Archive for the 'Roll Cake' Category


Dragon Con 2018: Postmodern Robotics Comes Home With a Vengeance

Nov 08, 2018 in Bots, Dragon Con, Roll Cake

Coming hot off my end-of-July trip to the mountain range I keep seeing next to I-81, and aiming to complete my original intended run down to Atlanta via the Blue Ridge Parkway, I decided to do Dragon Con a little differently this year.

Normally I’d be transporting down a whole production of robots from everybody and myself. Life, it seems, hits everyone eventually in the most inconvenient of ways, and a lot of my usual crew couldn’t make it (it seems startups are the robots and elaborate cosplays du jour).  And so this time, I decided early on to stop any plans for developing the next Überclocker /30-haul during the 3 or so weeks of August I had, and only bring Overhaul for display and what of Roll Cake I could get together. I wouldn’t even pack a Markforged printer this time! Imagine that.

So really this was one of the lightest loadouts I’ve done for Dragon Con, ever.  This was actually going to be different for me, and I wasn’t quite sure how to feel about it: Taking an eternity to get down there, actually maybe doing con stuff for once, and only whipping out a robot when convenient. Who the hell would ever do THAT? Go to an entire 100,000+ attendance convention to do stuff that’s not build robots!?

I did a speed-run departing Monday night before the con (Now featuring actual speed!) down to the area of Fancy Gap, VA where, the next day, I jumped onto the Blue Ridge Parkway. I was more interested in the mountainous portions in North Carolina, more so than the scenery itself, so I elected to skip the portion in Virginia as well as not run the Skyline Drive park again. There were plenty of opportunities to take Vans Next to Nature photos.

This was somewhere in Western North Carolina. Where? Hell if I know. The group of motorcyclists in this same parking area didn’t seem to know either. It seems you’re not supposed to know, or pay attention, or care at all; just disappear into the woods and assume you’ll pop out the other side at some point.

I get it. Not as much as someone who actually likes Naturing would, but I do understand.

As night descended on Tuesday, I decided to call a stopping point in Little Swaziland Switzerland, a mountain resort town. This region is very popular with motorcycle tours, as can be seen with Mikuvan’s new and temporary friends here. When you need the size of a van with the cargo capacity of a motorcycle…. Polaris Slingshot. By the way, the NC222A loop around this area is absolutely fantastic. I’m sure it gets tiring if you have to drive it every day to go to work.

I lied. While I didn’t bring along a Markforged printer, I sure as hell did bring a printer in general! I grabbed my “derpy van of 3D printers” Flashforge to make some parts for…. something, on the way down. Yes, I ran it overnight in a hotel room. I ended up having to build a pillow fort around it so I could actually sleep.

By mid-afternoon Wednesday, I’d reached the outer limits of Asheville and….. decided to tap out.

Keep in mind the Blue Ridge Parkway are all very winding, slow 2-lane roads, and that’s no way to cover distance effectively. At some point, I actually had to reach Atlanta and check into my AirCNC.

I decided to express the rest of the way after a harrowing foot-to-the-floor 55mph 4000RPM, 3rd gear climb up the side of Mount Mitchell for what felt like 10 minutes straight. If there was one moment that I was going to blow up my freshly rebuilt engine, I felt like it was going to be right there, but it would have been worth it. Worse, I was low on fuel at that point – there are no gas stations on the BRP – and feared that the straight uphill pull was going to get cut short. I basically coasted down the rest of the way into Asheville to try and conserve fuel.

From Asheville, I took I-40 west until US-74, then followed that down to US-23 – a well known historical path of mine, which I followed to basically outside my old front door…..


…step? Nah, in the intervening year, my parents sold the house. I instinctively glided into my favorite gas station from throughout high school and visits during my years at MIT, on Exit 111 on I-85, but realized I had no more business there afterwards. That was a strange moment indeed.

So, onwards we go to my aircnc house in downtown Atlanta. My local chariot was awaiting as soon as I got there! How positively quaint. There’s been a ton of fuss around Atlanta about scooter rentals lately. I mean, my whole goal of getting a place this close to the convention – right over the Downtown Connector in the “Most Boston part of Atlanta”, the Old Fourth Ward, was so I could just (as we memed it) “Millennial my way over” – whether that means rideshare app or silly scooter rental.

To be entirely based: I completely support silly scooter rentals. The only reason, in my mind, that they don’t work is because governments have outdated patchworks of laws regarding vehicular traffic that isn’t private passenger cars – our current society arguably came of age with the expansion of suburbs centered around the private car, and legislation has ossified around this concept to the point of being cancerous, just like infrastructure spending that is continually strongly biased towards private passenger car use. Much of the battle over scooter rentals is what kind of vehicle to tax, title, insure, and operate them as, and the context of legislative preference for FMVSS-certified normal people cars is unavoidable.

Fight me – I own several shitty cars and several more shitty scooters.

It would also help if people didn’t throw them in rivers.


Hi. (灬♥ω♥灬)

Who I think were the owners of this Model 3 were watching me very intently from a restaurant outdoor seat, so I didn’t get any closer. I was otherwise going to get within an inch for this photo op.

Thursday Funday is over – onto the con!

Overhaul was going to live in the Robotics track room for the whole weekend, so I did that unloading Thursday night when everyone was getting in the area, so I had backup. It would be rolled out for the Battlebots-related panels and otherwise hidden under a tablecloth. It was part of the Battlebots watch party and the “How to Get on BattleBots” Q&A session.

So, how do you get on Battlebots? Well, hell if I know. It’s clear your robot doesn’t have to be good.


That night, I finished the last print at my AirCNC house (whose host said “Would the neighbors mind if I ran a 3D printer all night?” was the weirdest request he’d gotten). And what I’ve been printing all along were in fact parts for my Overhaul cosplay.

It’s a little known fact that Haru-chan, like all good characters, has a male analogue. We actually have a rough sketch of what I call “Haru-dude” made by Cynthia (Lushanarts) from after the rework of Haru-chan:

But he’s too hot for me to pull off, so I simply used it as a design guide and changed up a few things (and also making it more realistic to put together). The “Overhaul head scythe” would obviously not be happening for Dragon Con, but I’ll consider it for something in the future.

I basically modeled and printed the whole thing on the fly, literally during the trip down and on site. I brought a handful of things I figured would be helpful, such as Velcro straps and hot glue and the like. Most parts were not actually modified from the CAD models of Overhaul parts, but made from scratch to exaggerate certain features that would appear too small if I just printed wheel_hub_assy.iam.

3D printing: Replacing the time-honored artform of hand-crafting costume pieces with on-demand kitted disposable bullshit, as-a-service, just like every good millennial trend!

So here we are.

Oddly enough, I was once again behind the camera 99% of the time and only really got this one photo taken of me (credit to Aaron Fan). Oh well, it was the prototype anyway. I actually don’t know of any full-body photos at the moment.

Besides the wrist shanks, I got a gray utility vest and added button snaps to reflect the staggered bolt pattern on the frame rails. I made two “edgelord belt chains”, one out of actual number #40H roller chain from Sadbot’s pokey stick) and one with orange wire loom wrapped around it.

I also made some cartoony wheel sprockets (which are hijacked #80 sprocket models, because again, teeth that are realistic can’t be seen in real life) to clip onto my repurposed motorcycle boots which I used for “dude-Ruby”. I had a few different shades of blue going on, which I’d like to fix for a future more proper rendition.


Saturday night robots with a few more of the crew. A couple of AirBnB houses around the area definitely suffered some robot building shenanigans. This was primarily to finish the 30lber seen on the right, but also to work on beetleweights because Sunday SUNDAY SUNDAY is the Robot Microbattles!

Likewise, after I got done working on other peoples’ robots, I put the last few solder joints in on Roll Cake, then test drove it in the hallway (“Do you mind if I test robots indoors?”). It’s quick, but controllably so, and I was fairly comfortable handling it after a few minutes. I like this new drivetrain a whole lot – the slight lag and torque ceiling of the hub motors is completely gone. It’s almost too twitchy now, and I know for a fact the drive motor size can come down to the next smaller outrunner class and be fine – this will be on the docket for a revision.

Now, back to the post title. I said there was a lot of postmodernism in the robots this year. What do I mean by that as applied to robots?

Postmodernism, broadly speaking, encompasses schools of thought which criticize traditional rationality and notions of objectiveness, calling into question the nature of what we call objective truth. In its basest form it often revolves around the ironic deconstruction and decontextualization of something in a disseminable media format, whether it be visual, text, audio, etc.. In a postmodern reading of something, then, nothing is considered “sacred” or free from reproach, and ideas are stripped down to their essence, lampooned and prodded, and then promptly bolted to the front of your robot.

Stance Stance Revolution is an instantiation of postmodernism in robot fighting: from a vertical spinner foundation, the discs are rotated to lie at angles to create an entirely new robot concept. It simultaneously derives from yet rejects the notions of the traditional vertical disc spinner, and is a complete eyesore while doing so. It raises questions at once of why the epistemological fuck would you do that and huh, I never would have thought of that. That’s kind of cool. I guess.

i am the department head of the school of postmodern robotics. don’t question me.

In short, it seemed to me like more robots than usual this year were doing away with, or severely reinterpreting “being competitive”. For instance, this….

….is quite the expression of postmodern robotics. Elements from a proven topology – a 2WD vertical “eggbeater” style drum – are seen as the foundation, but it destroys all notions of being outwardly serious and competitive by the fully functional LED-lit rubber duck mounted to it.

