Auto Race AI

[🚧 Project may receive updates in the future (that are listed below) for better experience🚧]

Welcome to the Auto Race AI project repository! This project provides an interactive simulation of machine learning model-trained cars using a neural network. Written in TypeScript and devoid of external libraries, this project is designed to offer a deep exploration into the inner workings of neural networks and genetic algorithms.

Introduction

The Auto Race AI project provides an immersive experience into the world of machine learning and artificial intelligence through an engaging car racing simulation. By adopting TypeScript and without usage of external libraries, I aim to facilitate a comprehensive understanding of the intricate mechanisms behind neural networks and genetic algorithms.

Neural Network Architecture

At the core of this project lies a sophisticated neural network architecture. Each car is equipped with a three-layered feed-forward neural network. The sensor.ts file allows you to effortlessly adjust the number of rays emitted by each car.

Genetic Algorithm

Genetic algorithm drives the training process, enabling cars to continually enhance their performance across generations. This mechanism incorporates single-point crossover, mutation, and selection strategies, mirroring nature's process of evolution.

Fitness Calculation

A holistic approach to fitness evaluation underpins the progress of each car. The fitness score is calculated based on four essential factors: the number of crossed obstacles, distance from the center line, car speed, and total distance traveled.

Simulation Details

A simulation process unfolds with each generation. A group of 250 cars, each equipped with its neural network, is introduced per generation. Progress to the subsequent generation is initiated upon the achievement of a predefined threshold (e.g., -20000) by the best-performing car. The racetrack layout, including 50 randomly positioned cars, is defined in the data/traffic.ts file.

Adjustable Parameters

Tailor the project according to your preferences by adjusting these parameters:

• Modify the number of rays emitted by each car in the sensor.ts file.
• Customize the track layout and obstacle positions in the data/traffic.ts file.
• Define the advancement threshold for generation progression (e.g., -20000).

Future Development Plans

In the future i plan to add on these new features to expand on the ones i already have for better user experience.

• Diverse Training Environments

• Playable Mode: Enable users to test their skills by racing against the trained AI model.

• Advanced Obstacles: Incorporate curved paths and stationary obstacles to simulate real-world driving challenges

Usage

To dive into the Auto Race AI project, follow these steps:

Execute the following commands to install dependencies and run the simulation:

npm install
parcel ./index.html
tsc -w

here is a video of a nicely trained cars:

AI_DRIVING_CAR_SHOWCASE.mp4