Big Data EVT

“8 oz” you said. This dang bar, man.


Battlebots in the Cloud

Big Data EVT drive

To cel­e­brate mov­ing to the Bay Area, and by “cel­e­brate” I mean “for­got that I reg­is­tered for a robot com­bat event in a farm show com­plex in Har­ris­burg back when I was still liv­ing on the East Coast and now have to beast one so I can fly it into afore­men­tioned Penn­syl­van­ian snow pile,” I start­ed liv­ing in the Google Work­shops around Jan­u­ary with a 12.4 oz chunk of hard­ened steel, a fever­ish­ly draft­ed new design, and dreams of brush­less synergy.

Big Data bar

The con­cept for my “new1″ beetleweight (3 lb) Big Data is sim­ple: a “high” volt­age2 sen­sored brush­less dri­ve­train cou­pled to the beat­er bar from Cake ReMix. With a ver­ti­cal­ly-hit­ting weapon like the bar, the robot has to hit the wall in order to right itself to the cor­rect orientation.

This means that the robot needs be able to dri­ve upside-down. The dri­ve­train was designed around ridicu­lous sen­sored inrunners.

Maxon 200118

These Max­on EC 22 200118 motors are tuned for mo’ torque. They’re filled inside with globs of hun­dreds of turns of mag­net wire thin­ner than hair to cre­ate a 400 RPM/V motor that’s only 22 mm in diam­e­ter. To give you an idea of how “cool” these are wound, they have 11.7 Ohms of phase-to-phase resistance.

The first order of busi­ness is to take the gear­head off, since it was so well jammed on the motor that I could­n’t get it off even with a heat gun and much vice gripping.

Maxon meets hacksaw

The con­nec­tors are huge!

Bare motors

These were well used bar­gains from eBay. On one motor, a hall sen­sor which detects the rotor posi­tion was stuck at zero volts all the time. The motor had some vis­i­ble rust on it, so I pulled it apart to have a look at the damage.

Oh shi-


Pulling off the ass of the motor had the exact same tug­ging and twang­ing as pluck­ing three pairs of hairs out of your head. The hall sen­sor board also ter­mi­nates the six leads out of the wind­ings3, and I had pulled it clean off.

Sensor water damage

Also that round mag­net, shaft, and a sen­sor all def­i­nite­ly have rust on them. Time for some shady repairs.

Maxon winding "repair"

With tweez­ers I tied more mag­net wire onto the remain­ing stubs and sol­dered them in place. Pen­cil eras­er for scale.

As for the dead sen­sor… well, let’s see if you can fig­ure out what I did.

Sensor repair 2

Sensor repair 1

I may or may not have tried to sand down the new sen­sor to make it thin­ner and had to repeat this whole repair because elec­tron­ics don’t work like that. >.>




Also I took a trip to Squaw Val­ley woooo North­ern California.


Wire it up and it runs. Ship it!

Oh and what are they con­nect­ed to, you ask?


These are Corn­flakes, ver­sion 3.0 of the beloved Corn­troller series. Same good ol’ ingre­di­ents, new packaging.

The motor bell is a new design. The face of the bell has mag­nets pressed in that grip the steel weapon bar and trans­mit torque through fric­tion. This allows slip when the bar is hit, reduc­ing dam­age to the motor in the worst case, or pre­vent­ing sen­sor­less dri­ver desync in the best case.

Green sauce

The motor can is lit­er­al­ly glued in with “green sauce,” or Loc­tite press fit retain­ing com­pound. It’s anaer­o­bic methyl acry­late adhe­sive that acti­vates in the absence of air in tight spaces. There’s also a less vis­cous ver­sion of green sauce that can wick into slip fits through cap­il­lary action.

Big Data EVT

The whole thing was designed in a week and built in a week. It was seri­ous­ly tremen­dous­ly rushed job, with most of the design work put towards reduc­ing man­u­al tasks (more water­jet!) and cut­ting many cor­ners (inac­cu­rate mass mod­el­ing, miss­ing holes and fasteners).

I usu­al­ly spend months design­ing and com­mit­ting thoughts to paper notes, before trans­fer­ring to Solid­Works fea­tures over a week or two. But in this case, the whole process was com­pressed down to a week. I had doubts about it before even send­ing off the water­jet files.

Any­ways, ship it (to Harrisburg)!

Dan, Aaron, and Jim setting up

Everything about this robot sucks

Seri­ous­ly. Even when not dam­aged, the weight dis­tri­b­u­tion is so far for­ward that Big Data could use hard­ly any of its beefy dri­ve, the Hob­byK­ing 6S con­troller could sync only to the weapon motor once every ten tries, and even when it did the mag­net­ic clutch did­n’t grip the beat­er bar tight­ly enough to spin it well.

The real clinch­er was how frag­ile every­thing was. The dri­ve motors were only face mount­ed with three M2 screws each to a frame rail, so after the first hit the mount points stripped out and both motors were hang­ing out in free space.

Using a 5‑face box made of 1/4 in 7075 was also a bad idea, as was putting the weapon between two unsup­port­ed rails that eas­i­ly snapped off. Var­i­ous screws were loaded in shear. Heck, even the dri­ve gears did­n’t mesh that well. Every­thing that could be crap­py about an hon­est attempt at a non-ass bot, I made crappy.

But the Corn­flakes held up fine, so I taped the dri­ve motors back in, fixed the weight dis­tri­b­u­tion by remov­ing the beat­er bar, and tossed it once more unto the breach.

This is one cloud tech­nol­o­gy that failed to disrupt.


On the oth­er hand, Sheboygan.


  1. Remem­ber I’m back­track­ing to ear­ly 2014. []
  2. At 6S: six lithi­um poly­mer cells in series, for a max­i­mum of 6 × 4.2 V = 25.2 V. []
  3. Con­fig­ured in a “wye” ter­mi­na­tion, if you cared. []