrapper) is critical in helping keep the burger localized as it gets moved from station to station, the patties along with the other toppings inevitably have to be dropped onto the buns, which pretty much guarantees a mess, and the components handling the meat patties get grimy/greasy pretty fast.

Miso Robotics (more on them later) started off as a burger-flipping robot, cooking and flipping burger patties on a grill with a standard spatula, the old-fashioned way. It's very cool and ambitious, and perhaps it made sense on paper that force/vision feedback in combination with an industrial arm could reliably slip the spatula underneath cooking meat, but I think it says a lot that Miso is no longer pursuing that concept. Some other robotic-arm-based projects appear to primarily be technology demos and include a cold-cut sandwich builder from Vasma Robotics, and a hotdog maker from Velox Alpha, which had an unfortunate failure case captured and uploaded to Reddit.    

Deep Frying

The afore-mentioned Miso has apparently found its stride in deep-frying chicken wings and chicken tenders. They provide a robotic arm/gripper on an overhead gantry that picks up baskets of raw meat, dunks them into deep-fryers for the optimal amount of time, and then transfers them to the next stage in the line. The system appears to be easily adaptable to various kitchens with just the addition of fiducials on the baskets and fryer locations. A Korean robotics company has set up a similar concept in Canada (of all places) called Space Robo Chicken (of all things). There's also a chain of fried chicken shops in Korea called Robert Chicken (not a typo) that also handles the dredging and coating of flour. Dedicating an industrial robot arm to deep frying feels excessive, especially since deep fryers with timers and electric lifts already exist. These systems also don't do anything to help with the portioning and packaging of the chicken after frying.

Eggs and Batter (Liquid to Solid)

In my research (aka browsing robotics videos on WeChat and Youtube), I came across two weird edge cases tied (or congealed) together with a key commonality: egg. Eggs start out liquid, easy to dispense haphazardly and evenly, but as they cook, the end product solidifies and (arguably) becomes possible to handle through non-prehensile means. They also bind together ingredients/toppings that otherwise would remain loose. By definition, we're reducing entropy through cooking, but eggs present a convenient situation in food where the application of heat is all you need to do so.

Kurve Automation put together a nifty omelette kiosk in Singapore where customers select their add-ins for a robot arm to add to an omelette. The system swaps out a variety of tools to add egg, mix the omelette in a teflon-coated pan, and then ultimately scoop it out and deliver it to the customer. This solution appears to be purely open-loop, so perhaps it demonstrates more the utility of the humble egg in cooking than any technological advancement.

The Beijing Weilizi Technology company may have tackled an even more difficult challenge in building a kiosk that makes JianBing, a thin Chinese crepe w/ savory fillings. Much like the Kurve Automation omelette-making station, this setup is still open-loop but spreads, lifts, and folds a very thin layer of egg-based batter. It leverages the same rotary cooking base that any food hawker would use to make the same dish.

Reheating and Vending

There are a couple of mixed bowl and pizza vending machine concepts that I removed from their native categories above because it appeared that the kiosks or machines themselves merely heated up pre-packaged meals, and that strategy deserved its own category. In the states, you may occasionally come across a fancier-than-average vending machine that dispenses sandwiches or maybe even fruit. Across the ocean, the options are significantly more extensive, to say the least, and Youtube channels like Ericsurf6 and DancingBacons give a nice sneak peek into that world. In the most basic form, hot-food vending machines aren't too complicated: the internal temperature is just kept high, and maybe the owner needs to refresh/restock a lot more often to deal with perishability. Shelf stability is not necessarily the same as shelf viability. However, the former is a nice thing to leverage if you can apply heat quickly on-demand.

Korea/Japan apparently have self-service ramen stands with shelves of instant ramen, various toppings (including raw eggs!), and hot water dispensers. My interpretation of Yokai Express's videos/patents ([1],[2],[3]) is that they're trying to cram all of that in a box and up the quality at the same time. The fast-casual version of a self-serve, instant ramen stand, if you will. Fast and even heating is not easy, especially if Yokai's patent images are in fact a good representation of what they have and continue to work on. I think we're all familiar with trying to quickly nuke something in the microwave on high and then somehow managing to both burn our mouths and taste cold mush. Maybe (and this is full-on speculation) adding in piping hot broth with agitation is an opportunity to finish off the heating of par-cooked foods.

Another aside: Dave Chang's podcast episode on microwave myths is delightful

I also uncovered several vending machines that seem to just heat up frozen pizza; hence why I didn't include them in the Pizza subsection above. Aldi's Pizzabot, Basil Street, and Pizza Forno all appear to just slide pizza across some heating elements and into a box for the consumer. The videos for the Aldi Pizzabot shows off that process the best, and it's pretty identical to how any fast-food pizza joint would do it: just move the pizza at the right speed through a hot oven zone. Fortunately, pizzas are thin enough to cook thoroughly and quickly even if we only heat convectively from the outside. I've sometimes wondered if the endless racks of frozen pizzas in our grocery store aisles hampers the attraction of automated pizza vending: how much appeal is there for something we can easily re-heat at home and for which we already have a multitude of options? On the other hand, simplifying an automated kiosk or vending machine down to what is essentially a microwave (or science oven) and a conveyor allows for the food to be produced in a much larger and far less constrained facility at scale elsewhere.

"Freeform" Anthropomorphic Cooking

On the other end of simple re-heating and dispensing, there are a few highly ambitious robotic systems built to more closely mimic human chefs and how they "normally" prepare food. Moley and the Samsung Chef Bot may be two of the more well-known "general-purpose", robotic cooking assistants, in theory promising a one-to-one replacement for the kitchen cook through a combination of AI, advanced machine vision, motion capture, and the latest in robotic hardware developments. In practice, both appear to replay a series of complicated-looking, open-loop motions for what ultimately are well-choreographed and well-constrained tasks. Ingredient portioning still appears to happen independently of the robot a priori, and the actual motions in the cooking task do not appear too different from the mechanized mixed bowl concepts described above.

Nala Robotics has deployed what I'd describe as a true industrial, robotic kitchen, designed to compensate for the shortcomings of automation available today while still staying true to conventional cooking methods. They've modified the handles of pots, pans, and other kitchen tools here and there, while adding dispensers where they make sense (why pour out of bottles/pitchers when you don't have to?), but their demo reels show very human-like handling of utensils and cooking vessels. If you were to block out the robotic arms and just focus on the motion of the food, it's a pretty good approximation of what happens in any non-mechanized kitchen. The staging and portioning of the finished food is missing from their published videos, so I'd imagine their primary product is more for a ghost-kitchen setup, serving multiple storefronts/brands with a consolidated back-of-house. It concerns me a tad that their more recent projects ([1],[2]) focus on some of the specializations above rather than expand their general-purpose capabilities. Perhaps the upfront capex differences are substantial and too limiting.

Connected Robotics is another company developing automated kitchens set up explicitly for robotics. Their primary offering leverages a collection of robot arms and customized brackets to prepare and serve ramen, but they've also developed a concept to cook takoyaki balls, complete with a vision system to evaluate when the cooking is complete. Most impressively, many of their prototypes show off modifications to existing kitchen machinery to make them more accessible for a robotic arm to operate through non-prehensile motions, and some of their other demos include retrieving cooked goods from a heated cabinet, serving beer, and cooking burgers.

Other tool-swapping cooking systems include DaVinci Kitchen, Gastronomous, and Dexai. Whereas the first two appear to rely on well-designed mechanical fixtures and hard-coded motions like Nala Robotics, Dexai seems to utilize machine vision and learning to swap out tooling, which can be in a variety of poses/configurations (not specific to a fixed tooling rack), and certainly seems to operate in a more reactive (as opposed to pre-scripted) way (Artly Coffee seems to do the same). From a research perspective, it's a really cool first foray into enabling a system to deal with the unpredictability and messiness of a typical kitchen. From an automation perspective, why increase the complexity when you presumably have full control over the design and structure of the task environment? There's a bigger goal on the farther horizon here for Dexai (and also Connected Robotics), but the Nala, DaVinci, and Gastronomous examples suggest that extensive scripting/planning with just open-loop motions can still get you pretty far.

Intra-Restaurant Logistics

The most mature restaurant robotic tech may not get very far, despite being the most mobile of them all. AMRs/AGVs are arguably having their moment right now, as warehouses continue to get larger, and customers need to shuttle more and more packages back and forth faster than ever. Companies have developed mobile bases for the hospitality, hospital, and security spaces for years now, and companies like Pudu and Bear (among many many others) have some of the more visible demos and installations in restaurants. I wouldn't say basic navigation in indoor spaces is a 100% solved problem, but it's certainly mature, and saving servers the physical labor of carrying loaded and dirty plates back and forth between the kitchen and front of house seems like a tractable task to tackle.  

