Learning to Coordinate without Communication under Incomplete Information

Abstract: Achieving seamless coordination in cooperative games is a crucial challenge in artificial intelligence, particularly when players operate under incomplete information. A common strategy to mitigate this information asymmetry involves leveraging explicit communication. However, direct (verbal) communication is not always feasible due to factors such as transmission loss. Leveraging the game Gnomes at Night, we explore how effective coordination can be achieved without verbal communication, relying solely on observing each other's actions. We demonstrate how an autonomous agent can learn to cooperate by interpreting its partner's sequences of actions, which are used to hint at its intents. Our approach generates a non-Markovian strategy for the agent by learning a deterministic finite automaton for each possible action and integrating these automata into a finite-state transducer. Experimental results in a Gnomes at Night testbed show that, even without direct communication, one can learn effective cooperation strategies. Such strategies achieve significantly higher success rates and require fewer steps to complete the game compared to uncoordinated ones, and perform almost as well as in the case direct communication is allowed.