Robolearn Environments

A Python package with OpenAI-Gym-like environments for Robot Learning

This repository contains a shorter version of the Pybullet-Environment interface of Robolearn.

The code has been tested with Python 3.5 (or later).

Robolearn is a python package that defines common interfaces between robot learning algorithms and robots. More info in the following link.

Warning: This package is gradually becoming public, so the public version is still in development. Sorry for any inconvenience.

Pre-Installation

It is recommended to first create either a virtualenv or a conda environment.

• Option 1: Conda environment. First install either miniconda (recommended) or anaconda. Installation instructions

# Create the conda environment
conda create -n <condaenv_name> python=3.5
# Activate the conda environment
conda activate <condaenv_name>

• Option 2: Virtualenv. First install pip and virtualenv. Installation instructions

# Create the virtual environment
virtualenv -p python3.5 <virtualenv_name>
# Activate the virtual environment
source <virtualenv_name>/bin/activate

Installation

1. Clone this repository

git clone https://github.com/domingoesteban/hiu_sac

2. Install the requirements of this repository

cd robolearn_envs
pip install -e .

Use

Standard-OpenAI-Gym way:

import gym
import robolearn_envs

env = gym.make('RoboLearn-CogimonLocomotionRender-v0')
# A 'headless' version of the same environment would be:
# env = gym.make('RoboLearn-CogimonLocomotion-v0')

# By default, Robolearn Bullet environments starts rendering when they are reset.
obs = env.reset()

for _ in range(500):
    obs, reward, done, info = env.step(env.action_space.sample())

env.close()

Easy-to-customize way:

from robolearn_envs.pybullet import CogimonLocomotionEnv

env = CogimonLocomotionEnv(
    active_joints='WB',
    control_mode='joint_torque',
    is_render=True,
    sim_timestep=0.001,
    frame_skip=10,
    seed=1510,
    max_time=None,
)

# If is_render=True, rendering starts when the environment is reset.
obs = env.reset()

for _ in range(100):
    obs, reward, done, info = env.step(env.action_space.sample())

env.close()

Environments

Robot	Task	Gym Name
Centauro	Reaching	CentauroReachingEnv-v0
Centauro	Locomotion	CentauroLocomotionEnv-v0
Cogimon	Reaching	CogimonReachingEnv-v0
Cogimon	Locomotion	CogimonLocomotionEnv-v0
Walkman	Reaching	WalkmanReachingEnv-v0
Walkman	Locomotion	WalkmanLocomotionEnv-v0
Coman	Locomotion	ComanLocomotionEnv-v0
Hyq	Locomotion	HyqLocomotionEnv-v0

Citation

If this repository was useful for your research, we would appreciate that you can cite it:

@misc{robolearn-envs,
  author = {Esteban, Domingo},
  title = {RoboLearn Environments},
  year = {2018},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/domingoesteban/robolearn_envs}},
}

Acknowledgements

• Erwin Coumans for Bullet, PyBullet, and his didactic examples (Bullet repository)
• Enrico Mingo for Coman, Walkman, and Cogimon urdf models.
• Malgorzata Kamedula for Centauro urdf model.
• Dynamic Legged Systems Lab for HyQ urdf model.