gym_multi_env

gym_multi_env is a python module allowing you to join multiple OpenAI Gym environments.

Installation

$ git clone https://github.com/martinseilair/gym_multi_env/
$ cd gym_multi_env
$ pip install .

Tested with Python 3.5.2 and Ubuntu 16.04.

Quick start

import gym
import gym_multi_env as menv

domain_names = ["CartPole-v0", "Pendulum-v0"]  # list of environments
gym_id = menv.create(domain_names, join_spaces=True)  # create new environment

env = gym.make(gym_id)

for _ in range(1000):
    rgb = env.render(mode="rgb_array")
    obs, reward, done, _ = env.step(env.action_space.sample())  # take a random action

Documentation

Spaces

If join_spaces is set to False, observation and action spaces of the chosen environments are joined in a tuple. If join_spaces is set to True and the types of the spaces are compatible, the spaces will be joined into a space of the same type (with corresponding dimensionality). Otherwise they will be joind in a tuple. To check resulting action and observation space you can use the function env.info().

Reward

As a default the reward of the multi environment is the sum of the individual environments. If you want a particular weighting of the rewards, you can pass a weighting vector with argument reward_mix. Use the function env.rewards() to retrieve all individual rewards.

Termination

Environment terminates, if at least one individual environment terminates.

Rendering

Every environment is rendered and shown separately. In mode="rgb_array" a tuple of the RGB images is put out.