| Definition | A variety of reinforcement learning environments, including robotics simulations, soccer, navigation puzzles, and classic ATARI games. |
| No of environments | 76 |
Shimmy is an API conversion tool that provides bindings for various reinforcement learning environments within the Gymnasium and PettingZoo frameworks. It features popular external environments such as DM Control, which offers a physics-based robotics simulation, and DM Soccer, a multi-agent cooperative soccer game. Additionally, it includes DM Lab for 3D navigation and puzzle-solving tasks, as well as the Behavior Suite, which serves as a test suite for evaluating model behaviors.
Moreover, Shimmy supports the Atari Learning Environment, featuring over 50 classic Atari 2600 games, and Melting Pot, which focuses on multi-agent social reasoning games. It also includes OpenSpiel, a collection of more than 70 board and card game environments. This diverse array of supported environments makes Shimmy a valuable tool for those developing and testing reinforcement learning models.
| Example observations | |
| Example actions | |
| Example reward |