From Walking to Running: 3D Humanoid Gait Generation via MPC

• DOI: 10.3389/frobt.2022.876613
• Publication Date: 2022-08-23T13:12:49Z
• Authors: Filippo M. Smaldone; Nicola Scianca; Leonardo Lanari; Giuseppe Oriolo
• Keywords: Zero moment point; Ground reaction force; Model Predictive Control

We present a real time algorithm for humanoid 3D walking and/or running based on Model Predictive Control (MPC) approach. The objective is to generate stable gait that replicates footstep plan as closely possible, is, sequence of candidate positions and orientations with associated timings. For each footstep, the also specifies an reference height Center Mass (CoM) whether robot should reach by or running. scheme makes use Variable-Height Inverted Pendulum (VH-IP) prediction model, generating in both CoM trajectory adapted footsteps. VH-IP model relates position Zero Moment Point (ZMP); avoid falling, ZMP must be inside properly defined support region (a extension 2D polygon) whenever contact ground. nonlinearity handled splitting generation into two consecutive stages, requiring solve quadratic program. Thanks particular triangular structure dynamics, first stage deals vertical dynamics using Ground Reaction Force (GRF) decision variable. Using given stage, horizontal become linear time-varying. During flight phases, collapses free-falling mass model. proposed formulation incorporates constraints order maintain physically meaningful values GRF, keep during ensure footsteps are kinematically realizable. Most importantly, stability constraint enforced time-varying guarantee bounded evolution respect ZMP. Furthermore, we show how extend technique perform tilted surfaces. describe simple receives input high-level velocity commands generates form required MPC scheme. validated via dynamic simulations full-scale HRP-4, well experiments small-sized OP3.