A physics-based simulation engine that utilizes Genetic Algorithms to evolve optimal 2D vehicle topologies capable of traversing procedurally generated terrain.
EvoCar is a research application designed to demonstrate the power of evolutionary computation. Instead of manually designing a vehicle to cross rough terrain, the program generates a random population and "evolves" the best design over generations using biological operators.
The car consists of a chassis (skeleton) and two wheels. The simulation leverages Box2D for physics and SFML for rendering.
Each vehicle is defined by a specific chromosome containing:
- Wheel Size: Radius of the two wheels.
- Chassis Topology: Vertex distribution of the car's body.
- Density: Material density of both the wheels and the chassis.
- Wheel Density.
Watch the evolution process in real-time. Mutation hyperparameters can be adjusted in config/EvolutionaryAlgorithmConfig.cc. You can force the next generation by pressing "N".
Vehicles are generated based on a specific chromosome. You can tweak:
- Target velocity (
config/CarConfig.cc). - Min/Max density and attribute limits (
config/EvolutionaryAlgorithmConfig.cc).
The terrain is generated randomly. You can adjust the minimum and maximum slope changes in the configuration files to make the track harder or easier.
We implemented a custom Genetic Algorithm with:
- Tournament Selection
- Gaussian Mutation
- Configurable Hyperparameters: Population size, mutation rate, tournament size (all editable in
config/EvolutionaryAlgorithmConfig.cc).
- Save/Load: The genome data and results can be saved to external files (Toggle
SAVE_TO_FILEinconfig/Config.cc). - Charts: Real-time plotting of vehicle speed and current position.
The project is structured around the EvolutionaryAlgorithm class, which manages the population lifecycle.
- EvolutionaryAlgorithm: Handles population initialization (
getPopulation), fitness evaluation (setFitness), and the transition to the next generation (nextStep). - Shape.h: Defines the physical structures (Box, Circle, Polygon).
- Language: C++
- Physics Engine: Box2D (
libbox2d-dev) - Graphics: SFML (
libsfml-dev) & ImGui (libimgui-dev) - Testing: Google Test (
libgtest-dev) & LCOV - Build System: CMake & Make
You can install all dependencies using the provided script install_packages.sh.
sudo apt-get install build-essential cmake libbox2d-dev libgtest-dev \
libsfml-dev libimgui-dev libudev-dev libfreetype-dev libxrandr-dev libx11-dev
To build the project from source:
# 1. Clone the repository
git clone https://github.com/CrustyCracker/EvoCar.git
cd EvoCar
# 2. Build the project using the helper script
sh fresh_build.sh
# OR build manually:
mkdir build
cd build
cmake ..
make
- Run Simulation:
./EvoRacer - Run Unit Tests:
./Test
| Key | Action |
|---|---|
| P | Pause / Resume simulation |
| N | Force skip to the Next Generation |
| Q | Quit application |
The project includes unit tests using Google Test. We track code quality using LCOV.
To generate a Test Coverage Report:
# 1. Ensure lcov is installed
sudo apt-get install lcov
# 2. Run the coverage script
sh test_coverage.sh
# 3. Open the report
# Open CodeCoverageReport/index.html in your browser.
This project was developed with a focus on rigorous software engineering practices.
- Total Development Time: ~167 hours.
- Estimated Time: 118 hours.
- Scope: From library selection and architecture design to implementation of physics, genetic algorithms, and data visualization.
Project Link: https://github.com/CrustyCracker/EvoCar