Game Hanabi

Hanabi, a cooperative card game requiring players to infer information from limited cues, serves as a compelling benchmark for research in multi-agent reinforcement learning (MARL). Current research focuses on improving zero-shot coordination (ZSC), where agents must cooperate effectively without prior interaction, and addressing the challenge of ad-hoc teamplay, where agents must adapt to unseen partners. This involves exploring various MARL algorithms, including variations of policy gradient methods and models incorporating Theory of Mind and abductive reasoning, to enhance both ZSC performance and adaptability. Findings in this area contribute to a broader understanding of effective collaboration in complex, uncertain environments and have implications for the design of robust and human-compatible AI agents.