BeamNG.tech_RL_Fun

Reinforcement learning experiments by Mikey Dempsey, age 9, working toward training a self-driving car in BeamNG.tech.

This repo starts with a 2D practice track (Gymnasium's CarRacing-v3) so Mikey can learn the basics of training and watching an RL agent before moving to the full BeamNG simulator.

Files

• practice_train.py — trains a PPO agent with a CNN policy on CarRacing-v3 for 100,000 steps and saves the result as practice_racer.zip.
• practice_watch.py — loads the saved model and opens a window so you can watch the AI drive.

Setup

python -m venv venv
venv\Scripts\activate
pip install gymnasium[box2d] stable-baselines3[extra] tensorboard

Train

python practice_train.py

In a second terminal, watch it learn:

tensorboard --logdir ./logs/

Then open http://localhost:6006.

Watch

python practice_watch.py

License

MIT — see LICENSE.