Yet another aside: If you see Pudu's robots in any restaurant, take a quick gander upwards and see if you can find their fiducial markers on the ceiling

I've seen some of these up close, and I've worked on a couple of variants from sketch to prototype; they admirably do exactly what they're designed to do, but nothing more, and that may not be enough for a bustling, cramped restaurant setting. The basic differentially-driven base with a standard suite of proximity sensors mounted onboard is pretty reliable at navigating from point A to B while avoiding (sufficiently large) obstacles in the way. That said, they don't clear the table or load themselves or always know how to handle temporary obstacles in their path...like chairs. When I see these deployments in person, I usually start a timer to see how long it takes before it gets stuck near a crowded table and needs to be rescued by a server. That's not to say these systems are terrible ideas in restaurants, but if a server needs to accompany the bot anyways and now may have extra responsibilities in case of unforeseen edge cases, the net gains start trending negative.

Maybe there's more promise in the sushi conveyor system. Or maybe a sushi lazy river? Or how about an overhead robotic crane?

Where the Tech Continues to Fall Short

Under-structured and Under-constrained Tasks

Many of the examples described so far make compromises and adjustments in how the food is cooked or the ingredients are prepared. A general-purpose, or at least more-easily-adaptable, robotic system would theoretically increase the range of possible cuisines. However, I think efforts like Miso's original foray into burger flipping have shown that even simple, purposeful tasks involving contact with deformable/soft ingredients through tools can be tough to get right. In my opinion, the initial challenge lies in reliable tool swapping and handling (which Dexai appears to be heavily invested in), and the next (pretty much unexplored) challenge relates to how the system utilizes those tools in contact with a food mixture changing in consistency.

Even the simple act of scraping the bottom of the literal (food) barrel lacks a go-to mechanized solution. If the food can't be reliably dumped en masse into a container, a human assistant needs to help. Advances in bin-picking and even the handling of deformable food don't quite start to address the physical requirements in just assembling a plate. Foodly, via the RP Corporation, may be one of the more advanced solutions in this space, and the end-product is rudimentary at best. Not a single robotic installation for automated frying seems to have a method to portion the end-product for the customer.

It shouldn't be surprising then, that ingredient preparation, whether that's just portioning or slicing vegetables/meat to the right size, remains a manual endeavor. Despite there already being food processors (and the Slap-Chop, of all things), even fast-food restaurants may insist on hand-sliced ingredients for quality, texture, and mouth feel reasons. Researchers have certainly attempted the task, but I guess you could say efforts have stayed mostly academic to date. It's still far easier to manipulate a container than what's in the container, so there appears to be significant opportunity in developing new end-effectors for food and more adaptive methods for handling utensils.

Scaling Throughput

I've often been told that robotics is theoretically easy to scale, because you just invest in another one once you reach sufficient savings on the first one, but I disagree. Automation needs both support equipment and space that most establishments probably can't accommodate, while delivering a fraction of the flexibility that a minimal-wage employee with minimal training could provide. Without an overhaul of the line, there's no straightforward way to parallelize any intermediary step. Worse yet, a lot of the examples above show many of the subsystems simply sitting idle if the dish being prepared is at another step.

Despite the supposed programmable flexibility/adaptability of robotics, no setup that I've seen can re-task its manipulator for another sub-task on-demand. Rather than adopt the sense-think-act description of robotics, I rather like thinking of a robot as a mechanism that can execute different tasks with zero setup time. There's an interesting conundrum where a 6-dof arm theoretically gives you more capability, albeit capability that is unlikely to be fully leveraged once the installation is fully-developed and locked in. Even if it were an option for onsite technicians or savvy back-of-house workers, I think the re-setup and re-tasking cost would need to be exceptionally low (if not zero) for it to even be worth considering.

One thing I don't hear being discussed is how these automated systems would handle peak vs off-peak hours. The bulk of the food automation examples deployed so far also arguably operate slower than the greenest human equivalent. If we're to design for the worst case, that could end up being a substantial increase in upfront cost that will just have to sit idle. You can't adjust the schedule and bring in extra workforce when you need it. Aside from some of the beverage examples, I don't think any of the restaurant automation examples I've found operate in your peak, line-out-the-door conditions.

The costs (both upfront and upkeep) of scaling make me wonder if the true unrealized gains in restaurant automation require us to go back further back-of-house and embrace the gains we've already achieved in mechanizing the production of prepackaged and frozen goods. Instead of shrinking and moving that entire system as close to front-of-house as possible in the standalone kiosk model, has anyone (maybe Cloud Kitchens and Kitchen United?) focused on decoupling the par-cooking from the finishing and only moving the latter to the customer? How impressive does the customer-facing science oven really need to be as long as the food is good and cheap?

Exceeding Restaurant Quality

When discussing automation, I typically think first about compromises and tradeoffs. How can I simplify the task to make it as reliable and repeatable as possible, and how much of a loss in quality can I incur? For all the examples of restaurant automation we've discussed so far, I can't name a single one where the customer reviews raved about the taste and flavor being better than what you could get from a human cook. Automation strives to maximize margin, which could mean increased yield and quality, but it doesn't have to.  

For me, the story of The Melt perhaps best exemplifies misplaced trust in how technology could elevate the dining experience. The precision and accuracy with which we can follow a recipe doesn't enhance flavor if the recipe isn't all that good to begin with. I don't think anyone has yet stumbled upon the dish where sub-second time-keeping and sub-gram dispensation truly elevates the food to new, otherwise unreachable, heights. Does a recipe so complex or fragile exist such that it can only be properly made when executed end-to-end by machine? Unlikely (but I'd be happy to be proven wrong!). Perhaps the efforts in the food robotics space can only at best achieve the status of the industrial croissant: not close in quality to the handmade original, but accessible in ways the original could never be for those ready to settle for less.

Cleaning

Who or what is cleaning the cooking vessels? It seems that advances in food automation have conveniently ignored the most ubiquitous and least glamorous job: that of the dishwasher. Grease buildup, sticky/charred leftover food, burnt sauces could all add an unfortunate surprise ingredient in the next dish that's prepared. The vast majority of the examples above use disposable plates/bowls to serve the customer, but it would appear that a human assistant still needs to at least spray down the cooking surfaces between orders.

A few of the concepts making mixed bowls ([1],[2],[3],[4]) flip the cooking vessel upside down and spray the insides with some nozzle array, similar to how countertop glass rinsers work. Many of the table-top rotary woks now have this has a built-in feature as well. This seemingly happens without any feedback to ensure that whatever needed to be washed off was actually washed off, and that's yet another reason why a non-stick cooking surface is paramount for the first cooking automation attempts.

Both Connected Robotics and Nala Robotics have shown off  some complex cleaning demos that handle various dishware and glassware with suction cups and grippers in a manner similar to how a human dishwasher would. They look slow and deliberate, a far cry from how a frenetic back-of-house may handle the same situation. However, this is yet another semi-structure problem that needs to be solved to move automation and robotics in the food space towards more complex tasks where the system needs to compensate for irregularities from the expected norm. Dishcraft Robotics (defunct as of 2022?) previously worked on this problem for years, eventually resorting custom dishware with embedded magnets to more effectively "solve" the manipulation problem. The size of their system forced them to adopt a dishwashing-as-a-service model as opposed to something that could be deployed onsite at customers' locations. In the large-commercial-kitchen space, we're more likely to see specialized cleaners customized for a particular plate size, or a hybrid approach with people stationed along a conveyor making sure the dirty plates/trays fall within the expected bounds in terms of pose prior to going through a dish carwash of sorts.

How to Turn Up the Heat in Food Robotics?

Literally. Is there a more effective way to transfer heat to the food that the automated system is dispensing? I've recently developed this belief that there are some applications not worth automating if you can't automate them completely. When considering the spatial bounds/limitations of a typical kiosk or restaurant, I think the Japanese/Korean vending machine ecosystem has it right: prepare the food 90% of the way off-site, using whatever traditional automation tools or hybrid workflows are most convenient, and then quickly heat prior to dispensing to the customer. Arguably, airline food also executes this exact strategy with far more limitations than your robotic kiosk.

