Impact Disturbance Rejection for a Humanoid Robot with Optimal Footstep Regulation Trigger

Abstract Rejecting impact force by adjusting footsteps during walking is crucial for a humanoid robot in an interactive environment. This paper proposes an optimal footstep regulation trigger based on the framework of the singular-linear-quadratic-preview (SLQP) walking controller and our footstep adjustment strategy. The trigger avoids regulating the footstep in every cycle to reduce the computational cost. Moreover, adjusting the footstep at the optimal trigger time achieves lower regulation cost than before and after the optimal trigger time. Before implementing the optimal trigger, we propose a method to identify the impact force occurrence based on the feedback acceleration and zero moment point (ZMP). After that, a determining function about system states is calculated over time. According to our analysis, the regulation cost meets the least extremum when the value of the determining function is null. The moment is taken as the optimal trigger time. Our method is demonstrated by experiments with multiple directions of impact forces.