jaco-gym

Reinforcement learning on the Jaco Arm in Gazebo using Gym. An honours project at McGill University.

This is a WIP package that will allow the Jaco arm Gazebo simulation to be incorporated into OpenAI Gym.

Running

$ cd run
$ python gym-testing.py

Requirements

This package has a number of dependencies, and the setup is non-trivial

• gym-gazebo
  • gym
  • Gazebo 8
• ROS Kinetic
• jaco-arm-pkgs
• tensorforce
• cup_and_ball

Sources

• Extending the OpenAI Gym for robotics: a toolkit for reinforcement learning using ROS and Gazebo 1608.05742

To look at

• https://github.com/pat-coady/trpo/blob/master/notebooks/ant.ipynb
• https://github.com/wojzaremba/trpo/blob/master/main.py
• https://github.com/pat-coady/trpo/blob/master/src/train.py
• https://learningai.io/projects/2017/07/28/ai-gym-workout.html

Screenshots

Image 1	Image 2