APC-RL: Exceeding Data-Driven Behavior Priors with Adaptive Policy Composition

TL;DR

APC-RL introduces an adaptive hierarchical method that effectively utilizes multiple demonstration priors in reinforcement learning, improving learning speed and robustness even with suboptimal or misaligned data.

Contribution

It proposes a novel adaptive policy composition framework that selectively leverages multiple data-driven priors, refining or ignoring them based on their relevance to the target task.

Findings

Accelerates learning with aligned demonstrations

Remains robust under severe misalignment

Leverages suboptimal demonstrations for exploration

Abstract

Incorporating demonstration data into reinforcement learning (RL) can greatly accelerate learning, but existing approaches often assume demonstrations are optimal and fully aligned with the target task. In practice, demonstrations are frequently sparse, suboptimal, or misaligned, which can degrade performance when these demonstrations are integrated into RL. We propose Adaptive Policy Composition (APC), a hierarchical model that adaptively composes multiple data-driven Normalizing Flow (NF) priors. Instead of enforcing strict adherence to the priors, APC estimates each prior's applicability to the target task while leveraging them for exploration. Moreover, APC either refines useful priors, or sidesteps misaligned ones when necessary to optimize downstream reward. Across diverse benchmarks, APC accelerates learning when demonstrations are aligned, remains robust under severe misalignment, and leverages suboptimal demonstrations to bootstrap exploration while avoiding performance degradation caused by overly strict adherence to suboptimal demonstrations.