Big Data: Test & Refresh

When I built the first Big Data in 2013, I was just getting to grips with my new surround­ings. I even added the EVT moniker from my new job: in hard­ware devel­op­ment, “engi­neer­ing vali­da­tion test” means both test­ing the earli­est proto­types and the proto­types them­selves.

So overweight it needed its letters clipped...


I was over­whelmed by the Bay Area and toned down my typi­cally masochis­tic design philos­o­phy. I used a less-exper­i­men­tal (proven, even) weapon system1 and relied on tried-and-true water­jet-style design.

I allowed one indul­gence, which was build­ing a new brush­less controller appro­pri­ate for this scale and for the sensored 22 mm trac­tion motors.

This is all great except the mechan­i­cal design didn’t work. The wheels were nowhere near the center of grav­ity, so that it didn’t really move except in short bursts, and the trac­tion motors were each held in with five threads worth of four M2 screws on the face, which is to say they weren’t held in.


Then came Big Data DVT (design vali­da­tion test) and the start of the orange. Embold­ened by famil­iar­ity with new capa­bil­i­ties, I used the CNC mills avail­able to me and casted parts from polyurethane. Trac­tion motors got even beefier while the weapon became more opti­mized… to the point of unre­li­a­bil­ity. The wheels shift so close to the center of grav­ity that just 0.54 g of accel­er­a­tion causes DVT to pop a wheelie.

Naked wheels held on with un-thread­lock­ered shoul­der screws reflected an opti­mism and noncha­lance to robot build­ing of college days past. The trip to Atlanta, comped by asso­ci­a­tion to Google’s recruit­ing at Geor­gia Tech and house calls to the local data­cen­ter, laid excru­ci­at­ingly bare how not to construct a vaca­tion (or robot).

An aside: there’s a certain art to express­ing your­self during a bot match. Jamo is a virtu­oso both with the R/C trans­mit­ter and with situ­a­tion­ally appro­pri­ate utter­ances of raw emotion. I might be on the other end of that spec­trum with my twitchy fingers and a vocab­u­lary of only “oh” and “no.”

Big Data DVT CAD

Also of note is that I’ve ditched machined hori­zon­tal chas­sis elements completely; i.e. there is no top or bottom plate that holds other compo­nents in place. I’ve favored this verti­cal construc­tion style since build­ing sumo in high school. I reused the round-stand­off-between-plates tech­nique used in Flapjack’s base because of the easy assem­bly and smooth curves it lends to a design. In Flap­jack they were used to mount bear­ings for the shell; here, one of the beams—a hard­ened steel shaft, bored through and threaded on both ends—likewise supports the weapon from the milled aluminum “book­ends.” To not load the screws in shear, the chas­sis is coun­ter­bored for the shaft to nestle into.


With all the cool logic of a frus­trated toddler spurned at the toy store (“I WANT IT”), Big Data—and by Big Data I mean I—doubled down on the design out in the park­ing lot (“BUT BUT I NEEEED IT”). Produc­tion veri­fi­ca­tion test (PVT) rang me up demand­ing none other than a weapon and drive­train each taking up 30% of the weight. That left just 40% for a weapon motor, chas­sis (with all the bits & bobs), and elec­tron­ics.

Big Data PVT

Whereas DVT used aluminum trac­tion motor mounts fixed to the chassis’s two round beams, PVT’s drive motors were contained along with PCBs and batter­ies in a 3D printed “scaf­fold” wrapped around its three frame beams.

Machining right chassis case
Test fit right chassis case
Test fit whole chassis

PVT was the first Big Data that entered combat with any semblance of armor. Its outside aluminum rails (15 g) served as double supports for the geartrain dead shafts and protected the wheels and gears a bit. They’re pretty.

Big Data PVT side rails

Unfor­tu­nately, my custom motor drivers were less protected against the ESD hazards of dry winter air. Nor could the 3D printed scaf­fold with­stand direct attacks from Chad’s saw and Kyle’s hori­zon­tal bar.

Assembled Big Data PVT

Why do I have to fight Geor­gia Tech Robo­Jack­ets?



PVT was under­weight by at least 80 grams, and the frame shows basi­cally no damage. Its access hatch made it much easier to work on than any other Big Data. If it weren’t for the too-wimpy weapon motor and that I misplaced half the chas­sis, I’d even consider using it again.


I sold out. Brush­less trac­tion power affords me ridicu­lous speed but I’m not a good enough driver nor do I have enough time to prac­tice to take advan­tage of that. And to be honest, I was getting a tad bored with this design.

