You'll need to symlink our environments into you OpenAI Gym installation.

First, install OpenAI Gym. You may need to checkout an older version of the gym:

git checkout 52e803f36b3281a575bf98255947a5a74d3c59a0

Once installed (using pip), cd into your installation and ln the new env/ file into the OpenAI gym install directory:

cd /path/to/openaigym/gym/envs/safety;
ln -s /path/to/this/dir/env/acc.py acc.py

Also register this new environment by adding the following lines to gym/envs/__init__.py:

  register(
      id='acc-v0',
      entry_point='gym.envs.safety:ACCEnv',
      max_episode_steps=200,  # todo edit
      reward_threshold=195.0, # todo edit
  )

And add the appropriate import to gym/envs/safety/__init__.py:

from gym.envs.safety.acc import ACCEnv

Finally, add a new method to gym/spaces/Discrete.py:

  def sample_restricted(self, filter_fn):
    """ Sample from a subset of the space. """
    subset = filter(filter_fn, range(0,self.n))
    idx = prng.np_random.randint(len(subset))
    return subset[idx]

Nathan Fulton 2017. GPLv2