SIGMA: Sheaf-Informed Geometric Multi-Agent Pathfinding

This repository contains the implementation of SIGMA algorithm accepted for oral presentation at ICRA 2025. The framework integrates sheaf theory with multi-agent reinforcement learning for efficient path planning.

Requirements

conda create -n sigma python==3.10
conda activate sigma
pip install numpy==1.23.5
# CUDA 11.8
pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118
pip install -r requirements.txt

Project Structure

.
├── configs.py        # Configuration parameters
├── model.py         # Core network architecture
├── train.py         # Distributed training entry
├── test.py          # Evaluation and visualization
├── environment.py   # Multi-agent simulation environment
└── worker.py        # Ray parallelization components

Environment Types

SIGMA supports two types of environments:

1. Room-like Environment: Open spaces with obstacles simulating indoor scenarios
2. Maze Environment: Complex corridor structures with multiple pathways

You can switch between these environments by modifying parameters in configs.py. We provide pre-trained models for both environment types.

Training

# Start training
python train.py

Citation

If you use this work in your research, please cite:

@article{liao2025sigma,
  title={SIGMA: Sheaf-Informed Geometric Multi-Agent Pathfinding},
  author={Liao, Shuhao and Xia, Weihang and Cao, Yuhong and Dai, Weiheng and He, Chengyang and Wu, Wenjun and Sartoretti, Guillaume},
  journal={arXiv preprint arXiv:2502.06440},
  year={2025}
}

Authors

Shuhao Liao
Weihang Xia
Yuhong Cao
Weiheng Dai
Chengyang He
Wenjun Wu
Guillaume Sartoretti