Off-Policy Correction For Multi-Agent Reinforcement Learning

TL;DR

This paper introduces MA-Trace, a scalable off-policy actor-critic algorithm for multi-agent reinforcement learning that extends V-Trace, with proven convergence and strong empirical performance on the StarCraft Multi-Agent Challenge.

Contribution

We propose MA-Trace, a novel scalable on-policy actor-critic algorithm for MARL that incorporates importance sampling for off-policy correction and provides theoretical convergence guarantees.

Findings

Achieves high performance on StarCraft Multi-Agent Challenge

Exceeds state-of-the-art results on some tasks

Demonstrates scalability in multi-worker settings

Abstract

Multi-agent reinforcement learning (MARL) provides a framework for problems involving multiple interacting agents. Despite apparent similarity to the single-agent case, multi-agent problems are often harder to train and analyze theoretically. In this work, we propose MA-Trace, a new on-policy actor-critic algorithm, which extends V-Trace to the MARL setting. The key advantage of our algorithm is its high scalability in a multi-worker setting. To this end, MA-Trace utilizes importance sampling as an off-policy correction method, which allows distributing the computations with no impact on the quality of training. Furthermore, our algorithm is theoretically grounded - we prove a fixed-point theorem that guarantees convergence. We evaluate the algorithm extensively on the StarCraft Multi-Agent Challenge, a standard benchmark for multi-agent algorithms. MA-Trace achieves high performance on all its tasks and exceeds state-of-the-art results on some of them.