The duck is a liability. It’s a target for opponents. It prevents any form of operating upside down or self-righting. It might fall off and end the match right then… because all the electronics were stuffed inside it.

But it is the central statement of the bot, at the end of the day. This is truly the beauty of Postmodern Robotics.


From another branch of Postmodern Robotics, this robot is simply a knife with wheels. Why? Fuck you. It’s a knife with wheels. Are you really going to argue?


Roll Cake after the battles were all done. It gave some damage, and took some damage. But most importantly, it moved! And spun! And flipped things, mostly itself!

Full Disclosure: I actually slept through half of MicroBattles. ggwp. I showed up in time to watch a few more matches and participate in two rumbles at the end. Overall, I’m content with how the bot handled itself in the arena. One of the drive motor pinions ended up letting go and I was down one drive side for much of the first rumble, but used the gyroscopic forces of the drum to hobble around and make a few more hits.

One thing I ended up discovering was that the flipper arm wasn’t aggressive enough, due to the more conservative linkage travels I designed in. If your flywheel doesn’t slow down much per use, it’s oversized for the load power. I can probably make the linkage fan out more and trade some more efficiency points for a higher travel.

I ended up not really having time to recharge the battery for the second rumble, so it was running out of power around the middle. Overall, not a very competitive outing, but it showed me the drivetrain idea was sound now and the weapon drive system is fairly flexible in terms of actual layout. I’ll design up another version which is more Roll Cake 3.1 than 4.0 with some changes, but in the back of my mind is also a flipper-focused (non-exposed drum/flywheel) version, more like Magneato of NERC 30lb Sportsmans which Überclocker has fought a few times.

Now, we move onto Monday and the full-size Robot Battles. Remember when I said something about bad ideas just being bolted to your robot? Well some times it’s not even with threaded hardware….

That’s…. an interesting approach. Why so many ducks? Well, why not!? Better yet, they’re all squeak toys. Whenever this robot landed off the stage, it usually let out a protracted squeal sound from one OR MORE! of the ducks deflating.

It was otherwise a nondescript 4-drill-drive pusher bot. Postmodernism!

This is a robot which solely used a pool noodle for a weapon.

Appropriately named “Eyesore”, a newbie team (yay!) with a love for fluorescent paint.


nice bite force


Now we’re getting extra weird here. What do the ladybug balloons do on this robot? Nothing in particular. Yes, it fought every match with them!

There were not one, but two mildly-modified Roombas (and Roomba clones, as seen above) with things appended to them. One was in fact still “autonomous”, as autonomous as a Roomba can be, and was simply set loose on the stage.

This is a….

You know what? This isn’t even postmodernism any more. An entry with a FULLY FUNCTIONAL, ALL 3D-PRINTED tower crane on top of it? Now you could say we’re hitting on the territory of…


Get it? It’s a crane. Structures. Hhhhueuheue

This little saw-bot returns from previous years with an overpowering serving of new incongruently-themed stickers.

(Saws on an open stage?! I always take the time to explain that all freely-spinning appendages of robots are limited to 20ft per second tip speed in the Robot Battles rules, in case the message doesn’t carry through on first glance).

And finally, the robot I was helping work on, Skuld! Built by Leanne from Valkyrie, it’s a 30lber that has a very competently-powerd hammer with a 63mm-class outrunner. This thing could hit hard if it needed! It also had brushless drive with some highly geared inrunners.

That’s not a safety cover on the hammer arm end. That is the hammer arm end: One of several interchangeable plushies.



This is what the ideal robot fight looks like. You may not like it, but this is peak performance.

I don’t even know what the snout-on-a-piece-of-wood is from, but it paired with a powerful drive base is surprisingly effective.

Look! I entered a robot this year! It’s only 217 pounds overweight, no biggie.

On a last minute whim before the 12lb 30lb rumbles, we decided to heave Overhaul on stage to act as an arena hazard (it’s not running since I haven’t repaired the ESC damage from #season3 yet). It added an interesting play element: Suddenly, there was something to drive around and play hide-and-seek behind. People used Overhaul immediately to their advantage this way, skulking around looking for easy openings. Bots with giant wheels could escape by just driving up and over the forks. At one point, 3 or 4 30lbers teamed up to try and push Overhaul – they got a few inches in before the party got violent and broke out into fighting.

Probably one of the best rumbles I’ve ever witnessed come out of this event. It really makes me want to add a terrain element to the MassDestruction arena even more, in order to change up the small-bot game.

And this concludes your introductory lecture to Postmodern Robotics! I took a while to meander back north afterwards. Overall, I can say this Dragon Con was way more stress-free than any previous one. I was happy to see the competitive edge coming off of Robot Battles again, because a few years ago, when I and Jamison and many others were in (or around) colleges, we went through what I call the Tryhard Era of Robot Battles where the matches were becoming just as intense as any of the NERC parties and newbies were getting shut out or demolished. With the return of the TV shows, a lot more folks are cutting their teeth (metaphotical, drum, or otherwise) at these events, and I’ll happily step aside (or sleep through) them to let the interesting unjaded, sacrilegious designs fluorish.

On deck for the immediate arrival of fall is a lot of company-related pregaming before winter really sets in, so I’ll probably have limited content again for a little while. I’ll be slowly picking at 30haul, Overhaul, and the silly van nation in the mean time, but probably aren’t going to do any intense building until well after the new year.

Roll Cake 3: The Build!

Oct 09, 2018 in Bots, Roll Cake

It’s time to put a robot together! This post covers the entire fabrication of Roll Cake 3 in the week or so before Dragon Con. The goal of the bot was primarily to try and drive better with an indirect drive brushless setup and also further refine the packaging and action of the flipper linkage. Will it do anything?! Hell if I know, that’s the fun part to discover!

Even as I was finishing up the design the weeks prior, I was pregaming getting parts. Basically as soon as I determined the C2028 motor would fit in the triangle behind the wheels, I went ahead and ordered 3 more since I only had 1 to start with.

I also ordered these OMG THE MOST ADORABLE LITTLE 0.8 MODULE PINIONS EVER THAT COST WAY TOO MUCH MONEY from SDP-SI. I had to go solely by what they sold for 2mm bores in these gears. I wanted an 8-tooth gear to achieve the 8:1 reduction I wanted to get with the wheel gears, but they only sold the 8-tooth with a 3mm bore. So my drivetrain reduction is going to end up more like 10.6:1, which is fine by me – the bot will be too fast no matter what.

The problem with these tiny gears? The 2mm bores were slip-fit, not press-fit! Certainly irritating, but not the end of the world. Already noted for a revision in the future is maybe transitioning to R/C model pinions, such as R/C car transmissions. Those typically come in 3mm or 1/8″ bores.

I used Loctite 609 to stuff the pinions onto the motors in the end – I tried to silver solder one of them to little success – while I’ve managed to solder or braze pinions before, it seems that the steel they used in this motor shaft is an unusually high chromium grade which didn’t want to take with any flux I had on hand.

That was before I melted the pinion completely. Ah well.

Completed drive motors refitted with their shafts, which was a minor press fit job. I also replaced the stock stainless steel prop-saver screws with drilled-out and re-tapped #4-40 set screws. Another minor point of trouble with these motors is that due to their length and the small shaft diameter, it was hard to get these things to spin entirely true again after resintalling the shaft. A shorter motor (more pancakey) will resolve this.

I put the gearbox parts on print in Onyx as soon as I finished the bot design. The two ring gears and main cam linkage (sticking off to the left) have hoops of carbon fiber; everything else just derives strength from a bunch of perimeters. On the very left are the two connecting links between the cam link and the arm. I also made two first-pass prints of the drive wheels, seen here with O-rings.

Frame parts freshly done and needing cleaning. I’d say this is the only irritating part of doing unibody bots for me – picking the support out of difficult crevices. There’s certainly orientations that I can print in for minimal support, but they sacrifice immense strength in the part itself.  Both of these frame halves were printed in the orientation shown to give continuous loops in the XY (flat) plane.

In Roll Cakes past, I’ve actually taken to making fiber laps around the frame in this orientation, but with the more recent-ish changes to Eiger, you can change the perimeter and fill with more flexibility such that I just stack on the perimeters.

Otherwise, pull out some injection-molded part design – relatively constant wall thickness and smooth transitions if you can manage it, using ribs and cavities versus just solid massive areas, etc. However, my habit with 3D printed frames is leaving a few bulky areas up front such that they remain somewhat hollow – this acts as a very effective energy absorbing medium. For instance, basically the front inch or so if this bot is solid in the model, which means it’s perimeter plus non-solid infill in real life.


Drive motor installed with some test fitting done on the wheels here. Most of the hardware on Roll Cake 3 is specified as plastic or metal threadforming screws. If you get really fancy, people like using heat-set or tanged flanged inserts in MarkForged prints. However, I like my expediency, so some fat #10 tri-lobe threads will do to hold the wheels on.

Here’s the final drive fitup. The 0.8 Mod (32-ish DP) gears printed perfectly fine. In my experience, this is about as small as you can print reliably because the nozzle diameter still is small enough to go in and out of tooth tips profiles.


Both sides now fitted with drives. The front binding screw is shown sticking out here.

…and now joined together. Already, this thing is way more rigid than last time side to side. Hell, I might as well just ditch the flipper and make it a pure drum :v

The screw length needed to do this assembly were compromises with what I had on hand, so the bot dimensions shifted a little to accommodate them. Roll Cake is in fact not a perfect circle, but slightly squished inwards to yield the rough double-D-flat shape.

