Brittle Star Locomotion with NEAT

This project focuses on simulating and evolving locomotion strategies for a brittle star using the NEAT (NeuroEvolution of Augmenting Topologies) algorithm. The simulation is built upon the MuJoCo physics engine and the biorobot library.

Project Overview

The primary goal is to train a neural network controller that enables a simulated brittle star to perform directed locomotion. The NEAT algorithm is employed to evolve both the topology and weights of these neural networks. The project includes functionalities for:

• Defining the brittle star's morphology (physical structure).
• Creating and managing the simulation environment.
• Implementing the NEAT algorithm for training controllers using TensorNEAT.
• Visualizing the simulation and the performance of trained models.
• Running experiments on HPC clusters.

File Structure

Here's a breakdown of the key files and directories:

.
├── README.md                     # This file
├── requirements.txt              # Python dependencies
└── src/
    ├── NEAT/                
    │   ├── train.py              # Main script to start NEAT training
    │   ├── config.py             # Configuration for simulation and NEAT parameters
    │   ├── neat_problem.py       # Defines the NEAT problem for the brittle star
    │   ├── neat_controller.py    # NEAT controller logic
    │   └── visualize.py          # Utilities for visualizing NEAT results
    ├── environment.py            # Defines the simulation environment for the brittle star
    ├── morphology.py             # Defines the physical structure (morphology)
    ├── initialize.py             # Handles GPU and MuJoCo initialization
    ├── render.py                 # Functions for rendering simulations and creating videos
    └── hpc_script.sh             # Script for running training on an HPC cluster 

Getting Started

Prerequisites

• Python 3.x
• MuJoCo physics engine installed and configured.
• Access to a GPU is recommended for faster simulations (see src/initialize.py for EGL setup).

Installation

1. Clone the repository:

   git clone https://github.com/Bendemeurichy/SEL3_2025_Groep_4.git
   cd SEL3_2025_Groep_4

2. Install Python dependencies:

   pip install -r requirements.txt

   It's highly recommended to use a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows use `venv\Scripts\activate`
   pip install -r requirements.txt

Running the Simulation and Training

1. Configuration:

   • Modify parameters for the simulation, brittle star morphology, and NEAT algorithm in config.py. This includes settings like the number of arms, segments per arm, simulation duration, population size, etc.

2. Training a NEAT Controller:

   • To start the training process, run the train.py script within the NEAT/ directory:

     python src/NEAT/train.py --mode <curriculum|no_curriculum> 

   • Trained models (genomes) will be saved in the ./models/ directory.

3. Running on an HPC Cluster (PBS):

   • The src/hpc_script.sh is provided for running training jobs on an HPC cluster that uses the PBS scheduler.
   • You might need to modify the script to match your cluster's environment (e.g., module loading, conda environment name).
   • Submit the job using qsub src/hpc_script.sh.

4. Visualization and Testing:

   • src/NEAT/visualize.py can be used to load and visualize the behavior of trained genomes.
   • Rendered videos are typically saved in the ./output_videos/ directory.