This project is a pipeline that connects a Matlab simulation (Simulink) to an OpenAI Gym wrapper for PyTorch Reinforcement Learning using DQN algorithm (and various ML/DL algorithms eventually).

Matlab simulations offer an excellent way to model real-world scenarios. This project aims to establish a connection between Matlab simulations and third-party libraries and open-source AI solutions. As such, we have chosen to develop an interface that bridges a Matlab simulation with Python AI libraries

The simulation that we used to test the interface is based on a Robot (Thymio) navigating through a maze, the robot has multiple sensors to measure the distance to nearby walls with the goal of learning how to efficiently navigate throught the maze and exit it safely.

What things you need to install the software and how to install them:

• Matlab
• Simulink
• OpenAI Gym
• PyTorch

1. Clone the repo : git clone --recursive https://github.com/Spinkoo/Matlab2TorchRL (to get the Matlab2Py submodule)
2. Navigate to the project directory
3. Install requirements.txt pip install -r requirements.txt
4. Install Matlab engine for Python

• Matlab based on the amazing implementation of The Pledge Algorithm - Thymio Robot Model on a Maze
• OpenAI Gym - The reinforcement learning wrapper
• TorchRL and DQN - The deep learning library used, various algorithms can be implemented easily other than DQN that we used

Initial work - Spinkoo

see the LICENSE.md file for details

• This work was supported by the French National Research Agency under the France 2030 program project IRT Nanoelec (ANR-10-AIRT-05).