Walker Environments

Important

So far, people can only use the framework with MuJoCo models. It should be actually quite easy to change the Mimic Environment to derive from a gym environment instead a mujoco gym environment and allow to use our framework with every gym env.

Straight Walker

2-segmented upper body and 3d hip walker

TODO: Find a shorter name for this model!

How to use your own environment?

Important

Do we need to limit ourselves to walker environments only? Actually not at all! The only thing so far that limits us to walking is the evaluation of the agents performance. When we refactor the code to allow implementing custom evaluation, we’ve created a easy to use DeepMimic Framework.

  1. Create a new class extending the MimicEnv class.

    1. Override the required methods (marked as methods to override at the end of the MimicEnv implementation.)

  2. Register the class in mujoco/gym_mimic_envs/__init__.py analogously to other environments.

  3. Import the class in mujoco/gym_mimic_envs/mujoco/__init__.py.

  4. In case some problems occure, please refer to the official documentation from OpenAI here: https://github.com/openai/gym/blob/master/docs/creating-environments.md