Autotuning Bipedal Locomotion MPC with GRFM-Net for Efficient Sim-to-Real Transfer

TL;DR

This work tackles the challenge of manually tuning model-based bipedal locomotion MPC parameters while bridging the sim-to-real gap. It combines DiffTune, a differentiable-programming-driven autotuning method, with a Ground Reaction Force-and-Moment Network (GRFM-Net) to augment a low-fidelity SRBM dynamics model, preserving differentiability and improving fidelity to real hardware. The approach is validated on the HECTOR robot: DiffTune with GRFM-Net learns MPC parameters that outperform expert-tuned settings, reducing multi-objective loss by up to 40.5% in hardware tests, and ablation without GRFM-Net highlights the necessity of sim-to-real gap mitigation. The results demonstrate a practical, generalizable pipeline for efficient, data-driven auto-tuning of MPC for legged locomotion with improved transferability to real robots.

Abstract

Bipedal locomotion control is essential for humanoid robots to navigate complex, human-centric environments. While optimization-based control designs are popular for integrating sophisticated models of humanoid robots, they often require labor-intensive manual tuning. In this work, we address the challenges of parameter selection in bipedal locomotion control using DiffTune, a model-based autotuning method that leverages differential programming for efficient parameter learning. A major difficulty lies in balancing model fidelity with differentiability. We address this difficulty using a low-fidelity model for differentiability, enhanced by a Ground Reaction Force-and-Moment Network (GRFM-Net) to capture discrepancies between MPC commands and actual control effects. We validate the parameters learned by DiffTune with GRFM-Net in hardware experiments, which demonstrates the parameters' optimality in a multi-objective setting compared with baseline parameters, reducing the total loss by up to 40.5$\%$ compared with the expert-tuned parameters. The results confirm the GRFM-Net's effectiveness in mitigating the sim-to-real gap, improving the transferability of simulation-learned parameters to real hardware.

Figures (3)

• Figure 1: Autotuning Bipedal MPC Parameters Through DiffTune and GRFM-Net. Supplementary video: https://youtu.be/bfrBW2hIT1M
• Figure 2: Force and moment profiles for GRFM-Net fidelity check
• Figure 3: Experimental snapshots of three trajectories and associated CoM position and Euler angles tracking plots.