Custom OpenAI Gym environments

Some custom environments for OpenAI gym.

Custom environments

cartpole_swingup.py is a modification of the original cartpole.py. Now the task is to swing-up the pole and optionally also balance after it swinging up.
double_pendulum.pyis a modification of the original acrobot.py. Now actuation is applied to the fixed end joint rather than the center joint as it is in acrobot. The environment supports multiple tasks such as swing-up, balance, and swing-up + balance.

Registered environments:

CartPoleSwingUp-v0 has been created which is a modification of the original CartPole-v1. This custom version includes having to swing the pole to the inverted state before balancing.

For more information on creating custom environments, see How to create new environments for Gym.

acrobot_wrapper.py: wraps the original acrobot environment to support new tasks such as balancing and swing-up + balance. acrobot alone only supports the swing-up task.

Example usage of wrapper:

env = gym.make('Acrobot-v1')
env = AcrobotWrapper(env, task="balance", max_steps=500)

Install package:

pip install -e gym_custom

The -e option is for 'editable mode' which allows the code to be updated after the package has been installed.

Tested and working with gym 0.18.0

By installing this package, the custom environments get registered (see gym_custom/envs/__init__.py) in gym and can be used by:

import gym
env = gym.make('gym_custom:CartPoleSwingUp-v0')