Follow the Gap + Genetic ML

An autonomous racing algorithm that combines the Follow the Gap (FTG) with Genetic Algorithm optimization to evolve optimal driving parameters for F1TENTH race cars in simulation.

Note: This Algorithm requires using the f1tenth_gym_ros simulator, you can find it here

This comes from my f1tenth repository, can be found here

Prerequisites

• f1tenth_gym_ros simulator installed and running
• ROS2 Foxy
• Python 3.8+
• Docker

Installation

# First clone into your local workspace

# Build the package
colcon build --packages-select genetic_driver
source install/local_setup.bash
# + source any other local files required

Running the Algorithm

1. Start the F1TENTH Simulator

ros2 launch f1tenth_gym_ros gym_bridge_launch.py

2. Open New Terminals and Source Setup

For each new terminal:

docker exec -it f1tenth_gym_ros-sim-1 /bin/bash
source /opt/ros/foxy/setup.bash
source install/local_setup.bash

3. Run the Algorithm

Terminal 1 - Follow the Gap Driver:

ros2 run genetic_driver ftg

Terminal 2 - Genetic Supervisor:

ros2 run genetic_driver supervisor

The supervisor will automatically start evolving after a couple seconds.

Interactive Controls

While the algorithm is running, you can interact via the supervisor terminal:

• Press f + Enter: View current champion genome stats (fastest lap)
• Press r + Enter: Manually reset the current simulation run
• Press t + Enter: Print recent valid lap times to console
• Press h + Enter: Write detailed valid-lap history to log/full_history.txt (overwrites each time)
  • Log file will be located in the container directory: /sim_ws/install/genetic_driver/lib/python3.8/site-packages/log/full_history.txt

Example output:

 🏆 CURRENT CHAMPION (As of Generation 5)
     GenomeID : 12
     Time     : 24.37s
     Score    : 31.55
     Genes    : [1.234, 5.67, 2.1, ...]

Implementation Details

Architecture

GeneticSupervisor (ROS2 Node)
├── Manages simulation lifecycle
├── Tracks lap times & crashes
├── Publishes evolved parameters
└── Handles user input

GeneticML (Core Algorithm)
├── Maintains population of genomes
├── Calculates fitness scores
├── Selects & breeds best performers
└── Mutates genes for diversity

FollowTheGap (FTG Driver)
├── Processes LIDAR scans
├── Implements gap detection
├── Publishes steering/speed commands
└── Reports crash detection

Configuration

Edit parameters in genetic_ml.py:

self.population_size = 20      # Number of genomes per generation
self.max_generations = 50       # Total generations to evolve
self.elitism_percent = 0.2      # Top 20% always survive

Genetic Parameters

Each genome contains these evolving parameters (in GeneticParameters):

• MAX_LIDAR_DIST - Maximum LIDAR range for gap detection
• STRAIGHT_SPEED - Target velocity on straights
• TURN_SPEED - Target velocity during turns
• STEERING_SCALE - Steering sensitivity multiplier
• ...and more (defined in parameters.py)

Fitness Function

fitness_score = (1000 / lap_time) - (crash_count * 500)

• Positive side: Rewards faster lap times
• Negative side: Heavy penalty (500 points) for each crash
• Result: Evolves safe, fast driving strategies

Selection & Breeding

1. Elitism: Top 20% of genomes automatically advance
2. Crossover: Random parents breed offspring
3. Mutation: Offspring genes randomly mutate for diversity
4. Replacement: New generation replaces worst performers

Project Structure

genetic_driver/
├── genetic_ml.py           # Core genetic algorithm
├── parameters.py           # GeneticParameters class
├── supervisor.py           # ROS2 supervisor node
├── ftg_driver.py          # Follow the Gap implementation
└── README.md