Counterfactual multi−agent policy gradients

Conference item

Many real-world problems, such as network packet routing and the coordination of autonomous vehicles, are naturally modelled as cooperative multi-agent systems. There is a great need for new reinforcement learning methods that can ef- ficiently learn decentralised policies for such systems. To this end, we propose a new multi-agent actor-critic method called counterfactual multi-agent (COMA) policy gradients. COMA uses a centralised critic to estimate the Q-function and decentralised actors to optimise the agents’ policies. In addition, to address the challenges of multi-agent credit assignment, it uses a counterfactual baseline that marginalises out a single agent’s action, while keeping the other agents’ actions fixed. COMA also uses a critic representation that allows the counterfactual baseline to be computed efficiently in a single forward pass. We evaluate COMA in the testbed of StarCraft unit micromanagement, using a decentralised variant with significant partial observability. COMA significantly improves average performance over other multi-agent actorcritic methods in this setting, and the best performing agents are competitive with state-of-the-art centralised controllers that get access to the full state.

Authors: Foerster, J; Farquhar, G; Afouras, T; Nardelli, N; Whiteson, S

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: AAAI Press
• Host title: 32nd AAAI Conference on Artificial Intelligence (AAAI'18)
• Journal: 32nd AAAI Conference on Artificial Intelligence (AAAI'18)
• Pages: 2974-2982
• Publication date: 2018-04-29
• Acceptance date: 2017-11-09
• ISSN: 2159-5399
• Keywords: actorcritic; deep reinforcement learning; multi-agent learning