Realization of dynamic walking for the humanoid robot platform KHR-1

Abstract

This paper presents three online controllers for maintaining dynamic stability of a humanoid robot and describes simplified walking patterns for easy tuning in real experiments. The legs of a humanoid robot are relatively long and serially connected with compliant force/torque sensor at the ankle. This architecture has the inherent characteristics of a lightly damped system. Most research on balance control overlook the deterministic vibration caused by structural compliance. In addition, the vibration was not positively considered to improve the characteristics of the system. Therefore, a simple inverted pendulum model with a complaint joint is proposed. The proposed model has an advantage in easy parameter identification by experiment. For this model, the damping controller that increases system damping is proposed as a balance controller. Furthermore, the performance of maintaining balance against external forces is experimentally shown. A landing orientation controller at the ankle joints is presented to manage fast and stable ground contact. A landing position controller is implemented in order to modify the prescribed trajectory of the swing foot and to reduce the landing impact during unexpected landing. The effectiveness of the proposed controllers is confirmed by walking experiments that have been applied on the KAIST humanoid robot platform KHR-1.