F1Racers

🏎️ Car Racing Agent using Reinforcement Learning

An autonomous car racing agent trained using reinforcement learning algorithms (DQN, PPO, SAC) in the Gymnasium CarRacing-v3 environment. The project evaluates model performance based on rewards, convergence, and control efficiency for autonomous driving.

---

📁 Files Included

• CarRacingDQN.ipynb – DQN implementation with discretized action space
• CarRacingPPO.ipynb – PPO implementation using continuous actions
• CarRacingSAC.ipynb – SAC implementation for exploration-based control
• Project Report.pdf – Full project report

---

📌 Project Overview

This project explores reinforcement learning for autonomous vehicle control in a simulated environment. Three different RL algorithms—DQN, PPO, and SAC—are evaluated for their ability to drive a car around a racetrack using Gymnasium's CarRacing-v3.

---

🧠 Algorithms Compared

Algorithm	Action Space	Type	Performance Summary
DQN	Discrete	Off-policy	✅ Best performance, fastest convergence
PPO	Continuous	On-policy	⚠️ Moderate performance, stable but slower
SAC	Continuous	Off-policy	❌ Did not converge in training period

---

👥 Authors

• Robayed Ashraf | robayedashraf@gmail.com
• Chong Fai Wong | jason1999689@gmail.com
• Asim Santos Poudel | asimsantospoudel@gmail.com