I slid in the trigger piece (which has a cutout for the servo arm) first, then I slid the servo into the DMs.

Alrighty, all the easy 3D printed stuff is assembled now. I still have to make a drum and motor, so off to Taki-chan we go!

(Also, check out the waterjet-cut feeder wedges made from leftover Hardox 450 from Overhaul!)

First order of business: separate the motor magnet ring so I can stuff it inside the drum. This can’t always happen cleanly, depending on the motor’s construction. The Donkeys have a single-piece stamped steel rotor, so I simply chucked the shaft in and very carefully parted at the outside corner where the magnet ring area joins the endcap.

With a motor that has an aluminum endcap, it’s often scarier since the parts can separate and self-destruct.  Luckily, this operation went very smoothly with higher speed and gentler feed to keep the forces on the rotor low.

Next up is to rough the drum shape. Most of this was easy, but I had to make the dual-disk shape, so I had to cut a valley in the middle. I just went with multiple staggered parting tool cuts to nearly the right shape, then very very carefully and gently made really shallow passes side to side to clean the profile up. Very gentle. These insert tools really should not be taking any side forces. To do this without wallhacking, I’d have gotten a single wider custom-ground HSS tool or something.


Then I bored the drum interior. There’s three stepped diameters – one for the ring bearing, then the magnet ring of the motor, then the “doesn’t matter” internal diameter. For the output side “bearing-stem-gear” I just did a straight drill to 12.0mm since it did not have to be that precise.

Next up: stator holder.

With the aluminum parts roughed out, I decided to make the one odd piece, which is the gear-on-a-stick that will carry the drum energy into the gearbox. This was a pretty easy turning job, but remember, it’s also the other bearing of the whole drum. The center bore, then, had to be 1. a clearance, but not by much, and 2. very clean and precise.

I actually went out and bought a 8mm-plus-0.001″ …. yes, this exists in America…. reamer to make this bore. You read that correctly – it’s a 0.3159″ reamer.

The gear gets a cylindrical boss on both ends eventually. After I finished both sides, I actually had to take a wire brush on an angle grinder to it in order to deburr the gear tooth edges – there were many little burrs and swarf hanging on which would have prevented the gear mate from working.

Parts in the midst of preparation! The stator bore has been drilled and the stem gear keyed on the top surface seen.

Throwing it together for a quick test fit! Here’s where it’s clearly visible that the stem gear skewers the cam ring gear and the cam linkage.

The next task was to drill the tooth holes in the aluminum drum body. I no longer have my old indexing head that made guest appearances in many of my undergrad builds, since I donated it to MITERS years ago. So to make sure I’m able to place the teeth 180 degrees apart, I milled a very shallow flat, then rotated it to sit on that flat, then milled another one.

This ensured I can drill and tap both sides with relatively equal precision. The teeth are just 3/8″-16 alloy cap screws, so they thread right in.

After the drum body was fully fnished, I broached it for a 3mm key – the stem gear is shown here mated to it. The magnet ring pressed into position without much trouble.

Unfortunately, I didn’t have time to make a new set of cluster gears, so I had to take them out of Roll Cake 2. Sad, since I wanted to keep that bot fully operational if I could!

The thought has crossed my mind of hiring out a whole basket of little cluster gears to be made if I’m going to keep evolving this bot.

The catch with these compound planetary cluster-fuck gears is that they have to be assembled at specific angle, or phasing, since there’s a hard relationship between the number of gears and the relative tooth counts. I scratched alignment marks into them when I first put them together for Roll Cake 1, and still use them for reference.

The gray goop I have dripped all over the assembly is liquid bike chain wax. I’ve been avoiding using petroleum-based lubricants like common greases because I don’t want to risk damaging the nylon-based Onyx material from the solvent action. Grease would also add immensely to the no-load drag of the gearbox, which is already a lot happening for a weapon drive.

The planet carrier installs easily after the gears are properly phased in place. It looks from here that there’s a ton of stuff going on, but I guarantee you this is not the case :v

As I mentioned in the design post, this carrier now has its own bearing to support itself. This is a pretty odd (but still apparently some kind of standard) 8mm bore, 14mm OD bearing – very low profile.

The drum drops into place from the front. The service order of this thing is a little suboptimal – to service the drum, I have to disconnect it from the folding linkage, but the central pivot bolt doesn’t rise up out of the body’s central cavity…. so really I have to undo the bottom linkage anchor (those three screws in a row from the CAD model) first, then remove the arm hinge. At least I made sure to make it all serviceable with at most 2 tools this time!

After the drum and arm are mounted, it was time for electrical dressup.

Along with the feeder claws, I cut another insectweight top plate set out of this piece of 1/32″ blue spring steel shim. I bought like 10 feet of this shim some time during undergrad, and have slowly been generating robots out of it since. This thing made its first appearance all the way back in 2009 with Pop Quiz’s rebuild! I’m now on the very last 8-10″ of it after Roll Cake’s parts are accounted for!

For better or worse, picking the parts out of the mother material was one of the last things I did before absconding on  3-day long slow meander down to Atlanta for Dragon Con. On the way, I hit up the Blue Ridge Parkway, like I promised myself I’d do. I only brought Roll Cake and its suite of service tools and a box of parts I think I’d need (mostly electronics), and Overhaul for display. This Dragon Con was going to be a pretty clean show instead of scrambling to take care of a few robots – the reason for this is I was pretty much solo this time, as there was a lot of bad timing involved for the various startups and schemes my friends and I are all involved in.

So what’s better than finishing your bot in the airBnB room when you got into town? Not all that much. I pre-gamed some of the electronics wiring (like battery leads and the like) before I left, so installing the two drive ESCs (AfroESC 12A) and weapon ESC (Spider ZTW 40A with SimonK) was a quick affair. I was able to drive Roll Cake around on the floor to get a gauge for its handling – which is much, MUCH improved over the hub motor drive version.

It’s very quick, but much more maneuverably so than Roll Cake 2, which tended to twitch everywhere. I swear the twitching is better than it was! At least now I could predictably send the bot places, which is an improvement over before!

The flipper geartrain was very tight, though, because I hadn’t had the chance to run it fully in up until that point. It would take several minutes of just running the drum at maximum speed and also a while of keeping the linkage engaged to wear everything in.

Here’s Roll Cake 3 fully buttoned up!

And shown with the flipper linkage at full height. In the next episode, hopefully you’ll get to hear about what it did, if anything!

The Roll Cake that Won’t Die: Roll Cake 3.0

Oct 03, 2018 in Bots, Roll Cake

I still like to pretend that I build robots here on this website! So in preparation for the fall round of events I like to go to (Dragon Con, Franklin, and any MassDestructions we try to hold) I decided in August to try and keep making progress on my recent persistent itch – Roll Cake.

Roll Cake had never “done well” – really done much of anything – at a competition since my main focus each and every time is kind of getting the vision of the bot finally in physical form. Remember its origin story and how I’ve been meaning to build a kinetic flipper for years, but never quite gotten around to it. It’s a robot built around a vision of a exterior shape and layout I came up with more than a decade ago, so it’s almost suboptimal on purpose.

Roll Cake 1 was sort of the grand puking of the idea in which nothing really worked. Roll Cake 2 made the mechanism shine, but still had a deficient drive due to sacrificing drivability for space conservation. My goal with Roll Cake 3 was to improve its driveability while making the weapon much more aggressive; Roll Cake 2 had a rather light drum/flywheel being belt-driven by an undersized motor that frequently overheated or shed belts.

The story of Roll Cake 3 actually dates back to not long after the previous Franklin Institute event, to which I brought it in order to talk about Alternative Flippers with a few other builders – it didn’t compete.

After mulling life on the return trip, I began the design by throwing some parts at the version 2 frame CAD and seeing what stuck. The principal design goals for V3 were:

  1. Moving to indirect drive on the wheels – the direct drive, while workable, was obviously still not very controllable. I figured a very small motor could tuck in the wasted volume (there’s a lot of wasted volume on Roll Cake) behind the wheels and could get me a considerable reduction just with open spur gears alone.
  2. Moving, on the contrary, to DIRECT drive on the weapon. The little 22mm outrunner just wasn’t enough to drive the whole geartrain continuously while also spinning a weapon. I could, with the increased drum interior volume, actually have a beetle-class weapon (so, you know, if the flipper plan just falls off a quarry cliff and explodes, it will at least be just a spinner)
  3. And lastly, moving to a dead shaft instead of V2′s live weapon shaft. This was more or less driven by going to a hub motor weapon. I’ll talk more about the Implications of this design change a little further down.

I had a Turnigy  C2028 motor model already, so I used it as a modeling guide for the positioning of the motor. I placed a Stance Stance Revolution motor, a Multistar pancake outrunner, in as a placeholder for the drum motor, though I wasn’t keen on using it. A motor that was a little more primitive seemed a better fit for the weapon.

To explore the space, I ordered some of the cheapest, shittiest motors you could buy on Hobbyking in the 35 to 45mm size range:

I love Hobbyking for having the sheer gall to sell you a motor that doesn’t have bearings. The Donkey line has bronze bushings and is seemingly made precisely to be the parts-recycling minimum viable products they are. A lot of builders have used them as foundations for their own weapon drives, and so will I!

I miss making motors, so this will be a fun distraction too.

I ended up selecting the Donkey 3511 motor for its stator size, but more importantly, the 12mm stator bore. The blue and silver “DT700″ motor had a thicker stator that I liked, but sadly only had an 8mm bore.

