Adapting Gait Frequency for Posture-regulating Humanoid Push-recovery via Hierarchical Model Predictive Control

TL;DR

This work tackles humanoid push-recovery under unknown disturbances with an emphasis on posture regulation during recovery. It introduces a hierarchical-MPC architecture that fuses high-fidelity whole-body dynamics for the initial response with a centroidal-augmented reduced-order model for efficient prediction, complemented by a posture-aware gait frequency adaptation planner and a convex locomotion MPC to execute tailored stepping. The key contributions include a CSRB-based dynamic model, a push-recovery NMPC formulation, and two gait-frequency laws that modulate swing duration to regulate body attitude. In both simulations and hardware experiments with the HECTOR V2, the proposed framework achieves larger recoverable impulses, earlier and more robust stepping, and reduced body attitude changes, demonstrating practical improvements for posture-preserving push-recovery in dynamic loco-manipulation tasks.

Abstract

Current humanoid push-recovery strategies often use whole-body motion, yet they tend to overlook posture regulation. For instance, in manipulation tasks, the upper body may need to stay upright and have minimal recovery displacement. This paper introduces a novel approach to enhancing humanoid push-recovery performance under unknown disturbances and regulating body posture by tailoring the recovery stepping strategy. We propose a hierarchical-MPC-based scheme that analyzes and detects instability in the prediction window and quickly recovers through adapting gait frequency. Our approach integrates a high-level nonlinear MPC, a posture-aware gait frequency adaptation planner, and a low-level convex locomotion MPC. The planners predict the center of mass (CoM) state trajectories that can be assessed for precursors of potential instability and posture deviation. In simulation, we demonstrate improved maximum recoverable impulse by 131% on average compared with baseline approaches. In hardware experiments, a 125 ms advancement in recovery stepping timing/reflex has been observed with the proposed approach. We also demonstrate improved push-recovery performance and minimized body attitude change under 0.2 rad.

Figures (6)

• Figure 1: Push-recovery Experiment Snapshots on HECTOR V2. Robot rebalances after a kick. Full experiment video: https://youtu.be/pUdVy0RSaiE
• Figure 2: System Architecture In high-hierarchy push-recovery NMPC, we employ a hierarchical dynamics model, fusing whole-body dynamics (WBD) in the first horizon and reduced-order dynamics in the rest. The WBD captures the effect of external impulse on float-base acceleration $\ddot{\mathbf q}_{fb}$, providing an accurate state evolution to start with. Our gait frequnecy adaptation planner gathers the unstable duration $\Delta t^u$ in NMPC prediction and associated body attitude deviations, determining a tailored stepping frequency in terms of MPC step duration and adapting contact sequence $\sigma$ for locomotion MPC to follow.
• Figure 3: Illustration of CoM States Prediction under External Push. With hierarchical-dynamics-based modelling in MPC
• Figure 4: Push-recovery Comparative Experiment Snapshots with Baseline Methods. Snapshots and plots of NMPC-prediction-based stepping trigger vs. current-position-based stepping trigger. CoM $y$-position plot is zoomed in to showcase the prediction trajectory of NMPC for time-advanced stepping triggering.
• Figure 5: Maximum Recoverable External Force and Moment Impulse Radar Plots in 8 Directions. Comparative analysis of (a) with different control approaches and (b) with different stepping frequency laws.