AI Navigation

Overview

A simulation of a race game using natural selection algorithm NeuroEvolution of Augmented Topologies, NEAT in short.

The goal of an agent is to travel the most the distance in shortest amount of time in direction of a road without hitting the edges.

Demonstration

Learning

learning.mp4

A few generations after

learnt.model.mp4

Instalation

• Download or clone the repository
• Navigate to the root of the repository
• Create and activate a virtual environment and install packages (see bellow)

Linux

virtualenv venv
venv/bin/activate
pip install -r .\requirements.txt

Windows

virtualenv venv
venv\Scripts\activate.ps1
pip install -r .\requirements.txt

Execution

A user can freely control an unit if the program is run from __main_movement__.py else if the program is run from __main_ai__.py you need to make sure that the config value let_me_drive is set to True

Run the program from the root directory, e.g.:

python __main_ai__.py

Controls

User can control an unit with WSAD keys, it's meant only as a demonstration as to how it feels to be an agent. To skip a generation you can press the Del button, meant to manualy speed up the natural selection process (bevare; the next generation is derived from the one that was terminated).

Program can be launched:

• as an AI simulation __main_ai__.py
• as a movement simulation __main_movement__.py

AI Config

• road_index - index of road pre-saved as a pickle
• show_active_fitness_lines - shows where is the agent currently earning fitness
• show_car_vision - shows how far is the wall (main information inputed into net)
• show_all_fitness_lines - shows where all the fitness lines
• show_parent - highlights most prominent genome (doesn't guarantee reproduction)
• save_best_genome - saves the most prominent genome as a pickle to utils/genomes directory
• reverse_fitness_lines - makes agents compete in the opposite direction
• let_me_drive - allows user to try what it feels like to be an agent
• max_frames_elapsed - determines after how many frames will simulation start a new generation
• no_progress_frames_elapsed - determines after how many frames of agents not crossing fitness lines in the forward direction will the simulation start a new generation

Tests

Unit-tests are checking the core functionality of the program, tests run on majority of functions and methods. Tests can be run with the pytest command.

Mechanics

mechanics.mp4