Why the bigger bore? I said earlier I was intending to move to a dead shaft for this iteration. Roll Cake historically has been rather tenuously held together side to side, with relatively small cross sections of material in the center due to the need to fit the flipper linkage. Moving to a dead shaft design allows me to use the shaft as an additional structural member of sorts up front.

The downside is I’d have to make a stator hub that the shaft presses into, and to do that, the stator needs to have a larger hole in the middle. I was intending on keeping the 8mm steel shaft that Roll Cakes have used for time immemorial (as in, literally the same one from v1 and v2), so that meant the stator needed at minimum, say, a 10mm center hole.

I pulled my usual motor designing tricks of making a hub for the stator to mount on, upon which a tpye 68xx ring bearing fits over. I sized this for a 6806 bearing, which has a 30mm bore. It’s a design balance between clearing shaft and wires versus simply being overly large and heavier.

I next generated a rough drum weapon shape that’s hollowed out in the middle. All of the dimensions and spacings were adjustable at this point – the final weapon would be a different width entierly. This just gets me something to start throwing into the CAD model so I can do the fitment of the gearset.

The teeth are, in typical beetle fashion, some big countersunk alloy steel cap screws.

With an eye for weight, I made some parametric adjustable cuts into the drum to turn it into more of a ‘dual disc’ configuration not unlike Witch Doctor and Hypershock. The final drum weight is tuned by just making the cuts smaller or larger.

Next up was the magical Roll Cake gearbox. I made some design changes to make it up to 1/4″ narrower to make the weapon itself occupy more of the front width. The clutch ring now has a single tooth and no longer its own outer support bearing – instead, it simply has a smooth shoulder inside to gently ride on the planet carrier. The carrier itself will have a 8mm thin-section bearing bored into it instead. There will be some extra friction from the technically bearing-less clutch ring, but the much more OP drum should more than make up for it.  I also got rid of an equivalent support bearing on the offset cam ring and it now only has a single bearing also in the center.

To pass power from the drum into the gearbox via the now dead shaft, I had to do something rather unconventional. There is now a lot of stuff going on here, so bear with me….

The distal endcap of the drum (opposite the motor end) is bored out to 12mm and has a key broached in it. A stem gear with a 8mm clearance bore has a keyway milled in one end and is inserted into the drum endcap with a key. The gear’s stem is 12mm OD and passes through the 12mm support bearing of the offset cam ring, and both transmits rotational torque and supports impact loads from the drum.

There is no “bearing” inside the stem gear in the conventional sense – there is only reamed steel on polished hardened steel with some oil in between.  Hey, if there’s one thing I learned from begrudgingly rebuilding an engine, it’s that steel on steel with a bit of oil in between is how every car works. What could go wrong!?

I knew going in I was betting a lot on this… technically a fluid bearing, but whatever…. working out. The friction would be higher than a ball bearing by far, but I was going to bank on the length of engagement making up for it with a light pressure resulting from the contact area.

Anyways, the gear end of the stem gear interfaces with the existing split-planetary gearbox and makes the thing go up and down.

One of the biggest challenges of Roll Cake has always been where to put the battery. When you scale robots down, you inevitably hit a “component Planck Length” of sorts – essentially, at some scale of robotting, the parts stop getting usably smaller. For me, the prismatic battery has always been troublesome for packaging inside a wedge-shaped robot. I played around with several methods, such as this one placing the battery forward and vertical…

…and including the unpalatable approach of splitting the battery up into two smaller ones. I really didn’t want to deal with the extra wiring and now squeezing on space for other components.

But one night I had a moment of come-to-Plastic-Jesus clarity – perhaps it is a reasonable compromise (it is – there ain’t no Perhaps, I’m just stubborn) to ditch the notion that the flipper has to span the whole width of the bot. At this point in the bot’s evolution, I should be thinking in terms of what makes the design work and what it actually needs, versus still trying to stick to my vision of “robot go flap-flap”.

In mulling over the compromise for the arm design, I also included a mockup of the new wheels, which will have a spur gear included on them to mate with a pinion on the motor.  I ended up just rolling with the C2028 motor and ordered a few from Hobbyking. Optimal? Maybe not. Fits back there and in stock? Hell yeah!

I decided to keep the battery arrangement shown – where the battery is placed widthwise in the bot, leaving the left half or so open to be used for the flipper linkage. This suddenly freed up a WHOLE LOT of volume inside the bot, and it was honestly a relief.

Before I went further with that design, I actually backed up and basically started over on another thought in my head about the design. There’s technically nothing stating Roll Cake had to be round. In fact, V1 was not round. Having the corners back could result in the difference between fitting in the battery vertically versus not, so I tried generating a square (chamfered octagon, I suppose) version of the frame to see if that was profitable.

Admittedly I did get pretty far along here – I found a vertical cavity for the battery and even was able to make space for mounting the drum motor. There wasn’t really anything preventing me from going with this design.

What swung me back the other direction was actually the sheer amount of usable space opened by repositioning the battery in the round design. I did want a Roll Cake in which trying to injection-mold ESCs and hyraulic press wires inside wasn’t even going to be an issue. From there, with an ideally working bot, I would make space optimizations as needed.

Well that does it for me. I returned to the round design and began cutting out cavities for everything. There’s gratuitous volume now to put things, and almost makes me wonder if Roll Cake could be a little smaller. However, for now, the final diameter was driven by giving the most space to the drive wheels and batteries while retaining an acceptable arm width that didn’t reduce it to just a stick.

Adding internal boss features to support the drum hub and drive motors now. I also made a crossing retainer for the battery – it sits in a neat little cavity and is prevented from bouncing around by the low wall and the eventual top plate. The geometry for the trigger pin is also taking shape.

Inside the left half of the bot, I made more space-filling features to mount the trigger servo. The dimensions did require cutting a hole of sorts in the underside to clear the servo cable. I moved the servo up as far as I thought was reasonable while still keeping the trigger pin on a radial path into the clutch ring gear.

Smaller but important features are now rolled up including the integrated wheel pegs and the arm pivot. The wheel pegs were going to be a machined piece, but gradually became so short there was no point in machining a part and using a mechanical attachment method. Instead, just printing the peg would do!

Notice how the bottom of the peg has a flat on it – to counteract the messy nature of homogenous-material support lattices in 3D printing which never really prints bottom-sides cleanly, I just made the peg a D-flat to give a single flat surface for the support to finish and the part to build. The eccentricity potential of the wheel from the bearing being on an incomplete circle is going to be negligible; worst case I’ll stuff a shim into the gap.

I put the arm pivot at the very front tip of the bot to make the arm as long as possible in order to get free height at the flipping end. This constraint would drive the arm shape and the placement of the linkages.

I made a first approximation of the arm after that. It had to start out higher than the bot’s upper slope surface in order to clear the cam and ring gears, which is why there’s a mild kink in it. You can see the cam ring passing through the arm here – I used this as a guide to make a cut that was just barely enough to clear it.

From there, it was relatively easy to make the linkage by fixing the cam linkage centered in the bot and fixing the arm in the lowered position. The lower anchor is a simple pin joint I modeled on the backside of the center crossing span of the bot, so all I really needed to do is adjust lengths and the position of the pin joint on the arm itself.

I tuned the linkage lengths to give me in the end around 4″ of rise. The end driving constraint was not hitting the crossing span in the middle when extended, but also when folded, not closing more than about 60 included degrees (or 30 per side measured from the centered cam link as a baseline).

At that point, which is my mental cutoff for “sensible linkage”, the orthogonal loading force from pulling the linkage open is twice the actual opening force, and just gets worse as you collapse the linkage more.

(This, kids, is why your elaborate scissor lifts never work in 2.007…. wait until you take 2.12 and understand the Jacobian matrix)


Here is that “JUST BARELY” cut I talked about – when the cam’s coming up, there’s only a few thousands of an inch of clearance as the arm goes up.

In real life, as I found out, the slop in the system actually made it such that the cam lifting shoved the arm upwards, helping the linkage get started out of the folded position. Sci………ence?

As a matter of habit when designing the smaller bots, I began adding hardware to make things more realistic for weight. On a bot that’s tight in space selectively like Roll Cake, I also wanted a sense of where fastener heads were going to go. So it turns out McMaster actually has intricately-cut models of their shitty sheet metal screws for download. I didn’t even know shitty sheet metal screws were 3D modeled, really. Either way, it was handy to see how much fastener head you actually need to clear.

The arm and linkages also get their dose of “Real-ification” by the addition of shoulder screws and properly sized counterbores, thru-holes, and tapped holes.

The linkage length allowed me to finally add one feature that Roll Cakes Past have never ever had: a rear structural link. There was enough leftover space to put a crossing span in the back with a bolt to tie it together! This might be the most structurally sound Roll Cake yet.

Space was tight enough that I needed to put the bolt model in and make sure I could still tighten a nut. The length of the righthand cavity was adjusted a little bit to be just barely longer than the bolt, so I can, you know, actually install it.

The three holes in a row above the rear crossing span? It’s a removable piece, serviced from the rear of the bot, so I can install and uninstall the arm linkage.

It’s a little hard to see what’s going on here. This is a counterbored, hidden cross screw which holds the front half of the bot together. It’s installable from the outside by squishing it past the O-rings in the wheels. You’d otherwise barely know it’s there!

This might be a better view of where it goes. It is sunk about an inch into the right hand (here, left side) half of the bot before the thread diameter begins, in order to put that much plastic in compression first.

