A Policy Efficient Reduction Approach to Convex Constrained Deep Reinforcement Learning

TL;DR

This paper introduces a memory-efficient method for constrained deep reinforcement learning that reformulates the problem into a distance optimization task and employs a novel conditional gradient algorithm, achieving better performance with less memory.

Contribution

It proposes a new reformulation of constrained RL as a distance optimization problem and introduces a variant of the conditional gradient algorithm for policy efficiency.

Findings

Reduces memory costs by an order of magnitude in experiments.

Achieves comparable convergence rates to existing game-theoretic approaches.

Demonstrates improved performance in navigation tasks.

Abstract

Although well-established in general reinforcement learning (RL), value-based methods are rarely explored in constrained RL (CRL) for their incapability of finding policies that can randomize among multiple actions. To apply value-based methods to CRL, a recent groundbreaking line of game-theoretic approaches uses the mixed policy that randomizes among a set of carefully generated policies to converge to the desired constraint-satisfying policy. However, these approaches require storing a large set of policies, which is not policy efficient, and may incur prohibitive memory costs in constrained deep RL. To address this problem, we propose an alternative approach. Our approach first reformulates the CRL to an equivalent distance optimization problem. With a specially designed linear optimization oracle, we derive a meta-algorithm that solves it using any off-the-shelf RL algorithm and any conditional gradient (CG) type algorithm as subroutines. We then propose a new variant of the CG-type algorithm, which generalizes the minimum norm point (MNP) method. The proposed method matches the convergence rate of the existing game-theoretic approaches and achieves the worst-case optimal policy efficiency. The experiments on a navigation task show that our method reduces the memory costs by an order of magnitude, and meanwhile achieves better performance, demonstrating both its effectiveness and efficiency.