bullet_envs

[Watch a demo]

This library extends some original PyBullet environments provided in pybullet_envs.

PyBullet environments are similar to those provided by MuJoCo which are fully compatible with OpenAI Gym, with the difference of being open-source.

Only the following environments are fully supported:

• TurtlebotMazeEnv-v0
• ReacherBulletEnv-v0
• HalfCheetahBulletEnv-v0
• InvertedPendulumSwingupBulletEnv-v0

Contributions

• All environments have a camera rendering in def render which is wrapped into an OpenAI gym wrapper.

• TurtlebotMazeEnv-v0 is proposed here as a new environment, built from the original Turtebot implemented in pybullet_robots. It includes a version where one of the walls has a randomly sampled color at each time step. The observation space corresponds to a first-person perspective camera.

• ReacherBulletEnv-v0 has a new version with a randomly moving ball as a distractor which corresponds to the file reacher_distractor.xml

Installation

Clone this repo and have its path added to your PYTHONPATH environment variable:

cd <installation_path_of_your_choice>
git clone https://github.com/astrid-merckling/bullet_envs.git
cd bullet_envs
export PYTHONPATH=$(pwd):${PYTHONPATH}

You can install the dependencies as:

pip install gym==0.17.2
pip install pybullet==2.6.4
pip install opencv-python==4.1.2.30

Usage

Example to run and visualize ReacherBulletEnv-v0 with a randomly moving ball (distractor=True), where the observation space is chosen to be the camera:

import gym

"register bullet_envs in gym"
import bullet_envs.__init__

env_name = 'ReacherBulletEnv-v0'
actionRepeat = 1
maxSteps = 50
"OpenAI Gym env creation"
env = gym.make('PybulletEnv-v0', env_name=env_name, renders=True, distractor=True, actionRepeat=actionRepeat,
               maxSteps=maxSteps * actionRepeat, image_size=64, display_target=True)

"running env on 5 episodes"
num_ep = 5
for episode in range(num_ep):
    obs = env.reset()
    done = False
    while not done:
        # follow a random policy
        action = env.action_space.sample()
        obs, reward, done, info = env.step(action)
        "get the image observation from the camera"
        obs = env.render()

Example to run and visualize TurtlebotMazeEnv-v0 with a randomly sampled wall color (wallDistractor=True), where the observation space is chosen to be the first-person camera:

import gym

"register bullet_envs in gym"
import bullet_envs.__init__

env_name = 'TurtlebotMazeEnv-v0'
actionRepeat = 1
maxSteps = 100
"OpenAI Gym env creation"
env = gym.make(env_name, renders=True, wallDistractor=True, maxSteps=maxSteps, image_size=64, display_target=True)

"running env on 2 episodes"
num_ep = 2
for episode in range(num_ep):
    obs = env.reset()
    done = False
    while not done:
        # follow a random policy
        action = env.action_space.sample()
        obs, reward, done, info = env.step(action)
        "get the image observation from the camera"
        obs = env.render()

See OpenAI Gym for more details on the Python env class.