I moved onto making some kibbles and bits for the bot – here are two little feeder fangs to try and get under someone and make sure the drum gets first punch. They’ll be waterjet-cut from some kind of leftover steel.

We’re getting awfully close with the CAD now. I took care of some more last details, such as the tiny half-round ears for if the bot gets flipped over to prevent the drum from hitting the ground. While Roll Cake can obviously drive upside down, the double-sided flipper was deprecated post v1, so it still has an “up” side. Maybe one day I will try bringing the double-sided flipper back.

I then made wire passthroughs and filleted corners and broke edges. This is the now-complete “unibody” frame.

I took snapshots of the side cavities and traced them in order to generate conforming top plates. The hole locations were patterned where I could and kind of freelanced where I couldn’t, at a fixed distance from the nearest profile curve.

To actually make the bot, the monolithic model had to split into three pieces – the two halves, plus a drum axle retainer for the non-motor side. I generated cutting surfaces for this operation. The split is not linear, but has a jog in the middle to accommodate convenient locations to begin and end the frame profile.

The cover plates are modeled from the sketch drawings individually. I suppressed the monolithic body and  imported all of the cut pieces and properly constrained them in order to yield the multi-piece modeled frame.

So that does it for Roll Cake 3′s design. In the next episode, I have to actually build the damn thing!

ORIGINAL CONTENT! The All-Around Robot Update: Roll Cake Rises Again; Dragon Con 2017 & Uberclocker

Oct 02, 2017 in Bots, Dragon Con, Events, Roll Cake

Whoa, what happened to this place? Everything’s so dusty and gross. Why is there a pile of circuit boards on the bench??

*trips over Chibikart on the ground and dies*

Hey, remember: whenever I disappear for an unexplained period of time, it’s always because I’m working on something hilarious. This time it’s extra hilarious, I promise! Obviously I’m always itching to keep everything updated here on my latest, but just like the first BattleBots build season, externalities which if broken would make other people look like assholes prevent me from saying anything at the moment. See, I don’t mind me alone looking like an asshole…

Anyways, backing up a little in life, I decided to redesign Roll Cake from the ground up following my hub drive experiments earlier. MomoCon came and went, but the Hobbyking orders kept stacking up, so I decided to roll it all in with Überclocker’s changes for Dragon Con!

roll cake

It all begins with a wheel.

Doing the drive test with the SimonK ESCs and the Multistar 460kv motors convinced me that the hub motor direct drive would work out, at least better than the previous BS I tried to do. I went shopping for high pole-count, low Kv drone motors since they’re pancakey. The plan was just to approximate wrapping an O-ring as close as I could to the motor. I ordered a few of these AX4006 motors for their combination of weight, low Kv, and high pole count.

Roll Cake is a bot which faces some packaging difficulties, since the middle of the bot has to be left pretty open for the flipper linkage. It would actually be easier in a 12 or 30lb design, since ‘noise floor’ of part sizes is much smaller compared to the bot size. If I scaled this design up to a 30lber right now, those would basically be 6″ hub motors, which is unnecessarily large.

There’s other architectures and shapes for the bot which might alleviate this, but for the time being I decided to try and keep the cheese wedge shape but  make it a little more…

…round. Remember that the flat sided shape was just an attempt at vomiting my vision of a bot that I’ve had for a while now, not making sure it works. When you ditch the need for 6WD, things get a little simpler! Even this is technically unoptimal packaging since there will be a lot of wasted space in the narrower parts of the cheese wedge. I’m basically just reskinning Roll Cake v1 and using all the same parts, since the goal is to get it driving and flipping things reliably, albeit not spectacularly, before deciding what aspect of the design to improve.

Once I had the parts placed reasonably, I started generating frame features to accommodate, such as wheel cutouts and future bearing blocks. The chassis will no longer split in the middle – that required so much extra effort to get everything to line up. Instead, I’ll be splitting the rounded caps next to the bearings off as its own print in the future.

The previous image showed the old linear slider trigger, but packaging necessitated switching to a swinging style. This means Roll Cake won’t fire when upside-down with the drum running in reverse – I’d still have to ‘self right’ so to speak. That’s fine, since I’m also ditching the double-sided linkage due to it taking up the entire center of the bot from swing space. At least keeping the flipper single-sided lets it still have structure in the middle!

The chassis is now taking shape pretty well, showing the swing trigger’s backing and “drilled” bearing cap holes and the like. I’m designing this to print ‘upside down’ on the flat top face.

After defining critical part anchor locations, I hollowed things out to accommodate the flipper linkage and irritatingly rectillinear things like batteries. Seriously, if there’s one thing this design is sorely lacking, it’s a battery worth having. I much prefer this to be 4S, but can only fit a 3S pack of adequate capacity for now.

As I model the body, I can give components final homes constrained by mounting holes and then adjust the cutouts and spacings to fit. So there was a fair amount of tuning going on at this point, including a change of wheel size to be smaller in order to shift the wheels more rearward (to give me battery space!)

After that, the fun part became linkage design. The goal is to get a linkage design which travels as far up as I can manage using the roughly 1″ throw of the cam ring, and generally has no linkage interacting at more than 45 degrees starting angle.

My insistence on a “pull” action on the main cam linkage means I have to transform the motion through a bell crank (the bottom and right side short line) to become an upward motion. Strictly speaking, I could potentially accept a push action from the cam linkage and that can directly interface with the flipper arm and move it upwards, but it would need to be designed much more heavy to stand the compressive force instead of tension (pulling) force.


This bell crank itself went through a few revisions in order to minimize the impact it has on the middle of the bot, the large bulkhead that runs across the two sides.

Here, I’m comparing designed linkage travel with actual part placement, seeing how much of the middle of the bot has to be cut out. The bell crank center distances and topology have also changed. The previous design intruded on the center of the bot with its full height, whereas this “T” design means only the short leg of the T pokes through the center bulkhead.

Then I decided to wrap the bulkhead around the bell crank instead of hollowing it out pre-emptively. It’s all going to be 3D printed in 50% density anyway, so no need to pre-emptively deny myself cross sectional area (which is very important to 3D printed parts)

After I was satisfied with the bell crank geometry, I made a crude flipper arm model to start out with.

The linkages will have to fold into themselves a fair amount, so I pre-emptively carved space for them before doing anything else.

The intermediate linkage is a bit of an awkward shape – here it is taking form. It has to adapt the narrow bell crank to the wide flipper linkage. I decided to do it here, and reinforce the middle of this linkage with a big flange, instead of trying to flare the end of the bell crank wider due to my desire to print it flat and have fully un-interrupted perimeters to maximize strength.

Here, see the aforementioned flange in the center of the intermediate linkage. I’ve now hollowed the flipper arm, which will be top-skinned with hardened spring steel.

The armor for this bot is quite simple – primarily Onyx in massive hollow-ish sections for the crumple zone effect, and blue-temper spring steel covering the important parts and providing access hatches.

I added a little feeder leg next to a region with unused material thickness. This will be a machined piece which is captured with nuts and flat-headed screws.

Finished and ready for printing!

I had to split the geometry first into the printable sections. I extracted the bearing cap by making a 5-sided surface box in Inventor and using a split body by surface function. Only one was needed – the other was disposed of. Other sections such as the mostly empty tail were cut off also in order to reach the print volume, and they were designed to be bolted back on.

A day later… Her’s the frame finished, printed in 3-perimeter 50% density Onyx. Ought to be plenty!  You can see where I cut the end of the wedge off and have modeled in a few tappable holes to hold them on.

Here’s a pretend-o-bot to make sure the dimensions all fit. The bearing cap was something I was particularly nervous about. I didn’t design clearances into the linkage parts to save design effort (read: I’m too lazy to make proper constraints) so some filing was needed to get them sliding freely.

Hardware installation time! I made sure to make little access ports for the motor wires, because wouldn’t that be embarrassing?

We move now to my old high school workbench down in Atlanta, which is somehow still there and in operation (maybe being 16 feet long has something to do with it). I got all the mechanical hardware installed before leaving, and decided to save the wiring for the Dirty South.

Pictured in the foreground is my new best friend: the itty-bitty-baby-offset-screwdriver-bit-ratchet. It’s McMaster part number 52725A31, and it’s positively adorable AND the only way some screws on Roll Cake are accessible at all. I designed it this way, so it’s legit, right!?


As usual with this thing, wiring is a disaster. The ESCs of choice are the Afro 30 Race with SimonK set up to do reversing with my usual tricks. They’re small, but not THAT small. I decided to keep the ESCs on the same side of the bot as their motors in order to reduce the amount of long wiring runs, so there’s two on the right side of the bot and one on the left.

All the motor connections have now been made, and I left one task for last before I soldered the 3-pin signal wires to the receiver….

I had to program the SimonK firmware to activate reversing and braking and my preferred goodies. I planned ahead and made this servo to tiny-clippy-jiggle breakout cable which I’ve termed “The Simonator 2.0″ in order to grip the signal wires of the ESCs. While I could have programmed them all beforehand when the servo conectors were all still there, I decided I needed this cable regardless just in case I had to change something in the field, post-installation.

I brought the finished bot to one of the robot panels at Dragon Con. Sadly this year I fell off the bus and did not host any panels, but I’ll make sure that changes next time! I’m glad that recently, my Makers presentation hasn’t really been needed – in the most recent years I delivered it, the percentage of the audience who’ve experienced CAD or soldering LEDs together, etc. has grown immensely, in my opinion greater than the rate of self-selection for these things.

