ACPO: A Policy Optimization Algorithm for Average MDPs with Constraints

TL;DR

This paper introduces ACPO, a novel policy optimization algorithm tailored for average constrained Markov Decision Processes, addressing the limitations of existing methods in average-CMDPs with theoretical guarantees and superior empirical results.

Contribution

The paper proposes ACPO, a new trust region-based policy optimization algorithm specifically designed for average-CMDPs, with theoretical analysis and extensive experimental validation.

Findings

ACPO outperforms existing algorithms in OpenAI Gym environments.

Theoretical guarantees are established for ACPO's performance.

ACPO effectively handles constraints in average-CMDPs.

Abstract

Reinforcement Learning (RL) for constrained MDPs (CMDPs) is an increasingly important problem for various applications. Often, the average criterion is more suitable than the discounted criterion. Yet, RL for average-CMDPs (ACMDPs) remains a challenging problem. Algorithms designed for discounted constrained RL problems often do not perform well for the average CMDP setting. In this paper, we introduce a new policy optimization with function approximation algorithm for constrained MDPs with the average criterion. The Average-Constrained Policy Optimization (ACPO) algorithm is inspired by trust region-based policy optimization algorithms. We develop basic sensitivity theory for average CMDPs, and then use the corresponding bounds in the design of the algorithm. We provide theoretical guarantees on its performance, and through extensive experimental work in various challenging OpenAI Gym environments, show its superior empirical performance when compared to other state-of-the-art algorithms adapted for the ACMDPs.