Skip to content

README.md

Latest commit

 

History

History
127 lines (108 loc) · 6.28 KB

README.md

File metadata and controls

127 lines (108 loc) · 6.28 KB

Autonomous Driving using NeuroEvolution of Augmenting Topologies

Overview

AutonomousDriver is a project that uses the NEAT (NeuroEvolution of Augmenting Topologies) algorithm to evolve neural networks for autonomous driving. The project allows users to train and test neural networks on custom maps.

Installation

Clone the repository:

git clone https://github.com/buzzpranav/AutonomousDriver.py
cd AutonomousDriver.py

Install the required dependencies:

pip install -r requirements.txt

Usage

python SelfDrive.py

Code Overview

Car Class

  • Initialization: Sets up the car's properties, including its position, speed, radar sensors, and visual representation.
  • Drawing: Methods to draw the car and its radar on the screen, providing a visual indicator of what the car "sees."
  • Collision Detection: Checks for collisions with the map borders, determining if the car is still "alive."
  • Radar: Simulates radar sensors to detect distances to obstacles, feeding this information to the neural network.
  • Update: Updates the car's position, rotation, and checks for collisions, driving the simulation forward.
  • Data and Reward: Methods to retrieve radar data for the neural network and calculate the reward based on the distance traveled.

run_simulation Function

  • Initialization: Sets up the NEAT neural networks and car instances for the current generation.
  • Simulation Loop: Runs the main simulation loop, updating car positions, handling neural network outputs, and checking for collisions.
  • Fitness Calculation: Adjusts the fitness scores based on the car's performance and renders the simulation on the screen.

plot_fitness Function

  • Real-time Plotting: Uses Matplotlib to plot the average fitness scores over generations, providing a visual tool to monitor the progress of the AI.

Main Function

  • Initialization: Sets up Pygame and NEAT configurations, including the game display and NEAT parameters.
  • Threading: Runs the simulation and fitness plotting in separate threads, ensuring the simulation remains responsive while updating the fitness plot.

Custom Maps

You can use your own maps for training and testing. Simply create a map using a simple drawing tool like MS Paint and save it as an image file. Upload the image to the maps folder in the project directory.

Steps to Create a Custom Map

  1. Open MS Paint or any other simple drawing tool.
  2. Draw your map. Ensure that the roads and paths are clearly defined.
  3. Save the image as a .png or .jpg file.
  4. Upload the image to the maps folder in the project directory with the name map[number].
  5. Update variable racemap in SelfDrive.py to your new map number

Configuration

The project uses a configuration file (config.txt) to set various parameters for the NEAT algorithm. It is optimized and should not be changed unless you know what you are doing:

NEAT

  • fitness_criterion: Determines how fitness is evaluated. Options are max, min, or mean.
  • fitness_threshold: The fitness level required to consider the problem solved.
  • pop_size: The population size for each generation.
  • reset_on_extinction: If True, the population is reset if all species go extinct.

DefaultGenome

  • Node Activation Options
  • activation_default: Default activation function for nodes (e.g., tanh).
  • activation_mutate_rate: Probability of mutating the activation function.
  • activation_options: List of possible activation functions.

Node Aggregation Options

  • aggregation_default: Default aggregation function for nodes (e.g., sum).
  • aggregation_mutate_rate: Probability of mutating the aggregation function.
  • aggregation_options: List of possible aggregation functions.

Node Bias Options

  • bias_init_mean: Mean of the initial bias values.
  • bias_init_stdev: Standard deviation of the initial bias values.
  • bias_max_value: Maximum bias value.
  • bias_min_value: Minimum bias value.
  • bias_mutate_power: Standard deviation of the bias mutation.
  • bias_mutate_rate: Probability of mutating the bias.
  • bias_replace_rate: Probability of replacing the bias.

Genome Compatibility Options

  • compatibility_disjoint_coefficient: Coefficient for disjoint genes in compatibility calculation.
  • compatibility_weight_coefficient: Coefficient for weight differences in compatibility calculation.

Connection Add/Remove Rates

  • conn_add_prob: Probability of adding a new connection.
  • conn_delete_prob: Probability of deleting an existing connection.

Connection Enable Options

  • enabled_default: Default state of new connections (enabled or disabled).
  • enabled_mutate_rate: Probability of mutating the enabled state of a connection.

Network Parameters

  • feed_forward: If True, the network is feed-forward only.
  • initial_connection: Initial connection pattern (e.g., full).
  • num_hidden: Number of hidden nodes.
  • num_inputs: Number of input nodes.
  • num_outputs: Number of output nodes.

Node Response Options

  • response_init_mean: Mean of the initial response values.
  • response_init_stdev: Standard deviation of the initial response values.
  • response_max_value: Maximum response value.
  • response_min_value: Minimum response value.
  • response_mutate_power: Standard deviation of the response mutation.
  • response_mutate_rate: Probability of mutating the response.
  • response_replace_rate: Probability of replacing the response.

Connection Weight Options

  • weight_init_mean: Mean of the initial weight values.
  • weight_init_stdev: Standard deviation of the initial weight values.
  • weight_max_value: Maximum weight value.
  • weight_min_value: Minimum weight value.
  • weight_mutate_power: Standard deviation of the weight mutation.
  • weight_mutate_rate: Probability of mutating the weight.
  • weight_replace_rate: Probability of replacing the weight.

DefaultSpeciesSet

  • compatibility_threshold: Threshold for compatibility between genomes.

DefaultStagnation

  • species_fitness_func: Function to evaluate species fitness (e.g., max).
  • max_stagnation: Maximum number of generations a species can stagnate before being removed.
  • species_elitism: Number of top species to protect from stagnation.

DefaultReproduction

  • elitism: Number of top genomes to protect from being replaced.
  • survival_threshold: Proportion of the population that survives to the next generation.

License

This project is licensed under the MIT License. See the LICENSE file for details.