Heating's seemingly not that sexy of a problem, but whether it's air fryers, sous vide, or the Turbochef Oven that's in more places than you probably know, that's the element predominantly dictating the final taste profile. I don't know if Yokai Express had that insight early on, or if they had other motivations, but I think focus on developing a better reheating method allowed them to simplify their customer-facing kiosk in a smaller footprint, while still (presumably) delivering a yummier product than if they tried to do everything on the spot. Minimizing time from time of order to delivery should be the primary goal, not any lofty aspirations for freshness or machine-driven quality through precision.

As with many (if not most) efforts in deploying robotics, the initial attempts focus on replacing the human element directly while retaining the existing (tried and true) workflows. We should recognize that the natural cadences of machine and man are different, even if we're bold enough to claim that we have full control over the former. If startups insist on automating the existing workflow in-place, I don't know that we've seen evidence of financial viability around the corner, or even the corner after that.

There's a need for speed,
The OG stock is lacking,
Brushes need to go

This post serves as a summary, as of the beginning of 2023, of the various drive gearmotor modifications that (to my limited knowledge) have been tried/tested for beetleweight (3lb) combat robot builds.

(I hope that this will also get me to finish my beetle, but let's not hope for miracles just yet)

Why Modify Drive Gearmotors

This can vary a great deal depending on the type of drive and wheel size you select, but a typical beetleweight build with direct-drive (wheels mounted directly to the gearmotor shaft) would utilize a 20-25mm diameter gearmotor with a standard 130 to 150 size DC motor, spec'd to run at 12-16V, and geared to a 700-1000 no-load rpm speed.  There are plenty of generic DC motors ([1],[2]) that fit these parameters, but those standard COTS options are (presumably) designed for elongated life and conservative operating conditions, not the aggressive bang-bang conditions of a (optimistically) 3-min fight. By mixing and matching gearheads and motors not normally paired together, we could get:

• equivalent speed/torque in a smaller size or lighter weight
• gains in max speed/torque at the same size/weight
• more robust gear transmission that's more resistant to impacts/hits
• mounting bolt pattern that's more secure
• better geometry for packaging

Selecting/swapping gearheads to better match the desired operating performance is a fairly common practice in mechatronics, but finding compatible combinations at an affordable price (for a hobby) wasn't all that easy until the past few years. While you could almost certainly get whatever gearhead you wanted via a comprehensive vendor like Maxon, it's (in my opinion) the combination of lucky finds on eBay/Aliexpress and the willingness of some builders to directly negotiate with and buy from Chinese vendors directly that has made these sort of higher-performance gearmotors all that more obtainable.

This also isn't some magical upgrade for success, especially for those of us just starting out, but there's significant room for optimization and improvement just through component selection. Learning and reading about different mods reminded me a lot of the Tamiya Mini 4WD kits I played with as a kid, where it was all about picking the right motor and gearing for the particular racetrack.

Gearmotor Modification Process Summary

The overall process is pretty straightforward and similar for most gearbox/motor combos:

1. Find compatible pair of motor and gearhead/gearbox that you want to combine.
   - Mechanical compatibility is largely determined by if the motor can attach to the appropriate pinion gear for the gearbox.
   - The pinion gear could be sourced separately, but it typically either comes with the gearbox or can be extracted from the motor that originally came with the gearbox
   - As the pinion gear is typically press-fit onto the motor shaft, it's sometimes recommended that you source a fresh pinion gear (of the same tooth count, tooth modulus, and ID) instead of pulling it off an existing motor
2. Make necessary mechanical modifications (if possible) so that the two can be fastened together.
   - Sometimes the motor shaft needs to be cut down so that the pinion mates properly with the gearset
   - There often is an interface/adapter plate (usually as part of the gearbox) that has a set of threaded holes for the rest of the gearbox and then a set of through-holes allowing you to fasten the motor. This may need to be re-drilled/re-tapped or replaced entirely for some combinations. Sometimes, it's a good opportunity to integrate a more robust mounting strategy
   - Many interface/adapter plates by default come with multiple sets of through-holes for compatible motors, so it's often possible to find combinations where you don't need to make mechanical modifications at all
   - With brushless outrunners, sometimes the shaft needs to be replaced entirely or pressed through to work
   - When possible, some builders replace gears (planetary/sun/pinion) with equivalent-sized ones of a stronger material. This is particularly prevalent w/ gearboxes that use nylon gears in the earlier stages.
3. The electronics may need to be upgraded/replaced to support the new motor
   - If switching from brushed to brushless, you'll need a brushless ESC modified to run on either SimonK or BLHeli firmware
   - If moving to a more power hungry brushed motor, you may need to move to a brushed ESC with higher current rating or consider modifying a brushless ESC to drive a brushed motor
   - Ironically, even though hobby brushless motors are newer, the proliferation of quadrotors have (seemingly) given us more brushless ESC options in more compact form factors than brushed

Here's a list of other (way better documented) guides on the process:

• The 5 minute, $20 brushless gearmotor (Russ Barrow)
• "5 Minute" Brushless Gearmotor for Beetleweight Combat Robots (StephenH102)
• How to make brushless 22mm planetary gear motors (Alex Mordue)
• Beetleweight Brushless Drive Tutorial (Robert Cowan)

Alternative Gearboxes

Haphazard Table of Alternative Gearboxes/Motors

For the purposes of this discussion, by alternative I mean alternative to the typical 22mm spur gearset you'd find from a vendor like Servocity, as that's the barebones, simplest option you could start with. You can see from the typical antweight ([1],[2]) or beetleweight kit ([1],[2]) that it's easiest to just directly mount the wheel onto the gearmotor shaft (as opposed to an indirect drive with dead shaft and belt/gear transmission) and fasten the motor to the side walls with the default mounting threads. By default, your standard 20-24mm diameter gearmotors will probably not reliably withstand the impacts or stall conditions of robot combat.

Double-Spur Gearhead

A typical spur gearhead transmits force along a single line of contact through the various gear stages. At each stage, there's generally a 1-to-1 tooth contact between independent gears (unless you're dealing w/ helical gears, but you shouldn't be). This makes construction simpler/cheaper, but reduces the load capacity. Apparently, you can get 25mm diameter gearboxes with a secondary, redundant set of gears for increased robustness. Absolute Chaos Robotics sells a variation of these with extended shafts in their beetleweight drive kits. A downside of these is that at 25mm diameter, they end up being quite heavy, and curiously enough, I haven't found such a double-gear setup in gearboxes of other sizes. The exact model number seems to be JGA25-370DG, but a search for "double-gear box" on Aliexpress should return a list of the different variations.

Flat/Box Spur Gearboxes

While a typical cylindrical spur gearbox has gears stacked in two primary columns, the gears could be spread out to minimize the axial length, like you're more likely to see in standard RC servos. For DC motors, they seem more prevalent for RC tank builds ([1],[2],[3]), and I think they're interesting because we could leverage the larger housing as a structural member in the bot's design, and the default shafts tend to be larger and include additional threaded features.

That said, these are pretty heavy, and I haven't come across any bots that have used these COTS gearboxes directly, but there have been a few custom solutions ([1],[2]) proposed/tested, built around gear train components salvaged from COTS boxes or bought directly.

Planetaries

If the design can package it, most drive modules will probably use a planetary gearbox nowadays. They have higher load capacity and distribute wear/tear/lubrication better than the simpler spur alternatives. You can find a multitude of options if you just search by gearbox diameter (22-24mm) on Aliexpress or look for gearmotors that come with a 370 class DC motor. Despite the increased weight, a typical early gearbox upgrade used to be switching over to the Servocity planetaries from the cheaper/simpler spur gearboxes.

Though a new builder may be better off nowadays going with one of the prebuilt/pre-tested, upgraded gearmotors listed further below, a crawl of the interwebs uncovered some pretty interesting alternative options in the past few years:

• 370 Micro Metal Planetary Gear Motor - Super lightweight (<25g) and with all steel gears (instead of brass/nylon), but the gear modulus is smaller and should be used only in indirect-drive setups, according to those I met who had used it. (Note: the link as of this post's writing seemed to work, but it also seemed to point to a listing for a gearmotor that's not quite the same as the one I remember when I first came across this motor)
• Rotalink Planetary Gearbox - The pictures show a rusty gear motor. The actual product has way more rust. I don't know what overstocked toy these came from, but they're as light as the micro metal option above (I think the ring gear housing is aluminum), but the initial stage has nylon gears, so it's common to buy a few excess so you can swap out those out for metal gears.
• PZ22GR9120R - Cheap (<$4 each) and available directly off of Amazon, but with a much higher reduction (1:84) than you'd typically want for drive (usually 1:10 to 1:40 depending on your wheel size). Also comes with a 6mm shaft (atypicaly for gearboxes of this size), though with a slot instead of a flat.
• Diameter 22mm Planetary Gearbox - Discontinued, but these look very similar to the Dartbox gearheads offered by Just  'Cuz.
• M22GXR - This looks like what's being sold as the MercuryBox by Robot Matter. The larger 6mm shaft offering makes this a lot more appealing for direct-mount drives.

