🚗 Self-Driving Car using Reinforcement Learning

This project focuses on training an autonomous driving agent using Reinforcement Learning in the CarRacing-v3 environment provided by Gymnasium. The agent learns to drive a car using visual inputs and the Proximal Policy Optimization (PPO) algorithm.

1. Overall Strategy

• Goal: Train a reinforcement learning agent to drive autonomously on randomly generated tracks.
• Input: Visual observations (96x96 RGB frames).
• Algorithm: PPO (Proximal Policy Optimization) from stable-baselines3.
• Output: Trained model that can navigate tracks efficiently.

---

2. System Requirements

Component	Specification
OS	Linux/Windows/Mac
Python	3.10.0
Processor	Intel Core i7 (13th Gen)
RAM	Minimum 16 GB
GPU (Optional)	For faster training

---

3. Required Libraries

Install the following libraries before starting:

• gymnasium
• stable-baselines3
• box2d-py
• numpy
• matplotlib
• swig (system package, for Box2D)

Install SWIG (Fedora):

sudo dnf install swig

---

4. Environment Setup

Set up a virtual environment using either Conda or venv to manage dependencies.

📌 Using Conda (Recommended)

conda create -n selfdrivingcar python=3.10.0
conda activate selfdrivingcar
pip install -r requirements.txt

📌 Using venv

python3 -m venv selfdrivingcar
source selfdrivingcar/bin/activate
pip install -r requirements.txt

---

5. Model Development Methodology

• Data: Agent uses pixel data from CarRacing-v3.
• Action Space: Continuous or discrete actions (steer, gas, brake).
• Training Script: train.ipynb
• Evaluation Script: evaluate.py

---

6. Usage Instructions

To train the model:

python train.ipynb

To evaluate a pre-trained model:

python evaluate.py

✅ Make sure you configure train.ipynb and evaluate.py according to your model paths and environment settings.

---