Search and Rescue Multi-Agent Simulation

This project simulates a multi-agent search-and-rescue mission using the PettingZoo MPE framework. Rescuers (adversaries) must guide victims to designated safe zones, navigating around obstacles and using cooperative intelligence to accomplish the task.

Getting Started

Prerequisites

Ensure you have Python installed and necessary packages for running PettingZoo and reinforcement learning frameworks.

Installation

1. Clone the repository:

   git clone https://gitlab.inf.elte.hu/student-projects-and-thesis/collective-intelligence/search.git
   cd search

2. Install dependencies:

   pip install -r requirements.txt

Project Overview

• Objective: Simulate a search-and-rescue operation where rescuers lead victims to specific safe zones based on clustering.
• Framework: Built using the PettingZoo MPE environment.
• Agents:
  • Rescuers: Adversarial agents guiding victims to the correct safe zones.
  • Victims: Agents that need to be rescued by being taken to matching safe zones.
• Safe Zones: Defined zones in the map’s four corners; each type of victim has a matching type of safe zone.

Key Features

1. Safe Zones:

   • Static Locations: Safe zones are positioned at each corner of the map.
   • Different Types: Each victim type corresponds to a unique type of safe zone, introducing a clustering challenge.

2. Reward System:

   • Rescuers earn rewards for successfully moving victims to their designated safe zones.
   • Victims are incentivized to avoid capture, reinforcing the search-and-rescue dynamics.

3. Collision Detection:

   • Refined detection ensures rescues occur only when victims and rescuers are close.
   • Introduced obstacle collision logic to prevent straightforward rescues.

Documentation

For a detailed description of the environment, reward system, training pipeline, and evaluation process, refer to the complete project documentation:
📄 Documentation.pdf

The documentation includes:

• Functionalities of the search-and-rescue environment.
• Code structure and modular components.
• Training and evaluation pipelines with performance metrics.

Usage

Run the following commands to start training or evaluating the environment:

# Training
python main.py --train

# Evaluation
python main.py --eval

Team Members

Jakab Etele, jangbo@inf.elte.hu​

Szarka Marcell, dia4sw@inf.elte.hu​

Pánczél Patrik, mbhr3h@inf.elte.hu

License

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