🚗 Self-Driving Car Project 🚗

Table of Contents

1. Introduction
2. Prerequisites
3. Data Preparation
   • Manual Data Collection
   • Download Pre-recorded Data
   • Data Preprocessing
4. Model Architecture
5. Usage

Introduction

This deep learning model uses Convolutional Neural Networks (CNN) to predict steering angles and speed for the self-driving car simulation: Udacity Self-Driving Car Simulator. The model takes input from the car's 3 cameras (left, center, and right) and makes predictions for the car's movement.

• Car Movement: The car can move left (←), right (→), accelerate (↑), and decelerate (↓).
• Camera Setup: The car is equipped with three cameras (left, center, right).

Prerequisites

• Operating System: Linux or Windows (Linux preferred)
• Python: 3.12.7
• Simulation App: Udacity Self-Driving Car Simulator

Data Preparation

Manual Data Collection:

1. Open the simulator and select the Training Mode.

2. Click on Record, choose a folder to save the data, and drive the car for about 10 minutes.
   • Driving for 10 minutes will yield around 18,000 images (6,000 images from each camera).

Download Pre-recorded and Preprocessed Data:

• Download here.

After data collection, you will have a driving_log.csv file, which contains information about the collected data:

Image Path	Center Camera Image	Left Camera Image	Right Camera Image	Steering Angle	Throttle	Brake

Data Preprocessing

For Manual Data Collection Only.

Follow Data Preprocessing

Model Architecture

Usage

Setup Environment

• Create a virtual environment:

python -m venv .venv
source .venv/bin/activate  # On Windows use .venv\Scripts\activate

• Install required packages from requirements.txt:

pip install -r requirements.txt

Training Model

• Set data_dir in training.py to folder holding both preprocessed driving_log.csv and IMG. For example:

data_dir = 'data/processed/track1'

• Run training.py:

python training.py

Autonomous Run

Once the model is trained, you can use it to predict the car's movements based on the camera input in the simulator.

• First, launch the Simulator App and select Autonomous mode; at this point, the car will remain stationary.

• Next, run the following command to connect to the Simulator (without recording):

python main.py <model path>

Or (with recording enabled):

python main.py <model path> <image folder>

Image to Video Conversion

• To convert the captured images into a video, run the following command:

python video.py <image folder> [--fps]

The --fps argument defaults to 60 if not specified.