Robustness and Push Recovery of Humanoid Robots Using Predictive and Whole-Body Control

Abstract

Robustness walking and recovery ability in uneven terrains and unexpected collisions are crucial for the practical application of humanoid robots. However, existing methods struggle to effectively balance stability, motion safety and walking agility. This paper proposes a hybrid control strategy combing Nonlinear Model Predictive Control (NMPC) and Whole-Body Control (WBC) to improve the robot's walking stability and reaction safety in collisions situations. The NMPC generates long-term and macroscopic movements, with capture point theory being utilized for foot placements determination under external impacts. On this foundation, the WBC regulates the robot refinedly in real-time based on task priorities and weights. Simulations verify the effectiveness of the proposed method, involving uneven terrain walking and external impacts rejection.