Alternative Motors

1806 Brushless

In the beetleweight class, the 1806 size brushless motor seems to be the de facto standard. The 2mm shaft is compatible with most pinion gears from the various gearbox options, and the default mounting holes match up well to many of the stock adapter plates. Robert Cowan's video mentions the Aerodrive SK3-2118 as another option, but I've yet to see that motor in stock, while a quick search for "1806 brushless" on Amazon, Hobbyking, or Aliexpress brings up a multitude of options (though quality/longevity may vary). Assuming you battle-harden the motors properly, variation in performance during robot combat may be minimal. (Note: again, keep in mind that this is written by someone who hasn't finished building his beetle)

Nerf DC Motors

Brushless (sensorless) motors don't work great at stall or low rpm's, and apparently you can get DC motors that can spin at speeds as high as brushless ones. Out of Darts (and probably other vendors) offer higher quality DC motors in the 370 form factor that use stronger neodymium magnets and subsequently draw considerably more current. I honestly still cannot comprehend that there can be small hobby grade motors of this size that can draw as much as 50A or more at stall. Unfortunately (and ironically), there aren't readily available brushed ESCs that let builders leverage the full power and stall capabilities of these motors. However, maybe that's not too big a deal: in drive applications, we should lose traction way before we even get close to hitting the stall current for some of these higher-end motors.

Stock Upgraded Gearmotors

It's interesting: when I first read about gearmotor mods (probably at least three years ago at this point), you had to Frankenstein your gearmotors from sketchy Aliexpress/eBay purchases and do some custom work on both the adapter plate and motors themselves, but as of early 2023, there are several vendors/builders selling these sort of upgraded gearmotors completely assembled and ready to go:

• Rectified Robotics Brushless Planetary Gearmotor (4mm shaft, 22mm diameter, 1806 2300kV motor, $45)
• OwObotics MKIV Brushless Drive (6mm shaft, 24mm diameter, 1806 2300kV motor, 63g, $55)
• Repeat Drive Max (6mm shaft, 24mm diameter, 2004 2100kV motor, 66g, $55)
• Megaspark Brushless Drive (6mm shaft, 24mm diameter, 1806 2300kV, 71.5g, $55)
• DartBox V2 (4mm/6mm shaft, 22mm diameter, Nerf brushed motor, 57g, $35.5)

It would seem that everyone's converging towards larger diameter shafts and planetary gearboxes, with the primary differentiating factor being how/where weight is reduced. Some options have portions of the gearbox housing milled away or replaced with aluminum (in place of steel) to minimize weight. Seems to make sense: planetary gearsets transfer torque and support the output shaft better, and the larger diameter shaft allows for direct-mounting of wheels without worrying about impacts as much as you would with the (historically standard) 4mm diameter shafts. The larger gearhead sizing increases weight but (probably) allows for a more secure mounting to the side walls and frame of the bot.

Despite all these options, I'd like to note that we've so far only been mixing/matching existing gearheads and motors from disparate (affordable) catalogs. Again, there's nothing really out of the ordinary in what builders are trying in terms of speed reducers, and the means of locomotion remain wheel-based or wheel-ish. There's been some attempts at friction/tangential drive, single-stage reduction via printed or metal gears independent of a standalone gearbox, direct brushless drive (pretty much only in meltybrains), and even "drive-less" setups ([1],[2],[3]). We may be now entering the stage where improving drive means improving the locomotion strategy rather than just the quality of components.

Used to have a dream,
Moving pictures in each home,
Cost both low and high

This is a story of hope, frustration, and sadness. Mostly the last one.

This past Black Friday and Cyber Monday, while most people were probably scouring the web for deals on new electronics, I somehow found myself on Facebook Marketplace looking up cheap broken TVs that maybe I could fix. Maybe I was trying to offload some of my own broken/used stuff at the time, or I was looking for a cheap/temporary replacement for the cheap projector of mine that was effectively useless in the daytime with the blinds open. Regardless of the reason, I saw potential in some of the posts, and a few days later, my living room was littered with 4 large televisions in various states of disrepair, while my Ebay shopping cart was slowly filling up with replacement boards that I hoped would result in making these purchases a steal instead of a very (literally) heavy sunk cost.

To be clear, I was only committed in seeing if the appropriate replacement board could revitalize the TV. I had no intention (or ability) to debug component by component, or do much of anything if the LEDs or LCD panels were broken/cracked/chipped. It still amuses me that there are still people posting TV's with cracked screens onto Marketplace for 100s of dollars. As for everything else, pretty much all TVs seem to have the following (possibly defective) boards: main board (for processing the input signals), power board, LED inverter (to power the backlight), and a T-con board for converting the signal from the main board to the LCD. Based on the (incredibly vague) descriptions on Marketplace, I had (at the time) a fair amount of unbridled confidence that the issue wasn't as bad as the sellers believed.

My judgement is mediocre at best.

TV #1: Hisense 50H6570FA

This one is a bit of a cop-out. The seller just said the remote power button was broken, and he didn't feel like replacing the remote (or didn't realize it'd be super easy), so he was willing to part ways with it for 1/4 the original price. The seller didn't even know at the time of sale that the TV even had physical buttons, so maybe I should've played hardball a bit longer and gotten a better deal, but I figured I'd hedge my bets and snag as-close-to-a-sure-thing as possible while I could. For a 2019 TV, it wasn't a terrible deal beyond taking on the risk that the seller hadn't disclosed all the issues with it.

I pretty quickly replaced the remote for $9 off Amazon, and it's been great, without any further issues to date. However, I did take apart the faulty remote and was a bit surprised at the root cause of failure: apparently some food crumb had gotten stuck under the elastic covering to the power button and jammed against the PCB every time anyone would push the button, and over time, all that aggressive pushing must've damaged the area on the PCB around the power button. You could see an obvious dent on the board, and that must've affected the connectivity on contact. It's not like any remote is IP rated against any significant egress, but death by snack infiltration was not on the top of my list for examples of hardware failure.

That said, I've built and continue to build prototypes that are "sealed" with tape, so I really shouldn't judge.

TV #2: LG 50LN5200

Next up, an older TV from 2013, but one that the seller claimed only suffered from a loose HDMI connector and otherwise was perfectly functional. I should've asked more questions, because if the connector wasn't working, and they couldn't see an image, how did the seller have any idea that anything was still working? Upon closer inspection, the HDMI connector has totally fallen off the board and was just sort of bouncing around inside the TV enclosure. The broken TV was $20, and a replacement main board was under $50, so I was initially pretty excited about the prospect of a quick/easy fix.

The only thing worse than rhubarb pie is humble pie. (I apologize to rhubarb pie enthusiasts reading this)

Screen stayed black after the replacement of the main board, and not just no-picture black, but totally dark without any of the backlights seemingly on. I'm guessing some of the LED strips had gone bad, or maybe the T-con board was faulty, but just the risk of the former no longer made it worthwhile for me to attempt any more repairs. There's actually quite a few Youtube videos detailing LED strip replacement, and a full set would've only run me about $30 more, but that whole process looked tedious as hell, and I didn't have confidence I wouldn't wreck the more fragile LCD in the process.

TV #3: Samsung LN55C630K1F

This bad boy was heavy as hell and didn't come with a mount. The seller said "it got wet" and did not expand further. The TV didn't show a picture anymore, but it seemed like the backlight was fine. It was old (~2010) and risky, but it was also a 55" screen for $25 without any cracks. I can't remember my line of logic at the time, but I went with getting a replacement inverter board first, ignoring the T-con board altogether, and that proved totally fruitless, especially since nothing was evidently wrong with the LED backlight in the first place. When fiddling with the T-con board, I cleaned off the ribbon cables and managed to get a complete image to pop back up on the screen, albeit with a greyed strip overlaid across about a quarter of the screen. That was enough evidence to compel me to get the proper replacement T-con board only to get:

A blank screen.

Don't know if I damaged a ribbon cable during the repeated disassembly/re-assembly (apparently that's a thing too), if something else on the LCD side of the panel was also gummed up, or if the momentary success I had before was just a blip, but I was never able to get another picture to show back up. Similar to the LG model above, the next step of disassembling the back LED panel from the LCD is doable, but not something I wanted to risky further time or money on.

