This project simulates a car racing environment using the NEAT (NeuroEvolution of Augmenting Topologies) algorithm to evolve neural networks that control the cars.

NEAT (NeuroEvolution of Augmenting Topologies) is a type of evolutionary algorithm used to generate artificial neural networks. Unlike traditional neural networks that have a fixed structure, NEAT evolves both the neural network's weights and its architecture over generations. This approach allows it to optimize complex tasks by:

• Starting Simple: NEAT begins with minimal networks and gradually adds complexity through mutations, ensuring efficient learning without overfitting.
• Evolving Topologies: It not only adjusts connection weights but also changes the network structure by adding nodes and connections over time.
• Speciation: NEAT maintains diverse solutions by grouping similar networks into species, allowing new structures to evolve without being immediately outcompeted by more developed ones.

This makes NEAT particularly powerful for solving tasks where the optimal neural network architecture is unknown or highly complex.

Due to the nature of this learning style, unexpected behaviors often emerge. Here are some key observations from the simulation:

• Circling Behavior:
  Initially, many cars were driving in circles at their first opportunity. This occurred because the reward system only incentivized distance traveled, allowing them to exploit the system without progressing.

  • Solution: I implemented a series of checkpoints to encourage more purposeful movement and proper driving behavior.

• Swerving Movement:
  Another surprising behavior was that the cars learned to drive in a swerving manner, sweeping their 12 o'clock radar in a way similar to how an insect uses its antennae.

  • Analysis: This behavior likely helps the model gather more accurate feedback from its environment, showing that the network is actively seeking more input.
  • Next Steps: To address this, I plan to modify the 12 o'clock radar to a 20-degree "line of sight" (-10 to +10 degrees), aligning the sensor's design with the behavior the cars are naturally adopting.

• src/: Contains the source code.

  • assets/images/: Contains image assets like car sprites and maps.
  • constants/: Defines game constants.
  • input/: Handles user input.
  • objects/: Contains object classes like Car.
  • rendering/: Handles rendering of the game.
  • simulation/: Contains the simulation logic.
  • state/: Manages global simulation state.

• config/: Contains configuration files for NEAT.

1. Install Dependencies:

   pip install -r requirements.txt

2. Run the Simulation: (WIP: For the time being be sure to run from inside src)

   python main.py

• NEAT Configuration: Modify config/config.txt to adjust NEAT parameters like population size, mutation rates, etc.

• ESC: Quit the simulation.
• R: Toggle radar visualization.
• UP/DOWN Arrows: Increase or decrease simulation speed.

• Car Image: Located in assets/images/car.png.
• Map Image: Located in assets/images/vegas.png.