Here’s the linkage fully opened! Note the preponderence of little shoulder screws forming the joint pins – I standardized all of these to the same length to save myself from my historic habit of making my robots all shoulder screw nightmares.

….and now announcing my new 6lb multibot entry??? This is the head of Lucy‘s Mei cosplay, the freeze-ray dispensing Snowball. overwatch has ruined my life run away now

I’ll post some of the test videos of Roll Cake soon – I was happy enough with its performance in the garage in terms of drivability and flipping, even if it won’t prove that impressive in the box due to being repackaged test rig parts.


We now move onto good ol’ Clocker, which has looked like this since Motorama…

Pretty depressing, eh? In the final rumble of the 30lbers, I burned out one of the SK3 4240 drive motors, so I was on the hunt for replacements, and Hobbyking didn’t have stock in that size at the time.

What they did have is a sale on their new NTM line, which had a similar size motor:

So I scored a couple of these – they were physically the same dimension, but unfortunately these motors were slightly faster again, so I was facing the very real prospect of Clocker hitting 25mph without much provocation, which could be a liability on the Dragon Con stage.

I emptied the bag of Clocker remnants to see what I could salvage and what I’d have to remachine – the answer was really basically everything minus the motor output gear :p

Good thing Clocker is legal in the new 30lb Sportsman’s Class rules enacted FRESH AND NEW this year for the Franklin Institute event in 3 weeks!

To extricate the motors, I had to disassemble the frame, which proved a little…. challenging after Glasgow Kiss gave it a once-over. There were some special extraction techniques I had to use here on this machined corner!

From the spare Clocker parts bin I extracted another section of the 1/2″-10 leadscrew and flanged bronze nut that fit it. I’d bought a few spares last year in anticipation of needing to machine them eventually, and here we are.

The bronze nut gets machined all the way down to be smashed into the bore of the modified Vex Pro spur gear. When the gear spins, the leadscrew gets sucked in and out of the nut, and its own reaction forces are taken up by the bronze bushings surrounding it. All solid, all friction, all the time, but it gets the job done.

I’d like to eventually rebuild Overhaul’s actuator in this way, except with preloaded tapered roller bearings, for #season3 whenever it ends up being :’(


Mate this up with new waterjetted plates that I drilled and tapped and we have a new actuator. The drill gearbox was reassembled from stock pieces from my giant decade-old (…) bag of Chinese cordless drill parts, using the original shaft which was not damaged in the fight. I have enough pieces now to straight up make two whole actuators, which is nice.

After that, I repaired the bottom plate of the bot by stitching new holes in between the hole ones. I’m not sure if I’d use #4 screws like this anywhere in a loadbearing path (which the top and bottom plates do count as) if I redesigned Clocker again, since the indirect shock loads from the 30lb Featherweight class alone (in the form of getting socked by a spinning weapon) is much higher than Sportsmans. One of the corner hits from Glasgow Kiss sheared off a half dozen of my bottom plate screws just by momentarily bowing out the frame enough.


While I was in there, I swapped in the spare wheels made from 60A Mystery McMaster Urethane (actually OEM’d by Forsch Polymer, the most 1997 company extant in 2017). The white Smooth-on Simpact wheels had worked well enough, so I wanted to see how these would do.

Well, everything is technically ready for reassembly!

I rememberd a much better way of taking the entire top off Clocker. Previously, it involved trying to drive the center lift shaft out through ALL of the components that were shaft collar’d onto it. This was patently painful. Unlike Overhaul’s unboltable lift towers, Clocker has solid ones built into the frame rails. It turns out if I just unbolt the outer and inner frame rail on one side as a unit (9 screws), there’s enough room to wiggle the shaft out of the bearings and pop the whole thing off.


Clocker was the last thing I wanted to put the Brushless Rage test units in before shipping them off for production. The severely under-geared high-Kv motors will be a good stress test for the architecture, since on the Dragon Con stage I’ll mostly be driving at low speeds and turning/reversing often.

Check out that little Onyx bracket I made to hold the units. I wanted to place them flat against the frame rail behind them here, but this arrangement kept the wiring cleaner and away from the outrunner motors.

A new waterjet-cut gear and some quality Taki-time later, and everything is now back together. I did some drive testing outside, which showed me that the Brushless Rages were working great even under duress – the gearing on the motors is low enough that the bot has trouble turning in place on a high traction floor. So here I was hoping that it would be even able to turn at all on the Dragon Con stage carpet! But once it takes off… boy does it want to keep going.

the charles and the dragon con

Welcome to Dragon Con! Have a van.

I would have loved to bring VANTRUCK instead this year, as it has been now impeccably reliable after its lobotomy and subsequent headcrab installation, but could not even begin to justify the 9 miles per gallon each way. It’s beginning to dawn on me that the kind of person who would have bought one of these things new, never ever thought about the cost of fueling and ownership. I’m not quite to that level in life yet.

Overall, this con worked out a lot differently than some of my past Dragon Cons. See, I wasn’t scrambling to finish a robot every day for once – Roll Cake’s finishing work and testing occurred before the con started. Instead of trucking around a giant transforming mechanical prop, Cynthia instead prepared a bunch of pieces for the Dragon Con art show (which as I found out was nontrivial to get into)

On top of that, it’s become more of a yearly reunion for some of the BattleBots competitors and friends who have moved around the country & world. For example, I found Lisa Winter!

The cotton candy committee has arrived.

I attempted to replicate her tattoos in the middle of talking at a panel. Nailed it!?!

And for the first time in probably over 10 years, I actually played in a gaming tournament. There was an Overwatch ruined my life run away now tournament being hosted at the convention gaming center, and a few of us essentially set up #BattleBotsPlaysOverwatch.

The house equipment was sub-par, though, so we didnt’ do too well – people who have clearly been to more than one tournament brought their own mice, keyboards, headsets, and pillows and stuff. Now that’s pro.



Alright, you know how Dragon Con goes down. Let me spare you the details and get to some robots!

MicroBattles has grown to the point where it has to be single eliminations only and running across two arenas to keep up. I’m glad that it’s a good problem to have! However, it does mean you’re pretty much one-and-done.

There wasn’t much to do with Roll Cake beforehand except get some driving in. I decided to move the tail on the flipper downwards one mounting bolt such that it was more likely to rest on the floor – otherwise, the bot tipped abouts its wheels a little. However, it kept weight on the feeder wedge, so that was a plus.

Robot Battles features mostly local builders who kind of keep to that series of events around the Southeast. It’s refreshing to see bots which haven’t been forced to become the small monolithic dense bricks that most competitions have forced them into being. These two, for instance, are hand-bent sheet metal from Home Depot, with a hand-soldered custom motor driver inside. I honestly miss these kinds of builds.

Pool noodle wheels were fully in fashion this year, made popular by the Dale robot Noodles. Hey, they’re totally not entanglement devices. The wheels aren’t supposed to come off, just incidentally if you hit them with your spinning thing! Wink wink. I suspect this kind of thing might get roundabout-banned somehow, but on the other hand, it’s 2017 – get a reversible ESC on your weapon already!

Sheet metal everything, down to the weapon! Now this is robot fighting.

Other builders who had too much time on their hands chose to adorn their robots in….. creative ways. That’s hand trimmed and applied fake wood veneer vinyl on Margin of Safety here…

I was pretty eager to fight Noodles since it’s high ground clearance and invertibility would have made for a whole match of flips with Roll Cake.

Besides the wacky builds, you had your usual array of kit-bots and modified kit-bots.

Roll Cake was matched with Margin of Safety first, obviously a fight that I was hard pressed to win. Aaron put on the miniature vertical drum module for the match, so we went head to head trading blows. Margins having the the smaller drum advantage,  Roll Cake got flipped over and I spent a while trying to self-right, but at the time didn’t have any skids on top of the bot, so I trundled it around a few times trying to get him to flip me back over.

See the two little hex nuts sticking up from the top? That was added after this match so I could get flipped over in the rumble and maybe get back up. With the drum bouncing off the ground, it wasn’t going to get enough momentum to roll it self back over, so after a while of trying, I decided to save the effort for the rumble.

In said rumble, the drum promptly threw the rubber o-ring belt and jammed as soon as it started. Well bugger me with a #1/2-20 tap, that sure didn’t come up in testing! So I spent the whole time running around like an idiot.

I suspect that spinning up quickly made the belt stretch enough (since rubber cord doesn’t have a tension element in the middle like fibers) to jump out out of the pulley enough to get grabbed by the drum. In Roll Cake 1, the pulley spacing was far enough apart that it would have just fallen off, but this time I had to move the drum closer and so there is a lot of overlap with the drum iself.

Hey, all things considered, I walked out with a working bot. It’s now time to get serious with Roll Cake. I’m extremely confident in the mechanism now, and so it’s time for it to stop being a test jig on wheels. The weapon motor is severely undersized – if there was one design which should have a motor-in-drum setup, it’s this one! And, furthermore, freeing up the space occupied by the weapon motor might mean I could use more conventional drive motors. The hub drive worked well enough, but I still prefer the positive feel of a geared motor.

And now it’s Monday!


With the return of all the robotty TV shows, we’ve seen a serious and sustained rise in the audience count. The room filled up to this level well before matches began, and the line continued out the door the entire day.

An entry being finished in the pit area before matches begin! How quaint.

