Advancing Humanoid Locomotion: Mastering Challenging Terrains with Denoising World Model Learning

2024/08/26 by Xinyang Gu, Gu, Xinyang, Yen‐Jen Wang +11 · 25 citations
Computer Science · Engineering · #Human Pose and Action Recognition #Human Motion and Animation

paper · pdf · doi:10.48550/arxiv.2408.14472

Abstract

Humanoid robots, with their human-like skeletal structure, are especially suited for tasks in human-centric environments. However, this structure is accompanied by additional challenges in locomotion controller design, especially in complex real-world environments. As a result, existing humanoid robots are limited to relatively simple terrains, either with model-based control or model-free reinforcement learning. In this work, we introduce Denoising World Model Learning (DWL), an end-to-end reinforcement learning framework for humanoid locomotion control, which demonstrates the world's first humanoid robot to master real-world challenging terrains such as snowy and inclined land in the wild, up and down stairs, and extremely uneven terrains. All scenarios run the same learned neural network with zero-shot sim-to-real transfer, indicating the superior robustness and generalization capability of the proposed method.

Cited by

• EGM: Efficiently Learning General Motion Tracking Policy for High Dynamic Humanoid Whole-Body Control
• Learning Terrain Aware Bipedal Locomotion via Reduced Dimensional Perceptual Representations
• Gait-Adaptive Perceptive Humanoid Locomotion with Real-Time Under-Base Terrain Reconstruction
• Coordinated Humanoid Manipulation with Choice Policies
• H-Zero: Cross-Humanoid Locomotion Pretraining Enables Few-shot Novel Embodiment Transfer
• Agility Meets Stability: Versatile Humanoid Control with Heterogeneous Data
• VIRAL: Visual Sim-to-Real at Scale for Humanoid Loco-Manipulation
• Unified Humanoid Fall-Safety Policy from a Few Demonstrations
• Towards Adaptive Humanoid Control via Multi-Behavior Distillation and Reinforced Fine-Tuning
• Learning Vision-Driven Reactive Soccer Skills for Humanoid Robots
• Embracing Evolution: A Call for Body-Control Co-Design in Embodied Humanoid Robot
• UniReLo: Learning a Unified Humanoid Policy from Fall Recovery to Locomotion across Diverse Terrains
• A Comprehensive Survey on World Models for Embodied AI
• Architecture Is All You Need: Diversity-Enabled Sweet Spots for Robust Humanoid Locomotion
• Towards Adaptable Humanoid Control via Adaptive Motion Tracking
• Preference-Conditioned Multi-Objective RL for Integrated Command Tracking and Force Compliance in Humanoid Locomotion
• D3Grasp: Diverse and Deformable Dexterous Grasping for General Objects
• Residual Off-Policy RL for Finetuning Behavior Cloning Policies
• HDMI: Learning Interactive Humanoid Whole-Body Control from Human Videos
• Track Any Motions under Any Disturbances
• Contrastive Representation Learning for Robust Sim-to-Real Transfer of Adaptive Humanoid Locomotion
• Robot Trains Robot: Automatic Real-World Policy Adaptation and Learning for Humanoids
• Geometry-Aware Predictive Safety Filters on Humanoids: From Poisson Safety Functions to CBF Constrained MPC
• MASH: Cooperative-Heterogeneous Multi-Agent Reinforcement Learning for Single Humanoid Robot Locomotion
• BarlowWalk: Self-supervised Representation Learning for Legged Robot Terrain-adaptive Locomotion

Related

• Learning Humanoid Locomotion over Challenging Terrain
• Learning Humanoid Locomotion with World Model Reconstruction
• Learning Generic and Dynamic Locomotion of Humanoids Across Discrete Terrains
• PPF: Pre-training and Preservative Fine-tuning of Humanoid Locomotion via Model-Assumption-based Regularization
• Real-World Humanoid Locomotion with Reinforcement Learning