DISCONTINUATION OF PROJECT.

This project will no longer be maintained by Intel.

This project has been identified as having known security escapes.

Intel has ceased development and contributions including, but not limited to, maintenance, bug fixes, new releases, or updates, to this project.

Intel no longer accepts patches to this project.

Accelerating Deep Reinforcement Learning Training with OpenVINO™

This code repo showcases how one could use Intel® Distribution of OpenVINO™ toolkit to accelerate Deep Reinforcement Learning (DRL) Training. Specifically, this is for DRL problems which leverage pre-trained goal classifiers for their reward function. The same idea can be applied for DRL problems which leverage pre-trained autoencoders for state-space reduction.

We created a Robotics Gym Environment using PyBullet. The goal of our DRL task is to have a robot move toward and hover over the blue object based on visual feedback. The observation of is the robot's x-y position. The action is the dx-dy control of the robot's gripper

Reproducing this Repo

This repository was validated using Python 3.8 on Ubuntu 20.04 & Mac OS Catalina 10.15.17

Installing Pre-Requisite Software

Step 1: Clone This Repository

git clone https://github.com/intel/openvino-drl-training-demo

Step 2: Install all the python packages (reccomended to this in a python virtual environment)

pip install -r requirements.txt

Training the Agent

Step 3: Run the training. Note the total time printed at the end

python sac_training.py -g [optional flag if you want to see the robot during its training]

Step 4: Run the same training but now using OpenVINO™ toolkit as the inference engine for the reward classifier network. The total time printed at the end should be lower than that of step 3

python sac_training.py -ov -g [optional flag if you want to see the robot during its training]

Step 5: Run inference.py to see the trained agent!

python inference.py -ov -g