This is a short research project conducted in cooperation with the MiRA Lab at National Taiwan University. The goal was to create a simulation for their robotic arm in pybullet and train it to grasp with reinforcement learning. This repository contains the urdf files for the robotic grasper, it's controller and two Gym environments to which Reinforcement Learning algrotihms can be applied. Furthermore it contains the code to train the grasper with RL algorithms such as DQN and DDGP, as well as the evaluation and testing.

The Simulation	The Real Robot

The code for the controller and the Gym environments is based on the example code provided by Bullet3 for the Kuka Arm (Links can be found in the comments).

In order to reproduce the code, please install the following packages, ideally in a separate virtual environment.

1. Install Pybullet via

pip install pybullet

git clone https://github.com/huetufemchopf/roboticgrasper.git

python baselines/train_tm700_cam_grasping.py --gym_env 'rgbd' --algorithm 'DQN' --discrete True --lr 0.0001

python baselines/replay_grasping_possensor.py  --gym_env 'rgbd'