MoRE: Mixture of Residual Experts for Humanoid Lifelike Gaits Learning on Complex Terrains

TL;DR

This paper introduces MoRE, a novel RL framework using a mixture of residual experts and multi-discriminators, enabling humanoid robots to learn lifelike gaits on complex terrains with controllable gait switching.

Contribution

The work presents a two-stage training pipeline with a mixture of latent residual experts and multi-discriminators for lifelike gait learning on complex terrains, including gait switching capabilities.

Findings

Effective traversal of complex terrains in simulation and real-world

Seamless switching between multiple human-like gait patterns

Enhanced control over gait behaviors using gait rewards

Abstract

Humanoid robots have demonstrated robust locomotion capabilities using Reinforcement Learning (RL)-based approaches. Further, to obtain human-like behaviors, existing methods integrate human motion-tracking or motion prior in the RL framework. However, these methods are limited in flat terrains with proprioception only, restricting their abilities to traverse challenging terrains with human-like gaits. In this work, we propose a novel framework using a mixture of latent residual experts with multi-discriminators to train an RL policy, which is capable of traversing complex terrains in controllable lifelike gaits with exteroception. Our two-stage training pipeline first teaches the policy to traverse complex terrains using a depth camera, and then enables gait-commanded switching between human-like gait patterns. We also design gait rewards to adjust human-like behaviors like robot base height. Simulation and real-world experiments demonstrate that our framework exhibits exceptional performance in traversing complex terrains, and achieves seamless transitions between multiple human-like gait patterns.