Model-Free Robust Adaptive Control of Humanoid Robots With Flexible Joints

Abstract

A model-free robust adaptive controller is proposed for control of humanoid robots with flexible joints. The proposed controller uses a time-delay estimation technique to estimate and cancel nonlinear terms in robot dynamics including disturbance torques due to the joint flexibility, and assigns desired dynamics specified by a sliding variable. A gain-adaptation law is developed to dynamically update the gain of the proposed controller using the magnitude of the sliding variable and the gain itself. The gain-adaptation law uses a leakage term to prevent overestimation of the gain value, and offers stable and chattering-free control action. The effectiveness of the proposed adaptive controller is experimentally verified on a humanoid robot equipped with flexible joints. Tracking performances of the autotuned adaptive gain are better than those of the manually tuned constant gains. The proposed control algorithm is model-free, adaptive, robust, and highly accurate.