Dynamic Bipedal Walking Using Real-Time Optimization of Center of Mass Motion and Capture Point-Based Stability Controller

Abstract

This paper proposes a walking pattern generator and a stabilization method for the dynamic walking of a bipedal robot. The capture point (CP) and center of mass (CoM) trajectories were optimized in real-time by using particle swarm optimization, and a desired zero-moment point (ZMP) can be freely generated within a certain area of the support foot by using the proposed method. Unlike a previous method, which optimized only the reference CP trajectory to generate the optimal desired ZMP, the proposed method optimizes the reference CP and the CoM trajectories simultaneously. Consequently, the control input size for tracking the reference CP trajectory is reduced, resulting in reduced CP error. The ZMP error could be reduced as the CoM acceleration was optimally determined to achieve the desired ZMP. In this process, it is possible to ensure that the measured ZMPs can be obtained in the area where a desired ZMP is to be generated, and the robot can walk stably even with reduced feet. The performance of the proposed walking pattern generator and stabilization method based on optimization was verified using a robot simulator, Webots. It was confirmed that in the case of the proposed method, the measured ZMPs are guaranteed to be within the desired area and the control performance of the CP and ZMP is improved.