A car that with a deep learning network that learns to drive purely from vision from a simulated camera.
This started as a practice project while reading the spinningup.openai.com guide on RL from Joshua Achiam @ OpenAI, and then became a hobby/weekends experimentation. Helped me to acquire tons of intuition about RL. Learning end-to-end through making the car more intelligent.
Install the PIP dependencies from requirements.txt.
Use Intellij Idea run configurations to run in modes:
- Debug (with rendering, normal environment speed);
- Research (with rendering, fast-forward environment speed);
- Train (no rendering, fast-forward environment speed);
- MPS (Apples Metal Performance Shaders);
Currently, the car can drive decently around turns and even make it through crossroads (based on the yellow turn signal from the drive), although stability can be improved.
Reach out to me: dmitrii.maliuga@gmail.com or linkedin/dmitrii-maliuga/.