Robot Learning Gym is a tool to test reinforcement learning algorithms for robotics tasks a la OpenAI's Universe and/or Gym.
The purpose is to evaluate how well various RL algorithms work on robotics (motor/manipulation) tasks by using standardized and consistent metrics. To do this, RLG supports the ability to easily add robots and tasks in order to evaluate performance on a large repertoire of tasks. This makes it possible to quickly and easily test which conditions algorithms work best and worse under.
To learn more, visit the wiki.
CAUTION: RLG is still in development, documentation may be incorrect, and code may be buggy
Dependencies:
- MuJoCo
- TensorFlow 1.0 (previous versions won't work)
- OpenAI Gym
- Others (will update soon)