TV #4: Sharp Aquos LC-C5277UN

Apparently my parents have this exact model from 2009 at home in Texas, still chugging away, occasionally basking some guest room w/ its hazy 1080p glow. The seller said it had issues staying powered on but otherwise seemed to work fine before it automatically shut itself down. During (limited) testing, the flashing error code (a rarety in this day and age!) implied something was indeed wrong with the power supply, so I was hopeful that would be the only thing I'd need to replace.

Inexplicably, I managed to order the totally wrong power board on my first try. The correct board didn't make things worse, but nothing got better either. Still the same symptom: power shuts down after a few seconds of operation. My next thought was that maybe the main board was struggling as well and entered some sort of failsafe mode when it detected or thought it detected a fault. However, that line of thinking quickly became moot once I determined I couldn't actually find a replacement main board for sale. I guess 13 years is a bit too long for anyone to keep such an old board in stock.

Thoughts

And so ended my ambitions of being some sort of shady Facebook TV flipper. I came out of this with a 3-year old TV and a bunch of heavy junk I'll need to haul to the recycling center one at a time because I have a Civic. I also came out of this with a lot of thoughts:

1. There's a market for used TV boards. For any given broken TV, even one that looks like it's been to hell and back, it's unlikely that every subcomponent inside is useless. Most are probably perfectly fine and could help repair or elongate the life of some other TV of the same model
2. The market for used TV boards is very small. Even TVs of similar size from the same manufacturer seem to use totally different boards, so what's the incentive for any vendor to keep an item in stock that wouldn't really be utilized until many years after it's first made? In my search of replacement boards, I came across a vendor that claimed it couldn't find the listed product in their warehouse because they moved three years ago. In some other cases, I found listings for components ripped out of broken TVs of unknown failure without any guarantee that they would work.
3. Plug-and-play can be more plug-and-pray. As components get smaller, I feel it's easier to mess up even the simplest of replacements, even as online tutorials for fixing various models get more detailed and commonplace. I may be haunted by one of my previous attempts at fixing my Google Pixel 3A, where I managed to crack the screen and mess up the speaker while trying to replace the charging port, but I do feel some surface-mount connectors are a firm tug away from completely ripping out. In every case where I was cleaning/re-connecting ribbon cables, I never felt confident I wasn't possibly damaging something there.
4. Cost of repair may exceed cost of replacement. I ran into a situation with a broken projector displaying a burnt brown spot where the vendor sent me a brand new projector, no questions asked, and when I asked, very firmly and directly informed me that they do not sell, nor would they sell, the replacement LCD module separately. In that situation, it's possible that the projector vendor was merely a redistributor of a re-branded generic projector, but as devices get more compact, and the assembly processes get more complex (usually requiring special tools), the labor costs alone may exceed the hardware savings you'd get from repairing instead of replacing. That notion is a bit sad, as the economics of this problem seem to incentivize waste more as (ironically) our manufacturing capabilities get more advanced.
5. There may or may not be a market for more interchangeable and easily swappable modules. I'm a bit split on this: would devices benefit from utilizing sub-modules that are easier for the average consumer to swap and replace themselves? There's additional incurred cost in adding the proper connectors and bulk in packaging the additional wiring as well, but surely there's value in making particularly fragile subcomponents swappable like fuses or cartridges. However, how much would we limit the functionality/performance of devices by forcing them to comply with the limitations of certain subcomponents? Would any manufacturers even agree to those design constraints if they didn't have to?

It seems that the most economically viable refurbish/repair operations are dependent on a significantly large number of units sold, and even some of the most successful products may not be enticing enough for the average joe to participate nowadays, and that's a real shame.

Hunger got you down?,
Here's many mini morsels,
Flavor costs extra

I don't remember if I was drunk or just emotionally vulnerable when I bought this dumpling-making machine off Aliexpress in the early Fall, but it finally arrived in the first week of December, and it was quite the early xmas surprise. I'm astounded it did anything, much less produce dumplings that are actually edible, and honestly, not that far off from the frozen packs of industrially mass-produced dumplings you'd snag out of the freezer at 99 Ranch. However, I'm also fairly confident that it won't last too many more uses, not that I'd really want to use this on the regular unless I had a repeated need to pull out a party trick.

Actually, we should ask: who is this product actually for? Someone or group developed (and presumably tested) this, designed it for assembly/production, and seemingly has produced enough to keep it listed on Aliexpress. Surely they can't rely on schmucks like me who like teardowns and foodtech to hit their sales goals. Who is looking for a tabletop, hand-crank mechanism that produces super doughy-dumplings not all that much faster (if at all) than your average Asian? I think you could even argue that an individual dumpling folder would provide better throughput and quality. The only scenario I can think of where this device would provide dividends is if someone who has never made dumplings before was tasked with making a significant batch for someone else who's never had dumplings before or simply has no preference for taste.

(If it's not clear: no, you most definitely shouldn't buy this thing either)

As has been the case with most things I've torn down off of Aliexpress, the product market fit is questionable (if it's existent at all), and the usefulness is negligible at best, but the mechanisms are pretty neat. A single crank couples the motions of a piston pushing the filling through a central conduit, a pair of augers/screws that extrudes a (somewhat thin) tube of dough around the filling, and a rotating "stamp" of sorts that punches out dumpling forms out of the aforementioned filling-dough sausage that's being pooped out of the machine. The crank also rotates a 'flour' duster that is supposed to prevent the dumplings from sticking, as well as a conveyor platform of sorts to avoid needing the user to manually separate the dumplings as they come out, but neither were all that effective and are probably best left out of the machine.

Operationally, I had a great deal of problems ensuring that the dough was adequately filling the auger/screw mechanism. Whereas the filling hopper has an even pressure applied from above, reliably funneling the filling through the central tube, it seems that the operator needs to manually ensure that the there's sufficient dough being extruded to keep pace withe filling's dispense rate. I wonder if a softer, wetter dough than what's typically used for dumpling wrappers would work better, as the screws by themselves do very little to pull in the dough. The dough extrusion mechanism forces the dough through a fairly small annular cross-section around the central filling tube. I'm surprised the plastic screws/augers are able to force dough through such a small opening, and it makes cleaning/maintenance afterwards a total pain. The tip of the dough-extrusion structure is a three-piece assembly with captured insert and screwed retaining cap. Due to the fastening constraints, there's a discontinuity in the dough flow introduced fairly close to the endpoint, which I think makes it more difficult to ensure a consistent extrusion.

For the filling, a plastic plate driven by a plastic screw, coupled to a series of gears connected to the primary crank, presses down on the filling and extrudes it through the central stainless steel tube. I'm not sure if the central tube is metal because that's the preferred low-friction material for dispensing foodstuffs or if injection molding it as part of the upper hopper would've actually been more cost-prohibitive. I also wonder if there was a similar reason for the taper around the central tube being so shallow, as that (intuitively) seems like quite a bit of excessive force being placed on the flimsy central plastic screw once the filling level gets low. To be fair, dumpling filling tends to be both quite loose and slippery, if you respect and embrace the joy that pork fat and sesame oil grants us.

The stamping sub-system acts like a well-oiled (if you oil it) and meticulously-timed sphincter in our dumpling-dumping machine. The rotating stamp has cutouts in it and spins against a curved steel surface, pinching the stream of dough and filling twice per rotation.  The dumplings are formed longitudinally, if that makes sense, with the spacing between roller surfaces set precisely to compress the dough into sealed edges. This means you don't get the folds/creases or plump body of a well-made, traditional dumpling, though I wonder if the rotating rate of the drum and the cutout geometry could be adjusted to better give you that effect. Similar to industrial bao-making machines, the end result seems to be mostly at the mercy of the incoming extruded dough/filling sausage.

The overall construction doesn't exactly scream "industrial" or "dependable." More like "adequate" and "not terribly flimsy." Most of the gears, even the ones that would never come in contact with foodstuffs, are plastic, rotating in loosely toleranced holes in the surrounding plastic frame. I think I found two metal bearings in the whole machine, neither of which were sealed or really greased. To be fair, the shipped instructions do warn the user to avoid ever submerging the entire device during cleaning, as that would damage much of the metal components. The extensive use of plastic throughout and the lack of proper bearings (for both support and low rolling friction) didn't inspire too much confidence with regards to the machine's longevity, though maybe there's a loose argument that replacement parts and repairs are easier to come by.

Nah. It's just cheap, and the manufacturer's trying to save a buck (as they should).

