Benchmarking Multi-Agent Deep Reinforcement Learning Algorithms

Chao Yu, Akash Velu, Eugene Vinitsky, Yu Wang, Alexandre Bayen, Yi Wu

Keywords: Multi-agent Reinforcement Learning, Benchmarking

Abstract: We benchmark commonly used multi-agent deep reinforcement learning (MARL) algorithms on a variety of cooperative multi-agent games. While there has been significant innovation in MARL algorithms, algorithms tend to be tested and tuned on a single domain and their average performance across multiple domains is less characterized. Furthermore, since the hyperparameters of the algorithms are carefully tuned to the task of interest, it is unclear whether hyperparameters can easily be found that allow the algorithm to be repurposed for other cooperative tasks with different reward structure and environment dynamics. To investigate the consistency of the performance of MARL algorithms, we build an open-source library of multi-agent algorithms including DDPG/TD3/SAC with centralized Q functions, PPO with centralized value functions, and QMix and test them across a range of tasks that vary in coordination difficulty and agent number. The domains include the particle-world environments, starcraft micromanagement challenges, the Hanabi challenge, and the hide-and-seek environments. Finally, we investigate the ease of hyper-parameter tuning for each of the algorithms by tuning hyper-parameters in one environment per domain and re-using them in the other environments within the domain.

One-sentence Summary: We provide an analysis of the relative performance of multi-agent reinforcement learning algorithms as well as an analysis of how easy they are to tune for new tasks.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Reviewed Version (pdf): https://openreview.net/references/pdf?id=E6CH2AvxhT