A Comparison of Multivariable & Decentralized Control Strategies for Robust Humanoid Walking

Abstract

Bipedal walking is one of the most interesting control problems in humanoids research. Walking is modelled as a hybrid system in the sense that it involves various phases such as single support phase, impacts with the ground (i.e. a state reset) and the double support phase. The control system has to provide good dynamic performance in these different modes to achieve fast walking speeds while guaranteeing its safe and robust operation. Most humanoids use local joint PID loops (decentralized) control systems while the robot is a multivariable system and walking involves significant interactions between the robot links. Hence in this paper a centralized LQR multivariable controller is designed for the robot and analyzed for its stability, robustness to noise and disturbances and dynamic performance. Then, an LQR based iterative algorithm is used to tune the local PID servos. A comparison between the two schemes is done, where it is shown that the multivariable LQR has better robustness and energy efficiency. Finally, both controllers are simulated using the linearized model of a 10 degree of freedom robot called “C-Cub”.