Taking a step back, this machine's a nice encapsulation of the tradeoffs and process adjustments we typically would need (or at least prefer) to enable automation of a manually-dexterous task:

• Maintaining a consistent rate or ensuring coupled rates of operation is easier than triggering specific, discrete events. This thing doesn't fill, fold, and dispense individual dumplings; it crimps subsections off two parallel streams of dough and filling being pushed out at the same rate.
• Compromises need to be made in quality due to changes in production methodology. The dumpling wrapper's an extruded consistency, not rolled/flattened/compressed. Wrapper needs to be thicker so that the machine doesn't tear it while extruding. The dumpling body can't be as full and plump as traditionally made for the same reasons. No crimping or folds on the edges
• There's a good amount of prep/preloading/priming needed to get everything flowing in sync. There's always some amount of dough and filling stuck midway in the machine once we've exhausted our ingredients. Once the ingredient packing drops below a certain point, the extrusion/dispensing becomes inconsistent, and without active feedback (which this gizmo certainly lacks), there's no way to compensate.
• The user gets stuck with a set of new tasks that otherwise wouldn't be necessary. In this case, cleaning the machine is a huge hassle. Much of the system needs to be totally disassembled and then soaked to make sure food bits aren't left behind, and we also need to take special care to not get the rest of the machine wet, since the metal bearing components are left mostly exposed.
• Friction will get you one way or another. We either had to oil the dumpling stamping unit or keep the extruded dumpling coated in flower for them to reliably separate and drop out of the machine. Otherwise, we had to manually poke/prod the dumpling as they formed to ensure they'd separate. Either approach requires some degree of monitoring and upkeep, and neither was something I particularly wanted to deal with.

Sudden closing thought: was this like the equivalent of an Easy-bake Oven for 30-something Asian engineers?

Cyclones a'swirling,
Round and round the small bits go,
A crowd is welcome

Been struggling a while now at work trying to a consistent vacuum flow working in a dirty environment. Most vacuums (that I've come across, set up, and tested) pull air through some filter that captures the dirty particulates to be captured. The filter both protects the filter and stops any of the particulates from escaping back into the air. As much as we'd like to ignore the agonized squeal of the motor in the un-serviced vacs we bought in the happier (and cleaner) past times, filters get clogged, decreasing the airflow over time. For a regular shop vac in standard, intermittent use-cases, that's unlikely to be a huge deal, but when utilizing these suckers continuously for much smaller particles, performance would be irregular at best even with frequent maintenance.

Enter the cyclone separator: an add-on unit between the inlet and vacuum motor that pulls bigger particles out of the air stream before hitting the filters at all. It's typically a conical unit pulling air flow through the center of the column, with the inlet stream entering the cone in an orthogonal direction, tangential to the curve of the cone. Combination of the conservation of mass and the weight of the particles directs them downwards and out instead of up through the central airflow to the motor. The most basic commercial offering is probably the Dust Deputy, but many variations exist, with some striving to minimize the space occupied by the add-on [1][2] or interface more easily with popular vacuums. The addition of the cyclone separator can end up being quite large and are more often found in stationary setups in woodworking shops.

Even though sawdust particles are quite small, and woodworking cyclone separators are optimized for such particles, they're not foolproof, as evidenced by the retention of filters in the vacuum subsystem, in handheld and stationary systems alike. This seems to be the primary struggle for so-called bagless vacuum cleaners, which either use a removable rigid filter (that can and should be cleaned regularly) or in Dyson's case, a multitude of small cyclone separators packaged as densely as possible. James Dyson perhaps summarizes their innovations most succinctly: they've miniaturized cyclone separation technology to maintain airflow/suction when picking up dust.

Other videos showing off more details on multi-cyclone separators:

• Dyson Root Cyclone w/ Core Separator
• Explaining/Showing Multi Cyclonic System Inside and Out
• Dyson Cinetic Technology is Amazing
• The technology behind the Dyson DC59 cordless vacuum cleaner
• Dyson 2015 | Dyson Cinetic Canister Vacuum Cleaner | Cinetic Science Explained

This thing is really small when compared to the cyclone separator add-ons you'd find at Home Depot. Granted, it's not meant to fill up barrels in a workshop and needed to be handheld, but it's still pretty spectacular the functionality that they've managed to shrink down into these collection of funnels. There's a total of 15 mini cyclone separators, surrounding a central filter sitting between the separator outputs and the vacuum motor. If you stare at the central axis of the canister, there are effectively 'zones' going from the outside to the center:

1. outer suction zone for large particles blocked by the metal mesh
2. inner suction zone leading to the small tangential ports of the cyclone separators
3. exit annulus for particles coming out of the mini cyclone separators
4. sealed channel for the vacuum source leading to the top of the separators

The air path is pretty convoluted, and I wonder if that's a consequence of the packaging requirements or if the extended effective length helps even out the airflow across the multiple separators in some way. The air first enters the dirt canister region at an offset, tangential to the outer contour, on the outside of the metal filter. On the interior of the metal filter, there's an annular region leading to a void that connects to the each of the tangential inlet channels for the conical separators. Each conical separator has a top outlet port that connects to yet another void in the center of the canister that leads to the vacuum source, sealed with a rubber grommet of sorts. The tip of the conical separators all dump into another annular region that connects back to the dirt canister volume.

The path to the inlets of the cyclone separators seems particularly messy and haphazard, which (to me) is further evidenced by how easily the dirt/powder builds up around the conical tips. I suppose that maybe the manufacturers are ok with the dust getting caught prior to the separators, since that still keeps the particles in the canister and doesn't clog any of the filter elements that may compromise the vacuum air flow? I would've loved to see the earlier prototypes that (presumably) had less tortuous air paths.

The injection molding and custom gasket work is pretty intense, so perhaps that explains we don't see this style of separator too often elsewhere, but it does make me wonder how far we can take this technology by continuing to make the separators smaller and more numerous. Seems like an interesting use case for 3d printing if only for testing/evaluation, and intuitively, I feel that we can get the performance a lot better with smoother/straighter airflow paths that don't need to be compromised to satisfy packaging constraints.

2022-11-17 EDIT - Snapshots from Patents

This teardown wasn't of some random Aliexpress purchase, and there's an opportunity here to go through some nifty patent images for better insight. It appears to me that the Dyson separator arranges their chambers a bit differently from conventional multi-cyclone separators.

For conventional multi-cyclone separators (patents US7771499 and US9016480B2, for example), it appears that the cyclone sits in serial between the inlet and the vacuum, meaning that whatever is picked up passes through the separator entirely before falling into the waste chamber. The flow pathway makes intuitive sense where the outlet of the cyclones are unlikely to get clogged.

However, in moving to smaller and smaller cyclones in an attempt to capture smaller dust particles without the use of a bag/filter, there's a higher chance of the cyclones getting clogged, despite the fancy vibration analysis Dyson mentions in their Youtube videos above. As such, it looks like there's a bit of a tortuous path added between the inlet to the separators and the waste chamber, so that bigger particles can skip the cyclones altogether. Patent US9603498B2 seems to illustrate this the best. As long we maintain enough suction that provides a net pull on the dirt we're picking up, the tortuous path around the outermost metal mesh seems to be enough to knock the bigger particles directly into the waste chamber by leveraging the momentum of the dirt particles in a similar way to how the cyclone normally work.

Bye bye octopi,
Your appendages can go,
This shell is sure swell

Automation often gets associated with robotic manipulators, especially the serial-chain variety, which also forms the basis for any intro robotics course. Arms and AGVs probably comprise the core components of nearly all robotic automation setups. It is the platform on which we place our end-effectors as well as our hopes and dreams. Companies like UR have made it increasingly easy to plug-and-play arms into various roles previously reserved for humans, and despite the (imo) market domination held by the major manipulator vendors (e.g. UR, Fanuc, ABB, etc), as well as the ever-growing list of failures (e.g. Rethink, Carbon, Innfos, Pneubotics), we continue to see new players developing new iterations on the humble kinematic chain (e.g. Ally, Flexiv, Dobot). Aside from a race to the bottom in terms of price (that somehow magically doesn't compromise key performance capabilities), new arm offerings strive to reach for more compliant/reactive capabilities, higher payload, improved usability, and novel form factors. The arm seemingly remains a key component for positioning your tooling in your workspace.

