Real-time DCM-based Agile Locomotion and Optimum Push Recovery of a 3D Humanoid Robot

This paper aims to achieve real-time optimum push recovery and high-speed pattern generation for a 3D humanoid robot.  For this purpose, a divergent component of motion (DCM) based locomotion is planned using a three-mass inverted pendulum model. To improve push recovery robustness in the presence of strong disturbances during walking, a quadratic optimization strategy algorithm is introduced, including weighted step time adaption, step position adjustment, and ankle strategy. The minimum error push recovery cost function is obtained through quadratic programming. To validate our proposed model different scenarios for real-time navigation of robot considering strong disturbances in various speeds of walking from 0/8 to 1/6 meters per second are analyzed. The verification methods indicate that the proposed method recovered from 34.5 N.s disturbances at 1/6 meters per second speed, while conventional push recovery methods guaranteed lower speeds robust stable locomotion in the same disturbances. In this paper, a real humanoid robot and actual physical conditions were considered for validation of navigation and robust optimum push recovery.