AI-robot-control

This repository contains code for industrial robot learning enabled with reinforcement learning algorithm that run in selected environments. The repository contains the following:

• algorithms : implementation of RL algorithms from OpenAI baselines and stable-baselines
• environments : robot environments
• evaluations : results of training
• experiments : scripts for running the training and for replaying the models
• models : saved models
• python_pkgs : python packages used
• simulation : robot definition, including pickbot

Prerequisites

• python 2.7
• python 3.5
• ROS

If ROS has not installed yet, go to installation for information on how to install ROS and catkin workspace then continue below

Get the code of the AI-Robot-Control packages

! Assuming your catkin workspace is named catkin_ws

cd ~/catkin_ws/src
git clone --recurse-submodules https://github.com/PhilipKurrek/AI-robot-control.git
cd AI-robot-control
git submodule update --init --recursive
sudo python3 -m pip install -e .
chmod +x install.sh && . install.sh

If pip is not installed, run

curl https://bootstrap.pypa.io/get-pip.py -o get-pip.py
python get-pip.py && python3 get-pip.py

How to run the Training:

Start the simulation

Currently we have three environments:

• simplified version of PickBot task

roslaunch simulation main.launch

• PickBot task in handling door handle

roslaunch simulation use_case-1.launch

• Generic environment for robot training

roslaunch simulation sim_research.launch

To run the Baseline DeepQ Algorithm

cd ~/catkin_ws/src/AI-Robot-Control/experiments/reach/DeepQ
python3 train_pickbot-v0_DeepQ_discrete.py

or

roslaunch experiments train_reach_deepq.launch

Tips: Add a new launch file in ROS package 'experiments' for a shorter way to execute the training script

To reuse the saved model in order to replay the learned policy

cd ~/catkin_ws/src/AI-Robot-Control/experiments/reach/DeepQ
python3 enjoy_pickbot.py

To use (in-house) algorithm implementation of DeepQ-Learning and Q-Learning

• Q-Learning

roslaunch robot_training start_training_Q_Learning.launch

• Deep Q Learning

roslaunch robot_training start_training_Deep_Q_Learning.launch

MoveIt: run simulation using MoveIt as controller

1. Start Simulation:

roslaunch ur_gazebo pickbot.launch

2. Start moveit controller:

$ cd ~/catkin_ws/src/AI-robot-control/simulation/motion_control
$ python pickbot_motion_control.py

3. Run trainning script (choose the training script you want)