genetic-car

Self-learning car using genetic algorithm

Usage

Please run manual version, do not run auto version, it has not been implemented!

1. Run program

$ python genetic_car_manual.py

2. Select elites manually

After each round, just click the blue cars who have Best performance, and their colors would turn to red, then press Space key to enter next loop

3. Repeat until cars can finish the track

Just repeat doing Step 2 and wait for several iterations, there would be some cars can complete the whole track. Don't stop, keep doing this, until most cars can finish the race (i.e. the network model has converged)

Caution

• Please don't forget selecting elites manually (mouse left button) after each round!
• For many reasons, Genetic_car_auto.py has not been implemented -.-
• Yellow car is the car that has the longest running distance, it doesn't matter

Details

• This repo uses genetic algorithm to optimize the neural network. I construct a network with 3 input layers, 5 hidden layers, and 2 output layers using numpy.
• Each car agent has three distances from itself to the track boundary (also the network's input)
• The network model has two output vectors which respectively represents angle and speed of car
• There are many hyper-parameters with my notations at the beginning of the code, you can customize them

Requirements

$ pip install -r requirements.txt