This is a fork of OpenAI gym that Robot Locomotion Group uses for custom environments. These environments will include:

• Gym environments wrapped around Drake
• Modifications to existing gym environments where observations are directly pixels.
• Custom environments written in pymunk.

Some of these environments will be exported to a pip package once they are mature.

• Carrot pushing environment. gym.make("Carrot-v0")
• Pendulum environment with pixel output. gym.make("PendulumPixel-v0")
• Planar single integrator with pixel output. gym.make("SingleIntegrator-v0")

This fork can be setup in the following way.

git clone git@github.com:RobotLocomotion/gym.git
cd gym
pip install -e .

If you are working on this repo then add the following lines:

cd gym 
git remote set-url origin git@github.com:RobotLocomotion/gym.git
git remote add upstream git@github.com:openai/gym.git
git remote set-url --push upstream no_push

You can create your own environment on a local branch or your own fork, then PR to this repo once the environment is good enough. Make sure you dont't PR to openai/gym, but to RobotLocomotion/gym!

See OpenAI's instructions on creating a new environment.

To isolate our environments from OpenAI's original environments, let's keep the group environments in gym/envs/robot_locomotion_group. Generally, the procedure will involve the following:

1. Create a folder under envs/robot_locomotion_group with the new environment name, and develop your environment there that contains your own gym.Env class.
2. Register your new environment class under envs/robot_locomotion_group/__init__.py.
3. Register your new environment name under envs/__init__.py. You will see a section dedicated to Robot Locomotion Grouop.
4. Now you can use your environment from any location with gym.make("my_awesome_new_environment-v0")
5. [IMPORTANT] To test your environment, you can deploy a random agent by python3 examples/agents/random_agent.py ENV_NAME

• pyglet
• pymunk
• drake