T-GMP: Terrain-conditioned Generative Motion Priors for Versatile and Natural Humanoid Locomotion

Real-world Deployment We present T-GMP, a terrain-conditioned generative motion priors module that enables robots to exhibit versatile and natural behaviors across diverse terrains. Trained only once, the robot learns whole-body coordination strategies, including natural arm swinging during walking, lowering the center of mass (CoM) on stairs and slopes, and extending the arms for balance when crossing gaps or narrow beams.

Abstract

Achieving both anthropomorphic naturalness and robust terrain traversal remains a fundamental challenge in humanoid locomotion. Existing Reinforcement Learning (RL) approaches typically rely on fixed motion priors, limiting their adaptability to varying environments. We propose Terrain-conditioned Generative Motion Priors (T-GMP), a module that captures a terrain-conditioned latent motion manifold from a few expert state-terrain demonstrations using a Conditional Variational Autoencoder (CVAE). The learned priors enable smooth style transitions, facilitating a unified policy that adapts to terrain variations. We integrate T-GMP into an adversarial learning pipeline with our proposed Foothold Penalty, where a discriminator dynamically modulates naturalness constraints conditioned on local terrain features, guiding the generation of versatile and human-like motions. Experimental results demonstrate that our method outperforms existing baselines in traversal success rate and motion smoothness, while preserving biomimetically natural and physically coordinated motions.

Overview

First research result visualization

Method Overview The overall learning pipeline consists of three parts: (I) collecting expert locomotion data using privileged expert policies and human motion capture, (II) training T-GMP using a CVAE, and (III) training a unified reinforcement learning policy with a terrain- conditioned discriminator. The terrain representation (height map) serves as a conditioning variable across all modules.

Once All

Robot traverse all terrains at once in mujoco.

Style Gallery

Comparison of reference data in MuJoCo with real-world deployment.

Stair Ascent

Stair Descent

Beam

Gap

Stage

Flat

Slope Ascent

Slope Descent

Ablation Analysis

Our Method vs. w/o Foothold Penalty

Stair Ascent

Stair Descent

Our Method vs. w/o Condition

Beam

Gap