Reinforcement Learning Racer

Evolution of an Autonomous Agent

Project Overview

This project demonstrates the development of a Reinforcement Learning agent trained to master 2D racing lines. Built using Python and Stable Baselines 3 (PPO), it documents the AI's evolution from basic obstacle avoidance on procedurally generated tracks to precision time-attack racing.

By upgrading the neural network inputs from simple Raycasts to include relative localization data (Checkpoint Angle/Distance), the agent optimized its lap times by over 9%, breaking the 10-second barrier. The project culminates in a custom "Draw & Drive" interface, allowing real-time testing on user-generated splines.

Project Showcase

Tech Stack

Key Features