Repository files navigation

AutoTech is a project from the INTech robotics club (Telecom SudParis and IMT Business School). Our goal is to design and build an autonomous vehicle forCoVAPSy (Course de Voitures Autonomes de Paris-Saclay) competition.

This repository contains the source code for our autonomous vehicle for the 2025 and 2026 competitions.

We chose to use a full Reinforcement Learning approach. We train an agent to drive the vehicle in a simulation of the race based directly on sensor inputs.

We train the AI model in parallel remote environments through nammed pipes connections between the main python script and the multiple webots instances.

Inside the simulations, each vehicle has access to data comming from a LiDAR and a camera.

• Simulator:Webots
• AI Training Library:Stable-Baselines3 with aPytorch backend
• AI Inference Engine:ONNX

We use uv for Python environment management. So if it's not already installed go check theofficial installation guide

Then, justuv sync to create the virtual environment and get all the dependencies.

# dependencies for AI traininguv sync --extra simu# dependencies for AI inference on the Raspberry PI 5uv sync --extra rpi

Navigate to the simulator directory.

cd src/simulateur

Run the multi-process training script.

uv run launch_multiprocess_training.py

This will launch the Webots instances and begin the SB3 PPO training loop. All the checkpoints will be in thecheckpoints directory. At every checkpoint, a compiled ONNX model will be stored asmodel.onnx.

To change the parameters of the simulation, just modify theconfig.py file.

(WIP)

For detailed information on architecture, hardware specifics and technical choices, please refer to theINTech wiki.

(Note: The Wiki is private and reserved to INTech members)

This project is distributed under the MIT License. See the Licence file for details.