Increasing Humanoid Locomotion Robustness with Black-Box Optimization of DCM-based Trajectories

Abstract

This thesis investigates the use of black-box optimization to improve the robustness of humanoid bipedal locomotion. Robust walking is a fundamental requirement for deploying humanoid robots in real-world environments, where unpredictable disturbances compromise robustness. To address this, a motion planner based on the 3-Dimensional Divergent Component of Motion (3D-DCM) method is extended with an algorithm based on black-box optimization, which tunes high-level motion parameters to influence 3D-DCM trajectories and reduce the effect of induced angular momentum. As a cost function, the quadratic average of the Center of Pressure (CoP) offset relative to the center of the foot is evaluated during each step of the optimization process. No analytical gradients or detailed dynamic models are required, which enables broad applicability and flexibility across different gait patterns. The proposed method is evaluated both in simulation and on the real Torque-controlled humanoid robot (TORO), developed at the German Aerospace Center (DLR). Results show a consistent improvement in robustness across various walking scenarios. The algorithm adapts to different movement types and demonstrates the ability to generalize beyond the motions it was originally tuned for.