Feasibility-Driven Step Timing Adaptation for Robust MPC-Based Gait Generation in Humanoids

The feasibility region of a Model Predictive Control (MPC) algorithm is the subset state space in which constrained optimization problem to be solved feasible. In our recent Intrinsically Stable MPC (IS-MPC) method for humanoid gait generation, means being able satisfy dynamic balance condition, kinematic constraints on footsteps as well an explicit stability condition. Here, we exploit concept build step timing adapter that, at each control cycle, modifies duration current whenever loss imminent due, e.g., external perturbation. proposed approach allows IS-MPC maintain its linearity and adds negligible computational burden overall scheme. Simulations experimental results where robot pushed while walking showcase performance approach.