Learning Robust State Abstractions for Hidden-Parameter Block MDPs

TL;DR

This paper introduces a robust state abstraction framework for Hidden-Parameter MDPs that enhances transfer and generalization in multi-task and meta-reinforcement learning, with theoretical bounds and empirical improvements.

Contribution

It extends HiP-MDPs to Block MDP-inspired abstractions, providing new theoretical bounds and demonstrating empirical performance gains in multi-task and meta-RL.

Findings

Improved transfer and generalization bounds based on task and state similarity.

Sample complexity depends on total samples across tasks, not number of tasks.

Empirical results show superior performance over existing baselines.

Abstract

Many control tasks exhibit similar dynamics that can be modeled as having common latent structure. Hidden-Parameter Markov Decision Processes (HiP-MDPs) explicitly model this structure to improve sample efficiency in multi-task settings. However, this setting makes strong assumptions on the observability of the state that limit its application in real-world scenarios with rich observation spaces. In this work, we leverage ideas of common structure from the HiP-MDP setting, and extend it to enable robust state abstractions inspired by Block MDPs. We derive instantiations of this new framework for both multi-task reinforcement learning (MTRL) and meta-reinforcement learning (Meta-RL) settings. Further, we provide transfer and generalization bounds based on task and state similarity, along with sample complexity bounds that depend on the aggregate number of samples across tasks, rather than the number of tasks, a significant improvement over prior work that use the same environment assumptions. To further demonstrate the efficacy of the proposed method, we empirically compare and show improvement over multi-task and meta-reinforcement learning baselines.