This is the modified mj_envs, which supports resetting to a given state, used in "Cross-modal Domain Adaptation for Cost-efficient Reinforcement Learning" based on the original repo.
mj_envs contains a variety of environements, which are organized as modules. Each module is a collection of loosely related environements. Following modules are provided at the moment with plans to improve the diversity of the collection.
HMS contains a collection of environements centered around dexterous manipulation with anthroporphic 24 degrees of freedom Adroit Hand. These environments were designed for the publication: Learning Complex Dexterous Manipulation with Deep Reinforcement Learning and Demonstrations, RSS2018.
| Hand-Manipulation-Suite Tasks (video) |
|---|
 |
mj_envs uses git submodules to resolve dependencies. Please follow steps exactly as below to install correctly.
- Clone this repo with pre-populated submodule dependencies
$ git clone --recursive https://github.com/jiangsy/mj_envs.git
- Update submodules
$ cd mj_envs
$ git submodule update --remote
- Install package using
pip
$ pip install -e .
NOTE: If the visualization results in a GLFW error, this is because mujoco-py does not see some graphics drivers correctly. This can usually be fixed by explicitly loading the correct drivers before running the python script. See this page for details.