overcooked_ai

Why we included this project

Overcooked-AI has become a benchmark that many coordination papers build on. It takes the popular cooking game and turns it into a structured decision problem where two agents split up tasks, decide on the fly who does what, and try to get soups out quickly. The repo ships the environment, planner logic, layout generators, and a web demo for playing against trained agents, which makes it straightforward to reproduce published results or pit a new method against prior work. One practical caveat: the bundled deep-RL training code is deprecated, so treat it as evaluation infrastructure rather than a source of ready-to-run baselines. For teams working on agents that need to coordinate roles with a human partner, this is a well-documented and heavily reused starting point.