Big Data 4

I acqui­esced to the voices call­ing for a bigger weapon motor and smaller brushed trac­tion motors. Big Data 4 finally has armor, and loads of it. I even tossed in my first ever wedges in combat.

Hoop armor wedgelet

Although at Drag­onCon 2015 I kind of just knifed those in.

Whittling UHMW hoop

Fear of Design Spaces

Another aside: I suffer from what I call creative agora­pho­bia. I’m para­lyzed by too open of a design space, too many free vari­ables, too few constraints. When I’m design­ing, this is the screen that panics me the most.

SolidWorks user interface

Having a free-to-use weapon bar and a gung-ho atti­tude to brush­less trac­tion were “hard” constraints that helped shape the last three Big Datas. I’ve heard that called “compo­nent-based design.”

Conversely if I consider the frame design a fixed estab­lish­ment, trans­formed from free vari­able to set constraint, it gets easier to tweak other compo­nents I took for granted, like the silly-ly large trac­tion motors. To get a design that worked, it was time to back off from those Portescap motors.

Maxon 139885 Gearmotor

These are 13 mm diam­e­ter Maxon 67:1 gear­mo­tors, sold as “235 RPM” units on eBay. They’re… OK. The construc­tion is great, with large robust steel gears and gear­head mating threads built into the motor itself. But, they’re defi­nitely slow for beetleweight combat and their precious metal brushes do not enjoy over­volt­age oper­a­tion at all. Maybe they become more useful after remov­ing a stage of plan­e­tary gear reduc­tion.

For now, they serve to provide slow but torquey drive trac­tion for Big Data 4.

In its debut fight, BD4 promptly gets bottomed out like a Google shut­tle at 23rd St.

Google Shuttle Stuck

At this point, the hoop armor has yet any wedgelets at the front, although they’d be inef­fec­tive anyways against the floor-scrap­ing scoop on Wedgee. Also at this point, I’m still tipsy from the night before and have had three hours of like­wise inef­fec­tive sleep. Which some­how leads to better driving than ever before. ¯\_(ツ)_/¯

Of course in my next match, I drive myself right out through the pushout.

Two months later I fight Aaron’s Margin of Safety, the new Drag­onCon cham­pion. It went as you might expect.

So before I get to the design vulner­a­bil­i­ties this revealed, let me get to the crip­pling misman­aged state that BD4 was in for the ComBugs Cup fight. None of the screws had thread­locker on them because I had taken Big Data apart for ship­ping. The loose left-side weapon shaft screw comes off with a “PING!” and a black blur at 0:39 in the video, so the shaft is single-supported after that.

Big Data 4 Eating Itself

Also, the set screw that holds the weapon motor in its fire­wall mount had never been tight­ened. This lets the the motor slide out of posi­tion and… into the weapon bar.

Big Data 4 Weapon Motor/Pulley

On the other hand, the UHMW hoop armor works well at keep­ing Margin’s disk away from the vulner­a­ble nylon shell. It’s not low or thick enough to protect the wheels, but that’s a constant danger in any fight against an under­cut­ter.

Big Data 4 Post-Fight

The killing blow (1:36 in the video) was an incred­i­bly vicious hit with Margin’s single tooth engag­ing the full wall thick­ness of the chas­sis piece.

Big Data 4 Frontal Damage

It applied forces towards the outside of the piece (i.e. towards the hogout wall). This concen­trates stress into the sharp inside corners of the hogout, caus­ing a frac­ture that prop­a­gated into the walls of the bulk­head.

Big Data 4 Chassis Damage

As designed, the piece is meant to break or wear down at the lower corner when hit with­out distort­ing the frame geom­e­try. Unfor­tu­nately it frac­tured way more than expected and the loss of the weapon shaft screw allowed the shaft coun­ter­bore to distort, making reassem­bly impos­si­ble.

Big Data 4 Dead Piece

However, the “hard parts” like weapon shaft and wheel axles survived with mini­mal distor­tion or thread damage, ready for reuse in another frame. All the motors and drive­train are fine as well. For my first fight with a spin­ner, it wasn’t alto­gether that bad.

It’s an engineer’s natural instinct to tinker with an exist­ing design. However, I’m not going to mess with Big Data too much; I promise to refine and ship a more reliable—not sexier—product.

See you at Motorama 2016?

  1. Although that would be a first for me, so does it still count? []