This is an archive of a very old project, made around 2012 in Java 1.7. It was created for my undergraduate degree that presented a concept of generating simple 2d models that adapt themselves to the track they have to pass using a genetic algorithm.
I decided to publish it more than a decade later, to keep it alive and maintained.
Each simulation round evaluates a population of cars. The initial population either is completely randomized, or can be created from the previously saved state. As the simulation progresses, cars that travel farther receive a higher probability of being selected for the next generation, where they produce offspring through crossover and mutation.
The simulation is highly customizable. You can adjust, for example:
- Track layout and environmental gravity
- Car properties such as elasticity, wheel friction, and density (weight)
- Structural limits, including the number of wheels and car body polygons
- Genetic algorithm properties, such as the selection method, crossover strategy, and mutation probability
Currently, the whole UI and internal docs are in Polish only. Sorry! I plan to translate it to English in the future, but due to how old this project is and how much other work it needs, it might take a while.
Genetic Car is based on LibGDX and follows a basic structure of a multi-platform LibGDX project.
All logic resides in the code project, while the desktop one provides just a single window to launch it.
The desktop project is responsible for creating the executable. The executable is a fat jar, but it might be still
buggy a bit. If you want to run it, then I recommend using a custom run task in Gradle.
Originally, this project did not use Gradle, but Ant, and I haven't ported it properly yet.
- Encoding issues in various places
- The build to distribute it (
dist) doesn't create a proper folder structure that can hold custom data such as tracks
This project is licensed under the Apache 2.0 License. Refer to the LICENSE file for the full text.