The Repository for the Autonomous Robots 2022 Journal article:
This contains a python toolkit for learning a grasping task with Franka Emika Panda Robot. The robot can be trained to grasp the cube, avoid obstacles and learn to manage redundancy using modern Reinforcement Learning algorithms of Proximal Policy Optimization (PPO) and Soft Actor-Critic (SAC). It is powered by MuJoCo physics engine
@article{shahid2022continuous,
title={Continuous control actions learning and adaptation for robotic manipulation through reinforcement learning},
author={Shahid, Asad Ali and Piga, Dario and Braghin, Francesco and Roveda, Loris},
journal={Autonomous Robots},
pages={1--16},
year={2022},
publisher={Springer}
}
The adapted policy can grasp the moving cube in 30 mints of retraining.
Before, the base grasping policy trained on a static cube is not able to grasp the moving cube.
To use this toolkit, it is required to first install MuJoCo 200 and then mujoco-py from Open AI. mujoco-py allows using MuJoCo from python interface. The installation requires python 3.6 or higher. It is recommended to install all the required packages under a conda virtual environment
This toolit is mainly developed based on Surreal Robotics Suite and the Reinforcement learning part is referenced from this repo