Policy Decorator: Model-Agnostic Online Refinement for Large Policy Model
This addresses the problem of limited demonstration data for robot learning practitioners, offering an incremental improvement by enabling stable online refinement of existing models.
The paper tackles the limitations of offline-trained imitation learning models by introducing Policy Decorator, a model-agnostic residual policy that refines these models through online interactions, improving performance across eight tasks in benchmarks like ManiSkill and Adroit while preserving smooth motion.
Recent advancements in robot learning have used imitation learning with large models and extensive demonstrations to develop effective policies. However, these models are often limited by the quantity, quality, and diversity of demonstrations. This paper explores improving offline-trained imitation learning models through online interactions with the environment. We introduce Policy Decorator, which uses a model-agnostic residual policy to refine large imitation learning models during online interactions. By implementing controlled exploration strategies, Policy Decorator enables stable, sample-efficient online learning. Our evaluation spans eight tasks across two benchmarks-ManiSkill and Adroit-and involves two state-of-the-art imitation learning models (Behavior Transformer and Diffusion Policy). The results show Policy Decorator effectively improves the offline-trained policies and preserves the smooth motion of imitation learning models, avoiding the erratic behaviors of pure RL policies. See our project page (https://policydecorator.github.io) for videos.