Ten days, one robot, three control loops. How our ESP32 balancing robot on brushed motors learned to stand, to come back after a push, to hold its spot and to drive over Wi-Fi, with the numbers from its logs and the firmware that made it work.
Brushed motors, an ESP32 and an MPU-6500 — and three control loops, each one fixing what the one before couldn't. The angle loop (PID, 500 times a second) keeps it upright, but it rolls 27 cm away. The speed loop brings it back to a 30 cm band. The position loop brings it home within 3 cm — then it drives over Wi-Fi and jumps off a bathroom scale. Three loops, one LQR, or a neural net — which would you bet on?
I push, it comes back. This frame balances in one plane on the thrust of a single propeller: an MPU-6500 sensor measures the tilt, and an ESP32 runs a PID controller (Kp 8 · Ki 2 · Kd 2.2 in this test) that drives the motor through a DRV8833. Bluetooth sends the tilt and the motor command to a page on my computer, where I can adjust the PID settings. Earlier attempts? Gravity was consistent. What next: two planes, or wheels?
Building a compact test bench for experimenting with a DC motor encoder and PID control.