Lisa brought itty-bitty Tento, weighing about 8 pounds, and entered it in the 12lb class for fun. This thing was built as a “how to build a robot” demo piece. Unfortunately, it suffered a gearbox failure literally right before matches began and it was of a type that nobody else was using, so spares couldn’t be located. Sad day – maybe next time!

Clocker just needed battery charging (and the replacement of a chain tensioner block) this whole event, so I’m quite pleased.

I only ended up having two matches – one against this giant purple thing (which had radio problems at the end – notice us both running onstage to disarm it), and the second against Dale’s Pushy Grabber. This thing has been sweeping RB events (literally) with the lynchpin strategy of wiggling under your bot almost no matter what. Now, normally I offer at least some resistance to Dale, but this thing I had to approach either at a very specific angle or risk getting plowed off the stage almost instantaneously. We had 4 total matchups, in the middle of which Dale had to reattach one of the rollers and I had to replace a chain tensioner block which finally decided to wear through and fall off.

This event really showed that, much like my arena-optimzed Test Bot v4 days in the Late Aughties, wide ground-hugging wedge surfaces really are more of a liability on the stage than an asset. Notice how in the final Pushy Grabber matchup, I tried executing the same strategy, but got hung up on the edge just long enough to become vulnetable. The only weakness of Pushy Grabber right now is a long-reach forked robot like Nyx with the lifter attachment – Clocker did not have enough “stickout” to really get a good handle on it – nor did it really on other bots.

Unlike version 2 and 3 where the clamp arm reached all the way to the end of the forks, this one for the sake of looking more like Overhaul has the ‘grab point’ more inwards, so I had a harder time getting opponents into the clamp in the first place unless I took a straight run at them with some velocity – upon which I would often run into the stage edges.

I stuck around for Rumble #1 which I won by virtue of trying to get around the damn stage and mostly ignoring opponents…. and in Rumble #2, I just took the wedges off and ran around like an idiot some more, accidentally handing the win to the purple thing after doing some kind of J-turn rocket jump off the stage. Oops.

This event was also the final straw for me in terms of gearing down the drive motors more. I’ve been threatening to go to 2-stage gearboxes for the drive, and now it’s more necessary than ever. Clocker v3 was geared for 19mph and was already a rocket, and there was barely any need for it on the stage. I’ll probably move to 11:1 2-stage P60s and use smaller 35mm drive motors.

Yes, this kind of thing is legal here, with a catch: It doesn’t exceed either 500 RPM, or 20 ft/s edge speed.  It’s driven by a geared motor, so it will more lift your bot up and chew at it.


Replicas of BattleBots entries are the in thing right now! This is Tuskin Raider, a 12lb Razorback-alike that Jamison built. It got all the way to the 12lb finals.

This is a 12lb shell spinner.

And here we have the assembled Power Rangers shot of all the scale models. Hey, we can film #season3 right now if we just get all the cameras up really close. I keep bugging Jamison about why he didn’t make Tuskin a 30lber instead of a 12lber.

So that’s it for Dragon Con! Two working robots remaining, shenigans abound, and…. no van adventures. Wow, when did my life become routine? Obviously it’s time for another all-vans update soon….

Brushless Hipsterism Intensifies: Returning to Brushless Rage. Brushless Mini-Rage!? And Trying Hub Motor Drive in a Beetleweight

May 12, 2017 in Motor Controllers, Reference Posts, Roll Cake

Oh, Brushless Rage… how far you’ve fallen. It’s been standing idle since late last year when I got the first version running. Thereafter, it began having some rather obdurate power supply problems that I couldn’t resolve with a few different attempts, and with #season3 still unknown (TO. THIS. DAY. UUUUUUGGGGGGGGGGGH.) and having to pick up and move shops, I lost motivation. Now, with the spring and summer silly go-kart season coming up, me really wanting to pregame getting Overhaul back in shape ( *cries deeply* ), and my comrades over at Robot Wars screaming for assistance, it’s time to put my robes and wizard hat again.

The last time I really worked on Brushless Rage was in October. After tuning out the first one, I went ahead and made a 2nd one. I wanted to get Sadbot running on them for a few test drives.

Here’s my innovative housing for the two controller! Bolted back-to-back with drilled holes in the Ragebridge shipping box.

And that was all! It was retained by a few zip ties running through the bottom ‘breadboard’ baseplate. I didn’t take much test video of Sadbot running on them, unfortunately;really the only one that exists within easy reach is, uhh, this one. While it doesn’t show them getting whipped, they definitely don’t not work! Yay!

But not for long. I soon lost both of the units in further off-bot tuning of settings. They didn’t blow up, but simply failed to ever power on, with the LM5017 regulator simply sitting there getting hot. The only “fix” was replacing the regulator, and I say “fix” because that really didn’t fix anything, and they would die again within minutes or even seconds.

No problem… maybe it’s just an issue with the two boards. I’ll just try another one of the five total I ended up making….

Nope. Nothing. They died one by one, all to the same symptom. I tried redoing my math for the regulator for the 4th time, thinking maybe  I made a mistake somewhere. I even tried mimicking the reference design to try and get something running. I literally never do that.

At this point, I figured it must have been something incredibly dumb and simple I missed. But why would the first two have worked at all, even for a little while?! Convinced the solution might just suddenly invent itself, I stopped thinking about it.

And so here we are, a few weeks ago, when I’m slowly building up a new rev of the logic board that fixes up some trace routing problems and Little Blue Wire problems. Again, the logic regulators kept exploding, some times dramatically taking out the input trace like seen above. The little light is strapped across the 15V gate drive supply to give me a visual indication of it being on.

What is with me and being unable to use switching regulators!? I recalled the Ragebridge Diode Debacle of 2015, and decided to take one last Hail Mary run through the datasheet along with friends to carefully cross-check each other for boneheaded mistakes and…….

TI, you assholes.

So here’s what’s going on. The Vcc pin of this chip allows you to power it from its own output voltage, which is often fairly low, so it prevents a lot of heat dissipation in the chip since otherwise it would have to derive its own power from the voltage input (up to 95V). But what I missed is this only works up to 13 volts. My gate drive supplies were 15 volts by design.

Beyond that? Who knows?! It might work, it might not. I’m guessing my first two were just high enough in manufacturing overhead that they worked for a little while. Subsequent statistics were not on my side.

Okay, whatever. I cut off the 11.3kohm feedback resistor and threw on a 9.1kohm to drop the voltage from 15V to about 12.5V and let’s see what happens.

Ah, it wakes right up.

Of course it does.

So I decided to respec the gate drive for 12.5V. Why do this instead of go for the full 15+ volts? Because I’m really aiming to make this design work at high-for-robots voltages of 36-48v, possibly up to 60V nominal with a different power stage, so I’d like to save the power dissipation in the chip’s onboard logic power supply.

The change in drive voltage will slightly affect the drive characteristics and switching time. For now, I’ll keep all the power stage parts unchanged, but I’ll probably tune the gate resistor values later.


To get rid of the noisy ripples on the feedback network and to stabilize the switching frequency, I added some more bypass capacitance to the chip. This was not included in the design at first, since I figured my large ceramic input and output caps were nearby, but it really really wants its own little private capacitor on Vcc. Gee, I thought I was a princess at times.

So now this thing is pretty much bombproof. Here’s a video of it throwing around one of the 30-pound old MIT CityCar prototype motors (which I inherited 4 of after the project was dismantled):

In that video, it’s running from 36 volts. I tested it with a smaller motor all the way up to 50V input before getting too scared for my power supply’s life; I’ll need to try it on a larger high-for-robots voltage power system later, but nothing smelled imminently unhappy!

With the regulator death issue apparently behind me (again) I decided to push another board revision. This time, I added all the necessary bypass caps and changed the layout of the logic power supply, as well as take out some parts I decided were superfluous.

The logic power supply got a little smaller and more electrically optimal. The whole thing is just less messy now. I like it – it takes up around 1/3rd square inch of PCB space on one side. At the behest of a professional PCB engineer friend, I turned the inductor 90 degrees and joined it with the LM5017′s switching node with a small trace instead of a larger groundplane. This would prevent the switching node (a source of huge voltage swings in microsecond timescales) from broadcasting as much noise.

Besides some other minor trace chasing, what’s going on down below on the board is also something experimental:

That there is a bidirectional optoisoated I2C bus for transmitting data between two microcontrollers which should never meet directly. I had a single-direction opto input on the board revisions so far, but this prevents updating of settings via the SimonK/BLHeli type bootloaders. That means tuning the settings require busting out my chip socket every time, which is annoying. I reviewed a couple of bidirectionally isolated bus schematics and decided to try this one out first, since it involved diodes only, not transistors.

The problem is, the I2C bus is a open-drain configuration with pullup resistors and ’1′ bits transmitted by pulling the line down to 0v. I kind of wanted to try keeping the opposite polarity, so to speak (even though SimonK supports an inverted input setting) just because I’m used to thinking about things this way. So I tried flipping the circuit over…. pullup resistors became pulldowns, and common-emitter became common-collector, and so on.

It makes sense in my head, but I’m sure excited to see this work!

On the board, this is the layout. It doesn’t consume much more space than my previous 1-direction optocoupler setup, and can be bypassed for testing with 2 wires if needed. That’s the nice thing about keeping things upright signal-wise.

So before I sent this board revision out, I stopped for a moment to think who would really be wanting to use Brushless Rage. I’d designed the 12-FET board to effectively replace Overhaul’s 250A DLUX controllers (with more realistic ratings, mind you). I’d say the majority of people who would buy such a thing won’t be running motors that big.

