AMO: Adaptive Motion Optimization for Hyper-Dexterous Humanoid Whole-Body Control

2025-05-06

AI Summary

Humanoid robots face significant challenges in achieving real-time, robust, and sim-to-real transferable whole-body motion—such as floor pick-up—under high-dimensional actuation and strongly nonlinear dynamics constraints. This paper introduces AMO, a novel framework that pioneers a synergistic paradigm integrating reinforcement learning (RL) with nonlinear trajectory optimization. To mitigate distribution shift in motion imitation RL, we construct a hybrid simulation–real dataset and enable online adaptation to out-of-distribution (OOD) task instructions. Evaluated on the 29-DoF Unitree G1 humanoid platform, AMO demonstrates substantial improvements in motion stability and workspace coverage, enabling end-to-end autonomous task execution. Quantitatively, it surpasses state-of-the-art baselines across key metrics including success rate, tracking accuracy, and robustness to environmental perturbations. The framework bridges critical gaps between learning-based control and model-based planning, advancing practical deployment of legged locomotion and manipulation in unstructured environments.

Technology Category

• Intelligent Robots: Manipulation
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Humanoid robots derive much of their dexterity from hyper-dexterous whole-body movements, enabling tasks that require a large operational workspace: such as picking objects off the ground. However, achieving these capabilities on real humanoids remains challenging due to their high degrees of freedom (DoF) and nonlinear dynamics. We propose Adaptive Motion Optimization (AMO), a framework that integrates sim-to-real reinforcement learning (RL) with trajectory optimization for real-time, adaptive whole-body control. To mitigate distribution bias in motion imitation RL, we construct a hybrid AMO dataset and train a network capable of robust, on-demand adaptation to potentially O.O.D. commands. We validate AMO in simulation and on a 29-DoF Unitree G1 humanoid robot, demonstrating superior stability and an expanded workspace compared to strong baselines. Finally, we show that AMO's consistent performance supports autonomous task execution via imitation learning, underscoring the system's versatility and robustness.

Problem

• Achieving real-time adaptive whole-body control for high-DoF humanoid robots
• Mitigating distribution bias in motion imitation reinforcement learning
• Expanding workspace and stability in hyper-dexterous humanoid movements

Innovation

• Integrates sim-to-real RL with trajectory optimization
• Uses hybrid dataset to mitigate distribution bias
• Validated on 29-DoF humanoid for superior stability