Google Research Football Environment is a novel Reinforcement Learning environment where agents aim to master the world’s most popular sport -- football. Modeled after popular football video games, it provides a physics based 3D football simulation where agents control either one or all football players on their team, learn how to pass between them, and manage to overcome their opponent’s defense in order to score goals.
New: don't miss the Tournament
Google Research Football Environment
About the Google Research Football League
The GRF League is a server that allows various agents (bots, trained ML models) to compete in the GRF environment. Once connected, each agent is assigned an opponent (based on the estimated ELO score) and both play a game for a fixed number of steps. There are currently three different tracks:
11vs11 - main challenge, with 11 players on each side
mini - a short 1 vs 1 scenario, for quick debugging
multiagent - multiagent challenge, similar to the main challenge, but with only 5 players in each team and you control all of them
The suggested way of participating in the league is to keep the agent always running in a terminal. This way, all new agents that appear can challenge it. Please also note that if the connection is dropped in the middle of the game, the environment will assume that agent actions for all remaining steps from this point are "idle".
For technical details on how to register and compete, please see the FAQ section.
The initial version of the Google Research Football League was developed by the Google Brain Team in Zurich. The core team consisted of: Anton Raichuk, Michał Zając, Piotr Stańczyk and Karol Kurach.
If you have any questions regarding this project, or you notice a problem with the server, please contact us at email@example.com.