Competition robot: designing for repeated high-energy impact
A reliability testbed that punishes every shortcut — and where a destroyed capacitor turned into a transient-survivability design rule.
Outcome Root-caused a recurring input-capacitor failure to hot-plug LC ringing, and fixed it in component selection rather than by adding protection.
Competitive robotics reads as a hobby until you look at what it actually demands: hardware that takes repeated high-energy collisions across a competition weekend, with no opportunity to repair between rounds, and no possibility of a firmware patch mid-match.
No product specification asks for that. Which makes it an unusually honest test of whether a design is genuinely robust or merely untested.
The failure that became a design rule
Input capacitors kept failing. Not immediately, not consistently — but often enough that it was clearly a design fault rather than bad luck.
The symptom pointed at overvoltage, but the supply was well within rating. The cause was the moment of connection itself: plugging a live battery into a board with bulk capacitance produces an inrush current through the wiring inductance, and that L and C form a resonant circuit. The transient overshoot can reach close to twice the nominal supply voltage for a few microseconds — long enough to exceed a capacitor’s rating, far too short to see without looking for it deliberately.
Two things fixed it, and neither was “add a TVS and hope”:
Voltage headroom sized for the ring, not the rail. Rating parts for the nominal supply is the mistake. They have to survive the transient.
Soft-termination MLCCs in a 1210 package. The other half of the failure was mechanical. Large ceramic capacitors crack under board flex — and a robot that gets hit flexes. Soft-termination parts use a compliant layer between the ceramic and the solder joint that absorbs that movement.
That combination — electrical derating for the real transient, mechanical selection for the real environment — is exactly what a medical or automotive design reviewer probes for. It is much easier to explain having actually broken something.
Motor drive migration
The drive system moved from a hobby-grade controller to a VESC-based platform, with a precision brushed motor handling the positioning axis. Command and telemetry over UART and CAN.
The interesting part was not the swap but the commissioning: motor parameter detection, current limits set against what the mechanics can actually survive rather than what the ESC can deliver, and failure behavior defined for the case where a motor stalls against an opponent — which happens in most matches.
A sensor board with nowhere to hide
The blade sensor board is built around an STM32L4 with an IMU and a comparator front-end, in a footprint dictated entirely by the mechanics.
Two constraints fight each other. The IMU needs mechanical isolation from vibration to produce usable data; the board needs to be rigid enough to survive impact. Resolving that is placement and mounting work, not schematic work — and it is the same problem as mounting an IMU on an airframe, at a smaller scale and with much higher shock.
Mechanical
Structural parts printed in polycarbonate with gyroid infill and shrinkage compensation on the critical dimensions. PC is unpleasant to print and worth it: it survives impacts that crack anything easier.
Why this is on a professional site
Because diagnostic ability is the hardest thing to advertise and the easiest thing to demonstrate.
Anyone can write “We do root-cause analysis.” Fewer can describe a specific recurring failure, the measurement that identified it, why the obvious explanation was wrong, and the component-level fix that resolved it. If your board is failing intermittently in the field and nobody can reproduce it on a desk, this is the kind of work you are actually buying.
The transient mechanism is worth its own write-up — see hot-plug transients and why your input capacitors keep dying.