Recently, the thought of a “Half-Rage” has been coming up in my mind as something worth pursuing. This would be a board with about half the footprint of a RageBridge 2 and supporting about 1/2 of the amperage. As some curious question-askers had innocently drilled into my mind, this would be an Actually More 30lber-Sized controller.

> mfw "When are you going to make a 30lber/12lber version of RageBridge?


With this in mind, I decided to make a copy of the power stage and began downsizing the hell out of it.

Step 1: Reap what I sow when it comes to the sheer number of vias I deposited under the FETs.

After bunching the FETs together, I referenced one of the earlier abandoned Brushless Rage layout ideas for the output wires. This board is now short enough that I’m comfortable pulling the phase outputs all the way to the right with the power. Keeping all my wires on one side is something I prefer.

Somewhat final routing of the fat bus traces here. I had to move a few gate drive traces, as there was no longer an opportunity to swap sides in the middle of the FET bank. Power+ runs straight from the bottom right corner, through the bus capacitors, into the high-side FET. Power- emerges from the current shunts and then has 3 paths to return to the buscaps before being slurped up by by the wire hole on the upper right.

Here’s an overlay of the signal board design on the power stage, showing roughly the size of things. The final power stage is 2″ x 2.75″. Not the tiniest thing, but I have more capacitors than you!

This board shares a lot of thermal characteristics with RageBridge, so I’m pretty comfortable calling this a 50A continuous class controller. 50 real under-partial-throttle amps, so that’s what, like 1,200 Hobbyking Amps?

In all likelihood, this controller will be able to handle an average 63mm SK3 motor in continuous duty applications like a silly go-kart. Robot-wise, it will probably be stressed handling the same in bidirectional drive mode.

Fast forward a few days and….

OhmygoditssocuteIjustwanttohugit and then make it run a 80mm outrunner on 12S violently. I’ve ordered parts to make a handful of these, and two are going on Sadbot ASAP to be driven until something blows up!

Direct Outrunner Hub Drive for Your Little Bot

Next up, something even smaller!

So I’ve long been a connoisseur of fine handcrafted hub motors. I got curious recently on using direct-drive small outrunner motors in an ant or beetle after thinking a while on the redesign of Roll Cake. Version 1 of Roll Cake was honestly just a braindump of a vision I’ve had for years for the shape of the bot, and everythng else came second to that. On the beetle scale, the multi-pulley serpentine pulley drivetrain simply had too much friction for the Fingertech motors (which were severely underpowered for the task) to overcome.

For the next version, I’m ditching the triangular cheese wedge shape for something more straightfoward. The cheese wedge will be back for a heavier weight class. Roll Cake’s design really wants to have the middle of the bot kept clear for the flipper linkage. I’m sure I could work around it with low-mounted drive motors and similar, but this was an excuse to play with brushless things!

I based my thoughts off Jamison’s mini-gimbalbot which used camera gimbal motors for drive with a small Hobbyking R/C car ESC. It drove “okay”, certainly capable of a weapon delivery platform. So naturally, I wanted to put some SimonK-capable controllers on it and see how the handling would change. I got a small selection of motors: A pair of DYS and Quanum 28mm motors as well as a pair of Multistar “HV” 460kv motors. 460 RPM/V is reeeeeally slow for that size of motor that isn’t a gimbal motor, so I was quite interested in them.

These are the gimbal motors. I like them for their pancakeyness – the Quanum motor is more 30mm and has a bigger stator.

Playing around in the CAD model a little for component placement. At this point was when I realized Roll Cake in this incarnation might end up looking a lot like The Dentist :P

I designed up a few hubs that bolt to the face of the motors and have a tapped middle hole to sandwich a wheel. The wheels are spare 1.625″ BaneBots wheels that I originally bought for Candy Paint & Gold Teeth.

Shown with those motors is a ZTW Spider 18A controller. My typical SimonK ESCs, the Afro series, were out of stock when I placed this order, so I took recommendations from people on what I should use. The Spider series are fairly popular these days among small bot folks.

The issue is, they come with BLHeli firmware, the other other open source drone racing / vaping rig development path. It’s a newer effor than SimonK and has a more polished interface. I’d read about it before, but not worked with personally. Other builders have said it doesn’t run robot drivetrains as well due to being much more optimized for propellers. So hell, why not – this was a chance to explore that side of things.

Here’s some real life CAD layout, featuring the Multistar motors.

I really wanted to use the gimbal motors, but they disappointed me in bench testing sufficiently that I didn’t even end up installing them. Basically, they can’t draw enough current to make torque at typicall little-bot voltages. With phase resistances of 10-20 ohms, they can really only draw ~ 1amp or so. I mounted one in a vise and could stop the motor with my pinky finger at full radio stick input.

These motors might be better at 6S and up, but for the time being, since all of my small-bot batteries are 3S, I decided to pursue making a test platform using the Multistar 460kv motors.


The platform of choice was…… one of Candy Paint’s spare weapon pulleys. I literally spilled my “preformed robot spares” bin on the ground and tried to see what was good to use as a base. Hey, it’s round and has convenient wheel holes in it already! All I needed to do was quickly whip up some motor mounts (3D printed) and I was in business.


Here’s everything hooked up. That nut is for a counterweight on the front to add some friction against the ground while turning. Otherwise, it had a tendency to keep spinning and spinning if you even thought about turning.

Communicating with the ZTW Spiders was a hell of an adventure in its own right, and I am putting this post under Reference Posts because I’m 99% I will need it again or someone else will randomly find it while needing the information. If there was any industry that continually pisses me off with how undocumented and tribal-knowledge focused it is, it’s the R/C anything industry.

So, here’s how everything went down. I lost my AfroESC USB communicator, so I purchased the Spider SPLinker advertised as working with the controllers. I also bought one of these stupid things:

That’s a “SimonK/BLHeli compatible” dongle called the ESCLinker. It allegedly can talk to either kind of ESC, but there was nothing remotely resembling a manual or operating guide; all of the search results for this brilliant device were people complaining that there was no manual.

So I’m writing the manual now: This thing does not want to talk to KKMulticopter Tool (my go-to for flashing SimonK ESCs). It will only talk to BLHeli Suite. As a matter of fact, I couldn’t get the Spider SPLinker to talk to ANYTHING. For all of my tuning here on, I used the ESCLinker tool.

Here is BLHeli Suite, which is hosted on the sketchiest possible website that is one tier above compiling it from the Git repository yourself.

Notice how I’m connected to the ZTW Spider now. The ESCLinker (and SPLinker) install as virtual COM ports.  The necessary baud rate is 38400 baud, not 19200 (Afro/Turnigy USB dongles, to my knowledge)

By the way, once I realized this, I tried to talk to the SPLinker and ESCLinker on KKMulticopter Tool again using 38400 baud; no dice.

Further investigation revealed that the ESCLinker needs these options to communicate to the ESC – both options 2 and 3 will work. So if you’re listening, people mystified by the ESCLinker: Talk to it on 38400 Baud and ask it to communicate to your ESCs with BLHeli/SimonK 4-way-if bootlader.

Ugh. One of my selfish reasons for wanting Brushless Rage is so it’s one known quantity and I never have to dick around with other people’s open-source bullshit again.

So with all that behind me, I decided to try out BLHeli drive on the little pulleybot. I went with intuitive settings based on my SimonK advice, which included “Damped Light” mode, a fancy euphemism for synchronous rectification/complementary PWM, medium to low timing and maximum start power. BLHeli also has a “demag compensation” feature which appears to delay commutation to compensate for current decay in the windings. Who knows!? I wasn’t given the imprssion that its users actually understood what it meant, nor does the manual really say anything useful.

I found that Demag Compensation turned all the way up gave the best performance, along with maximum start power. However’ it still couldn’t compare with my SimonK experience. It seems like even maximum start power is much weaker than what SimonK permits you to do.

Here’s the final test drive I made with the BLHeli Spider ZTWs:

I’m honestly not very impressed. I think BLheli is very much optimized towards multirotors and helicopters (hmm, maybe it’s even called BrushLessHeli for a reason!) and the settings are more high-level and mask the underlying mechanicals of the firmware. I think this makes it much more accessible to hobbyists, though. In the end, I’m not very enamored by it.

These were my final settings:

For a direct comparison, I decided to replace the ESCs with my old SimonK Afro 30 amp units. These have been on quite a few bots now, starting with the original Stance Stance Revolution, and they were completely beat up. But they still worked!

A direct replacement into the existing wiring harness later… we have SimonK!

I found myself in the awkward position of using KKMulticopter Tool to compile a customized SimonK formware, then uploading it via BLHeli Suite because my USB dongles didn’t talk to KKMulticopter Tool; I’d lost my AfroESC USB dongle a long time ago.  BLHeliSuite doesn’t seem to have a firmware editor window that I’ve found yet.

Here it is. I found the SimonK version so much more responsive that I actually needed more counterweight on the front. So, a non-fitting bolt gets zip tied to the nut! Now the bot’s a lot more controllable:

I like it a lot. It might even be worth doing 4WD to give me more yaw damping, or I’d have to design the bot to be well balanced enough on front skids, or something. I used my typical SimonK parameters: complementary PWM, maximum braking power, maximum braking ramp speed, and adjusted start PWM limits to something like 50%.

I’m aiming to get Roll Cake and maybe Colsonbot running for this year’s MomoCon in a couple of weeks, so hopefully I’ll post up some design news soon!