cAr-drIve - reinforcement learning for simulated self-driving car
This is a project that allows to train an AI able to drive alongside any track built from pre-defined pieces.
This is the code for an accompanying Medium article, for details on the content, please check the article.
This project needs the ML-Agents environment version 0.15.0 to be set up. At best I'd suggest to follow the official installation guide. However the following steps should suffice:
- Get Python 3.6 64-bit (the 64 bit seems necessary, at least on Windows)
- Get Unity 2020.1+
- Download the ML-Agents 1.0.0+ package (see https://github.com/Unity-Technologies/ml-agents/blob/release_12_docs/docs/Installation.md)
- Install dependencies:
pip3 install "mlagents"
- Clone this repository
- Open project in Unity
- CarTrain This scene is used for training a model.
- CarTest This scene contains a different track for verification of the trained model.
- CarPlay This scene contains a set of 3 Agents and a car controlled by the player to mimic a game scenario.