Off-Policy Correction For Multi-Agent Reinforcement Learning
This work addresses scalability and training efficiency issues in MARL for applications like gaming or robotics, though it is incremental as it builds on existing V-Trace methods.
The authors tackled the challenge of training multi-agent reinforcement learning (MARL) systems by proposing MA-Trace, an on-policy actor-critic algorithm that extends V-Trace to MARL, achieving high performance on the StarCraft Multi-Agent Challenge and exceeding state-of-the-art results on some tasks.
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.