Robustness Experiments for a Planar Hopping Control System

Abstract

We explore the robustness of a control system for a pneumatic monopod simulation by adding Gaussian noise to the sensors and actuators. The control system is based on Raibert's three-part control system decomposition; with significant modifications to two of the control loops. Our speed controller uses a neural network to approximate the neutral point function, and we use a model-based height controller. No changes were made to the attitude controller. Simulation experiments show that the control system performs stably in the noisiest case, with a relative error of approximately 20%. The control system is expected to perform comparably on the real robot since our actual sensors are more accurate compared to the sensors simulated in these experiments.