Walking Primitive Databases for Perception-Based Guidance Control of Biped Robots

Abstract

This paper presents a systematic approach to the generation of walking primitive (WP) databases for perception-based guidance control of humanoid 3D bipeds. Individual WPs are computed off-line by optimal control techniques such that they can be concatenated into a physically feasible reference trajectory for walking with step length adaptation, direction changing and stepping over or upon obstacles. Zero moment point and friction conditions ensuring stability of the foot contacts, as well as bounds on joint angles and control torques are considered as constraints. Numerical results are presented for ‘optimal’ WPs generated for the human-size 3D walker JOHNNIE taking into account typical energy based performance indexes. Resulting reference trajectories applied to locomotion control are validated by dynamic simulations of the 3D biped walker. An approach to predictive step sequence planning based on the WP database concept is proposed.