I've dealt with a range robotic arms for the whole duration of my career now (not sure if I should be grateful), dealing with everything from integration, motion control, and basic planning to routing air/electrical lines, mounting custom end-effectors (so many end-effectors), and long-term mechanical maintenance/repair. With each arm, there's been new interfaces, both hardware and software, that needed to be understood and then adapted for the target task. The idiosyncrasies in terms of reachable workspace and joint constraints are different enough to be frustrating, but consistent enough to be expected. Despite all the setup and resources we pour into building the application solution around a robot arm, I would argue that it's woefully under-utilized in most cases, and we should probably replace it with far simpler positioning devices, if only they were easier to customize/construct.

Do We Really Need 6DOF

I'm not convinced that the majority of applications employing robotic arms need the full workspace that a typical 6-dof serial chain arm provides, but let me be clear: the capability to position and pose the end of a manipulator in all 6 degrees of freedom is substantial, and it's not something that can be easily re-added or bandaided into existence on top of other motion systems. That reason alone may be sufficient justification for integrating a robotic arm wherever possible. I'll admit that I'm still a bit tainted by academia, where we're always trying to reduce the dimensionality of the problem. Entire theses have been written and will continue to be written on how principal component analysis (or some variation) reduces a super complex problem into something way more tractable (under certain conditions, blah blah blah).

That said, I think we would still mostly be in agreement in saying that you'd be hard-pressed to make up an application where you absolutely needed the robot arm to exhaustively move through its workspace in 6-dof. You may, with some difficulty, conceptualize a task where you need a highly dexterous arm, especially if the environment is dynamic. However, I'd ask: at what point does that operation venture into the nebulous world of 'general purpose' tasks, something so vague that by definition you need the full workspace range at your disposal? Once the task scope begins to crystallize and narrow, I'd argue that the utility of a 6-dof motion platform starts to quickly depreciate.

Palletizing and Container Handling

Boxes or box-like things meant to be stacked on pallets seem highly structured when compared to other candidate objects in more generalized pick-and-place problems, with consistent reference surfaces and stable resting configurations. In their typical use cases, they're likely to be well organized or at least singulated as well, so why would we need 6-dof arms to handle these things? Whether it's loading a pallet, re-organizing racks of totes, handling trays, moving luggage, mail-handling, or other similar applications, the motion itself seems predominantly (if not exclusively) cartesian, with the end-effector approaching from the top or bottom, and never needing to tilt the box itself in any way. There are even 3rd-party attachments sold to "fix" the payload-limitation issue of serial-chain arms. In a lot of these use-cases, we could probably even get away with chutes, and it seems like the primary benefit you get from serial-chain arms is the slightly more compact space usage.

The more "exotic" box-handling solutions are even more confusing to me. I don't understand the advantage that a humanoid like Digit or Atlas has over more conventional solutions. Even the more simplified/industrial implementation from Dexterous Robotics has me scratching my head. Boston Dynamics' Stretch implementation may provide benefits in terms of payload and speed, but is it still worth the complexity? How much additional utility do these solutions offer beyond the humble forklift?

I'm a much bigger fan of more "dedicated" box-handlers, so to speak, like Invia Robotics or Hai Robotics' solutions. Those seem like a much better middle ground between a complex manipulator system that is able to handle more chaotic package clutters and the complete "robotic warehouse" solutions like Symbotic and Ocado. At the smaller scale, more established examples of pharmacy automation have seemed to learn more towards the "smart vending machine" with gantries and custom shelves to dispense pillboxes instead of larger setups with industrial arms, though demonstrations of the latter still pop up from time to time.

I would actually also bucket the majority of the recent wave of food-robotics setups into the container-handling space. In my opinion, very few of them interface with the food directly; most attempts just move some container (e.g. tray, bowl, wire basket) between standalone dispensers and different heating zones. For example, I don't see why functionally, the arms of CafeX and Bobacino couldn't be replaced by the automated drink conveyors already developed by Miso (Sippy) and Cornelius (Automatic Beverage System), the latter of which has been around for ages. This isn't to say that all startups in that space lean a certain way: whereas companies like Pazzi and Hyper Robotics use 6-dof arms to handle certain tools, we see places like Stellar use more bespoke conveyors and gantries.  

Bin-Picking and Stowing

Bin-picking, especially with objects in clutter, presents a more complicated challenge, albeit one that predominantly only needs an approach from the top. For me, the key cop-out distinction that should be kept in mind for this task is that we only need to somehow attach the object to the end-effector, not necessarily in a particular relative pose. Now, there may be a limited (or perhaps only one) relative, pre-grasp pose in which the selected end-effector can reliably pick the object, but in my opinion, that particular scenario is far less likely with the move towards compliant suction cups with adaptive bellows. Somehow connecting an object to the end-effector doesn't always necessitate a secure and locked relative transform between the two.

Take your pick of some of the current stalwarts in the bin-picking space: Righthand, Berkshire Grey, Soft Robotics, Nimble AI, Kindred AI (and a whole bunch more I'm missing). For all the talk of optimized approach vectors, I'm not convinced that jamming a compliant suction cup into the object's center of mass from directly above wouldn't return a similar if not the same end result, given that the demonstrated results are pretty much just that. I have zero numbers to back up that accusation, but I hope the reader can accept my point that for the dominant, semi-"unstructured", commercial task that's challenging the robotics community at the moment, the required range of motion is fairly minimal, if not only Cartesian. Also, because it's convenient for my argument, I'd like to point out that while the winner of the initial Amazon Robotics Picking Challenge used a 7-dof Barrett WAM, the last winner used a far simpler, custom gantry of the team's own design. I think it's also interesting that the problem of sorting recyclables/waste, arguably a more unstructured problem (with less object knowledge know a priori) than bin-picking, has stuck with limited-mobility gantries over 6-dof arms, as shown by implementations from AMP Robotics, Glacier, and Waste Robotics.  

Part-handling and Assembly

I'm a bit conflicted on whether we should consider part-handling and assembly a more difficult robotics problem than bin-picking. On one hand, scope is fairly limited and usually the entire problem scene is well modeled and known ahead of time. On the other hand, the object pose relative to the end-effector is now significantly more critical, and the task may only be completable within a very narrow band of system states. However, a high degree of motional fidelity is not the same as range. As the integrator generally has significant control over the design of the support structures, functional steps, and additional tooling, the kinematic flow of the target part can be made to be incredibly simple, often locked to a single or limited set of nominal orientations.

While it's true that robotic arms may place components at some angle or perform some minor adjustments as part of the assembly motion, the trajectories are most commonly run open-loop, with end-effector modifications handling any slight deviations from the nominal trajectory where necessary. Even in applications like applying adhesive/primer to a vehicle windshield or de-flashing the plastic from an injection-mold part, the arm's primary motion should remain identical (or identical with some simple offset).

A lot of assembly demos you can find online nowadays largely show a robotic arm responsible for shuttling the part between various stations, each of which is custom-built for a particular task in the assembly sequence. The individual stations typically need alignment features and guides to ensure reliable operation, so the robotic part-handling element has some allowable error in dropping off and picking up the parts in question. We typically see gravity (or some other constant external force like a driven conveyor surface) leveraged to further drive the part against physical hardstops to better guarantee the system state, avoiding the need for the robotic subsystem to actively detect and adjust for errors mid-task.

What really troubles me (as much as this thought experiment should trouble anyone) is that we've had highly complex manufacturing assembly lines devoid of robotic arms for ages now, and as far as the fundamental kinematic motions are concerned, I'm at a loss for explaining what significant benefit having a serial-chain arm brings to the task. I feel that part of a possible explanation is that we have a tendency (and aptitude) to break down tasks into simpler motions that typically end up being rectilinear or grid-like. Take additive manufacturing for example (maybe one of the more freeform robotic assembly processes): three-dimensional shapes with any sort of curvatures are simply resolved to stacks of 2D contours without (arguably) significant loss in the end result.

Now, does this mean that we, as the literal puppet masters of these high-dof positioning tools, lack the imagination to maximize our tools' potential? Or are physical tasks really much simpler than we initially assume?

Where We (May) Need 6-DOF

The easy cop-out answer (for me) is that when the objects and environment gets increasingly unstructured, or our system state is especially prone to errors, increased kinematic reachability becomes a lot more necessary. Mobile robotic platforms like TRI's system, TIAGo Pal, or any of the DARPA robotics challenge entrants probably would not get very far without dexterous manipulators. The arm motion in those cases (I think) are necessarily predicated by the system's estimation of the world state and need to compensate for the mobility system's shortcomings. That said, I suppose I could point to Kevin Robot (built on Care-O-Bot's base) and Hello Robot's Stretch as counterpoints where the platform makes do with much simpler arms by either limiting the scope of the problem or leveraging the mobility system as part of the manipulation task.

