CreeperX robot and mainboard

CreeperX is two things of very different strength. The first is a 12-servo quadruped robot, built as a three-person course project in late 2025. The second is the mainboard I designed for it a year later, on my own: one 100 by 100 mm, four-layer board that replaces the robot's pile of breakout modules. The board is a finished, checked design. It has not been manufactured, so nothing on this page is a measurement on hardware.

The original robot was breakout modules wired to an Arduino Nano or a Raspberry Pi, and its faults were power faults. One small regulator module fed all 12 servos and the controller, with no bulk capacitance. The servo rail was over-volted. The Nano drove 12 servos at the limit of its servo library. There was no battery monitoring, and the servos went live the instant power was applied. The board had to fix each of these without tying the robot to one controller.

Power comes in through an XT60 connector or a screw terminal, then a 15 A fuse, a reverse-polarity transistor, a transient suppressor, and a 5 milliohm shunt that reports pack voltage and current. Three separate 5 A buck converters follow: one for the servos of legs 0 and 1, one for legs 2 and 3, and one for 5 V logic, so a servo brownout cannot reach the controller. The servo rails come up disarmed, behind two independent interlocks. A 16-channel PWM driver runs the servos, and an IMU with hardware free-fall and tilt detection sits on board. Any 3.3 V controller plugs in: an ESP32, a Pico, an STM32, or a Raspberry Pi on its 40-pin header.

The schematic is code. The netlist is written in Python with SKiDL, and placement, the board build, a collision check, the power budget, the design-rule report and the fabrication package are all scripts. Routing was done by an autorouter, Freerouting, plus a repair script of mine. Firmware exists for the ESP32 (a bring-up self-test and a Wi-Fi page to nudge single servos) and as a Raspberry Pi driver. It compiles. It has not run on hardware.

These are design-analysis results, not measurements: no board exists yet. The netlist has 142 components on 89 nets. The electrical rules check is clean apart from deliberate warnings. Routing is about 99% done: 7 short power connections are still open, with no clearance, short or keep-out violations. Parts come to about $14.19 a board at a quantity of five. In the worst case modeled, standing up from a crouch, the servo rails sit at 96% of their rating. The same checks caught 6 silently split nets and 9 colliding reference designators before any order was placed.

The robot itself has no results to report. It has a detection script for the Pi camera that uses a pretrained model, and a bring-up sketch for the servos and sensors. No successful detection run is logged, and there is no walking code. The course reports quote accuracy figures that no code, data or model backs, so none of them appear here.

Close the 7 open connections, fix the silkscreen labels that still overlap, order five boards, and film the first power-up. Until then this is a design, and the page says so.