PARTITION-AWARE STABILITY CONTROL FOR HUMANOID ROBOT PUSH RECOVERY

For successful push recovery control in response to unpredicted perturbations, a humanoid robot must select the appropriate stabilizing action from a wide range of available strategies. Existing approaches are restricted to a limited range of balancing motions because they are often derived from reduced-order models and ignore system-specific aspects such as swing leg dynamics or joint kinematic/actuation limits. In this work, a novel partition-aware push recovery controller is introduced that can select between the ankle, hip, and captured stepping strategies for balance by evaluating the robot’s center-of-mass (COM) state with respect to different levels of criteria. The criteria are the partition-based stability regions in the augmented COM state space, which are numerically constructed for each balancing strategy. For stepping, both free- and fixed-arm capturability regions were obtained to quantify the effect of arm momenta on balancing capability. The regions are precomputed for control with an optimization-based method that incorporates whole-body system dynamics, contact interactions with the ground, system-specific characteristics, and requirements of the corresponding strategy. Through simulation experiments, the proposed approach was demonstrated to allow the controller to fully exploit a humanoid robot’s balancing capability and validate the use of pre-computed stability regions as explicit criteria to initiate a proper balancing motion, in contrast to the use of incomplete or implicit criteria in existing controllers.