Reliably Re-Acting to Partner's Actions with the Social Intrinsic Motivation of Transfer Empowerment
This addresses the issue of unreliable coordination in multi-agent systems for researchers in MARL, though it appears incremental as it builds on existing intrinsic motivation concepts.
The paper tackles the problem of multi-agent reinforcement learning agents overfitting to their training partners' policies, which leads to brittle strategies that fail to react to others' actions. The result shows that their method, transfer empowerment, improves MARL performance in three simulated cooperation scenarios.
We consider multi-agent reinforcement learning (MARL) for cooperative communication and coordination tasks. MARL agents can be brittle because they can overfit their training partners' policies. This overfitting can produce agents that adopt policies that act under the expectation that other agents will act in a certain way rather than react to their actions. Our objective is to bias the learning process towards finding reactive strategies towards other agents' behaviors. Our method, transfer empowerment, measures the potential influence between agents' actions. Results from three simulated cooperation scenarios support our hypothesis that transfer empowerment improves MARL performance. We discuss how transfer empowerment could be a useful principle to guide multi-agent coordination by ensuring reactiveness to one's partner.