Agent Priors-guided Policy Learning

2026-09-28

AI Summary

This study addresses the information loss inherent in robotic skill composition and generalization, which constrains performance in few-shot scenarios. To overcome this limitation, the work proposes leveraging policy structural priors as a language interface to bridge skill acquisition and composition. During training, an agent automatically segments demonstration data to learn multiple prior-conditioned policies. At inference, the agent dynamically selects and composes these policies via the proposed interface, enabling zero-shot recombination of unseen skills. Extensive evaluations on the MetaWorld and ManiSkill benchmarks demonstrate that this approach significantly enhances out-of-distribution generalization performance while effectively supporting novel skill composition. Furthermore, ablation studies validate the critical role of the prior-based interface in achieving these improvements.

Abstract

Robots that learn from a few demonstrations often require two forms of generalization. Compositional generalization recombines skills to solve new tasks, and skill generalization lets the learned policy behind each skill work in new situations. The two depend on each other, yet information is lost between composition and the skills it calls. Where a skill works is determined by the structure its policy is trained with, while composition sees the skill only through a separate description, such as a name, an instruction, or a symbolic operator, that omits this structure. Our key idea is to use each policy's structural prior as part of the interface between composition and the skill. A structural prior states what a behavior depends on, for example that a grasp depends only on the gripper's pose relative to the object. Built into training, it shapes where the policy generalizes; stated in language, it tells composition where the policy applies. We instantiate this idea in Agent Priors-guided Policy Learning (APPL). A construction agent segments complete demonstrations into reusable skills, proposes several structural priors for each skill, and trains and verifies one policy per prior. A runtime agent then selects among these prior-specific policies and composes them toward new task goals using their interfaces. Across MetaWorld and long-horizon ManiSkill tasks, APPL improves out-of-distribution skill generalization and enables previously unseen skill compositions; ablating the interface information substantially reduces performance. These results support the use of training-time structural assumptions as a bridge between skill learning and skill composition.

Problem

• compositional generalization
• skill generalization
• structural prior
• policy learning
• few-shot demonstration

Innovation

• Structural Priors
• Compositional Generalization
• Skill Generalization
• Policy Learning
• Agent Priors