Beyond adaptive motions/tasks, I would be excited to see applications that employ the same arm for multiple operations in a single setup, to maximize the utilized uptime of the machine. Even in manufacturing lines running 24/7, it's not uncommon to see arms/gantries paused between steps. Various vendors sell multi-gripper setups, primarily to increase throughput, and many systems have implemented multi-tool setups to avoid multiple arms or tool changers in executing more complex tasks, but I don't think it's too much of a stretch to shift an otherwise idle arm to a secondary task. In that scenario, at least we'd be utilizing the arm for multiple toolpaths.

However, I'd still say that sensory feedback is key to fully utilizing the range of motion that high-dof arms have to offer. Several RaaS startups in recent years have focused their offerings on high-mix/low-volume welding, polishing, grinding jobs: contact-rich tasks that ideally require active probing during execution. These examples require various degrees of motion compensation that may be difficult to predict a priori (maybe less so in the welding case). I also feel that the auxiliary motion needed in these operations (and I'll admit that whether they're truly needed can still be open to debate) should be driven by the sensory elements themselves, not an arbitrary offset or buffer decided by the human operators before the task starts. At the end of the day, these sort of tasks (imo) just aren't that prevalent yet, and so I continue to be more than a bit befuddled that integrators don't do more with modular linear robots or reconfigurable gantries.

Peer through this here lens,
See the change you can truly be.
Distortions enhanced

I should have known better, but a recent late night trip down the Aliexpress rabbit hole landed me on a few suggested pages for handheld mini/pico projectors, the YT200 in particular. This thing is shockingly tiny and only $30, so I made it one of my indulgent purchases for the primary intent of tearing it down. A cursory internet search returned nothing on the company/logos in the product description, and there didn't even seem to be any reviews for this thing beyond the smattering of remarks on the Aliexpress page. While it certainly is still most definitely a waste of $30, I was impressed by what it could deliver for that cost at its size. It certainly shows a recognizable image (on a clean wall, in the dark) while drawing less than 10W, and despite the lack of VGA/HDMI/DVI ports, it plays a reasonable range of media filetypes off of connected flash drives. I'm hopeful that this thing supports DisplayLink and can be connected to an RPi in a compact Retropie or projective mapping project, but even if it can't, it's a cute, easy-to-pack toy that can run off most handheld power banks.

As with a lot of off-brand projectors, the listed specifications are laughably misleading. The little guy produces a 180p (no, not missing a 0), low-brightness image. Ignore the advertised performance pictures or other claims. One of DIY Perks projects demonstrates how this type of projector works: A super-bright LED source (collimated by some fresnel lens) shines through an LCD panel, and then a simple stack of mirrors and lenses projects that image onto the screen. It's really the old school, overhead, transparency projectors getting re-packaged into as small of a container as possible. If you remember from school (and a shockingly increasing number of you won't), those overhead projectors didn't work all that great without the room already pitch black, but hey, for sub-200 USD projectors, the end result isn't all that terrible.

The YT200 is that same basic layout shrunk even further down. The construction is pretty cool. Most of the internal pieces are slotted in place, sandwiched by the two enclosure halves. The single electronics board is actually the only internal element thing bolted to the enclosure, and I'm a bit surprised they didn't just add some plastic stakes in to pin that down along with the other components, though maybe that's preferred to give more stability for when you plug and un-plug peripherals into and from the board.

The cooling on these models leave a bit to be desired and (apparently) is the primary source of noise for projectors in general. Some DC fan or blower is usually positioned to blow onto some set of heat fins connected to the light source, whether that's a high-power LED in this case or an actual bulb in others. For projectors like the YT200, the fans draw in air from the opposite side of the machine, across the LCD and fresnel lenses, requiring the intake to go through a tortuous path along the interior of the enclosure. It seems like a great way to pull in dust and contaminate the lens, as well as the already-warm air inside the box. The heat sink in this machine also wasn't much to write home about: it's thicc in all the ways we didn't need in this product. The temperature-threshold shut-off sensor is pretty nifty though, even if it's only sorta, kinda jammed against the heatsink and not actually secured in place.

I also really liked the focusing lens, if only because I associate lenses and focusing with precisely-fixtured, polished pieces of glass, and YT200 is yolo with some plastic lens mounted in an injection-molded spiral guide. There is just barely enough friction to ensure that the lens doesn't regularly move out of focus, though admittedly it's hard to tell given the stunning paucity of resolution in this device. These sort of use-cases in products have always interested me: you can seemingly get away with a highly unreliable mechanism when manual adjustment to an arbitrary metric (how sharp the image appears to the user) is accepted to be part of the initial setup (and subsequent re-setups).

Without being able to reliably track down the details for the electronics boards and sub-components, I'm not sure how I can hack these things beyond their actual, OG use cases, as limited as they may be. I keep hoping I can stumble across some gizmo that has excellent re-purposing potential in some other project, but so far, it's looking like I'm the one getting repeatedly played by the Aliexpress algorithm.

Dampness underfoot...
That a soggy rag I see?
It's a clean skid mark!

I should preface this post with two main points:

1. No, I did not get any compensation from any person or entity for writing this post. This post only reflects my personal opinions and is not in any way sponsored
2. No one should ever buy the ALEE mopping robot. It is terrible.

I really enjoy watching and doing teardowns of random stuff. Two of my favorite Youtube channels are AvE and bigclivedotcom (AvE's Juicero video is amazing). For electromechanical stuff, it's always a great learning experience to see how engineers designed for assembly, wiring, and actuator selection. Even for simpler for stuff, it's cool to see how the design cut corners and squeezed out more functionality with less. Products off of Aliexpress are great examples of that last point. Funds permitting, I'm hoping from here on out to regularly buy stuff that catch my eye and break them down to see how they work.

The ALEE mopping robot is a mobile unit that drags a wet rag on the floor, available for $85-ish USD on Aliexpress. There's a water reservoir you fill up that slowly drips onto a cloth sleeve covering the bottom of the robot, and when you turn it on, it seems to bounce back and forth across your room similar to how the OG roombas used to pseudo-randomly traverse rooms back in the day. At the time of purchasing, I didn't see any obvious sensors mounted on the robot, and my limited research only turned up this little wheeled module as a replaceable drive unit. I had originally thought that maybe there's be some nifty interior suspension for bump detection, or some utilization of the active rotation of the mop as part of the drive. The truth is a lot dumber but also (imo) a lot more amazing.

The "Cleaning" Behavior

The Guts

There's one motor. That's it, and it's a teeny motor without any encoders. The drive module isn't some sort of tiny differential drive with concentric shafts or anything. It's just a bevel gear transmission that runs both wheels at the same rate. The detachable wheel-pair and actuator are mounted to a central panel that's spring-loaded vertically at four points relative to the outer body. The outer body enclosing the water tank and electronics can only translate slightly up/down (with minimal parasitic motion in xy due to assembly clearances). There's no rotation of the outer body and drive assembly, though the wheel-pair will spin freely when the wheels aren't contacting the ground. The actuator itself has an initial belt drive but then a standard spur drive transmission afterwards. The belt presumably is to absorb some of the impact when the bot body bumps into stuff.

The Drive Mechanics

I should note that I'm guessing at some of the behavior here. Like I mentioned above, there's no bumper or proximity sensors. In playing around with the wheel-pair, you can (sort of) see below how it changes direction once you forcibly stall it. My guess is that the "navigation" of the bot, so-to-speak, is limited to changing direction once a current spike is detected. That's all the control board does (aside from playing back a cute intro dialogue in Chinese).

So how does this thing do anything more than just run back and forth? Three points make a plane, and we have at least two reaction torques at work: the torque from the two wheels rotating, and the torque at the bevel gear. If I were to hold onto just the wheel pair and not allow it to touch the ground, the entire body of the bot would spin at the point where the wheel pair attaches to the body. With the wheels on the ground, the bot body is tilted back due to the drive wheels to form that 3rd point. Assuming the wheels don't slip, there will be some net rotation of the bot body relative to the direction of the travel dictated by the wheel pair, and that gives us the primary "cleaning" motion of the cloth pad. The four springs change how the two wheels grip the ground during collisions, which presumably is sufficient to generate some minute turning behavior such that this thing doesn't just bounce back and forth in a straight path between walls.

It's fascinating that this thing is able to achieve something that even somewhat resembles a cleaning/scrubbing motion profile with just a single motor. I'm both astounded at the cleverness of the mechanism and disappointed that I so easily shelled out $85 for it.

Again, don't buy this.