Action-Sufficient State Representation Learning for Control with Structural Constraints

TL;DR

This paper introduces Action-Sufficient State Representations (ASRs) for partially observable environments, leveraging structural constraints and variational auto-encoders to improve decision-making efficiency and generalization.

Contribution

It proposes a novel framework for learning minimal, sufficient state representations using structural models and variational auto-encoders, enhancing policy learning in complex environments.

Findings

ASRs improve policy learning in CarRacing and VizDoom.

Estimated models and ASRs enable learning from imagined outcomes.

ASRs enhance sample efficiency in reinforcement learning.

Abstract

Perceived signals in real-world scenarios are usually high-dimensional and noisy, and finding and using their representation that contains essential and sufficient information required by downstream decision-making tasks will help improve computational efficiency and generalization ability in the tasks. In this paper, we focus on partially observable environments and propose to learn a minimal set of state representations that capture sufficient information for decision-making, termed \textit{Action-Sufficient state Representations} (ASRs). We build a generative environment model for the structural relationships among variables in the system and present a principled way to characterize ASRs based on structural constraints and the goal of maximizing cumulative reward in policy learning. We then develop a structured sequential Variational Auto-Encoder to estimate the environment model and extract ASRs. Our empirical results on CarRacing and VizDoom demonstrate a clear advantage of learning and using ASRs for policy learning. Moreover, the estimated environment model and ASRs allow learning behaviors from imagined outcomes in the compact latent space to improve sample efficiency.