Welcome to gym-extensions! We’re building on top of OpenAI Gym in the pursuit of creating environments suitable for benchmarking a wide variety of auxiliary tasks. On the RoadMap right now are:
- Multitask, Transfer, and LifeLong Learning Benchmarks
- Learning to navigate and search tasks
- More realistic robotics tasks integrated with ROS and Gazebo
There’s a lot of info to go over!
- You can read about the modified Mujoco OpenAI continuous domain tasks here
- You can read about new 2D navigation continuous domain tasks here
- We also provide a set of wrappers to process agent observations in tasks as described here
- Finally, you can enjoy a Brand New SpaceX-like Falcon rocket that can be thought to land on the autonomous drone ship here
We want all the envs and pull requests we can get! Open an issue if you’re confused and we’ll fix it ASAP.
In progress documentation:
- A website for benchmarked multitask runs
- Pictures and videos
- A nicer layout
- How to create new environments and contribute
Brought to you by