Rocking the Bongo Board: Humanoid Robotic Balancing on Dynamic Terrain

Abstract

This thesis presents a comparison of multiple control algorithms designed to allow a robot to balance on unstable terrain. To evaluate these algorithms I use a bongo board, a simple apparatus consisting of a deck positioned above a free-rolling wheel. I program a small humanoid robot to stand on the deck, controlling its pose to keep the deck from falling off the wheel. I implement three different control algorithms, derived from solutions to the well-known cart-and-rod inverted pendulum problem, including PID control, Fuzzy Logic, and Always-On Artificial Neural Networks. These algorithms are used with two different control policies: Do the Shake, where the robot reacts to external forces and Let's Sway, where the robot introduces a rhythmic oscillation to the system to promote dynamic stability. Using identical experimental conditions with a physical robot, I show that both PID and Fuzzy Logic control are well-suited to active balancing on unstable terrain.