Sawyer MuJoCo

A MuJoCo and Gym based Reinforcement Learning environment using Rethink Sawyer robot for various manipulation tasks.

This is a research code. A lot can change, break or never work.

Installation

Reqiurements:

• MuJoCo
• mujoco-py
• OpenAI Gym
• spinup (if you want to train a model)
• hid (if you want to control Sawyer's end-effector with SpaceNavigator)

Note: MuJoCo requires a license, follow the instructions here. Anaconda 3.6 is highly recommended before you install the packages above.

Use:

Reaching Environment:

from envs.sawyer_env import SawyerReachEnv
env = SawyerReachEnv(n_substeps=1)
env.reset()

Grasping Environment:

from envs.sawyer_env import SawyerGraspEnv
env = SawyerGraspEnv(n_substeps=1)
env.reset()

TODO:

SawyerGraspEnv:

• Add camera
• Fix grasp reward function
• Fix friction and penetration issues

SawyerReachEnv: