PreCi: Pretraining and Continual Improvement of Humanoid Locomotion via Model-Assumption-Based Regularization

2025-04-14

AI Summary

Addressing the challenge of adaptive locomotion for humanoid robots under complex terrains and high-speed commands, this paper proposes an “imitation pretraining + reinforcement fine-tuning” framework. First, behavior cloning is performed via model predictive control (MPC) for pretraining; subsequently, proximal policy optimization (PPO) is employed for fine-tuning. A key innovation is model-assumption-driven adaptive regularization (MAR), which selectively enforces policy alignment with the MPC expert only when the dynamics model’s confidence exceeds a threshold—thereby mitigating catastrophic forgetting and ensuring safe knowledge transfer. The method integrates state-dependent confidence filtering, MPC-based imitation learning, and MAR regularization. Evaluated on the Digit robot, it achieves forward running at 1.5 m/s and demonstrates robust locomotion across diverse challenging terrains—including slippery surfaces, inclined slopes, uneven ground, and sandy terrain.

Technology Category

Machine Learning: Imitation Learning & Inverse Reinforcement LearningIntelligent Robots: Behavior Learning & ControlHumans and AI: Human-Aware Planning and Behavior Prediction

Application Category

Responsible Web: Machine-in-the-loop, human agency and autonomySearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for rankingUser Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems

Abstract

Humanoid locomotion is a challenging task due to its inherent complexity and high-dimensional dynamics, as well as the need to adapt to diverse and unpredictable environments. In this work, we introduce a novel learning framework for effectively training a humanoid locomotion policy that imitates the behavior of a model-based controller while extending its capabilities to handle more complex locomotion tasks, such as more challenging terrain and higher velocity commands. Our framework consists of three key components: pre-training through imitation of the model-based controller, fine-tuning via reinforcement learning, and model-assumption-based regularization (MAR) during fine-tuning. In particular, MAR aligns the policy with actions from the model-based controller only in states where the model assumption holds to prevent catastrophic forgetting. We evaluate the proposed framework through comprehensive simulation tests and hardware experiments on a full-size humanoid robot, Digit, demonstrating a forward speed of 1.5 m/s and robust locomotion across diverse terrains, including slippery, sloped, uneven, and sandy terrains.

Problem

• Developing a framework for humanoid locomotion policy training
• Adapting to diverse and unpredictable environments effectively
• Preventing catastrophic forgetting during policy fine-tuning

Innovation

• Pretraining via model-based controller imitation
• Fine-tuning with reinforcement learning
• Model-assumption-based regularization prevents forgetting