CoMA-SLAM: Collaborative Multi-Agent SLAM System

1. Introduction

CoMA-SLAM is an advanced collaborative multi-agent Simultaneous Localization and Mapping (SLAM) system. This system can handle collaborative navigation and map construction tasks for multiple agents in complex environments, supporting real-time processing and high-precision reconstruction.

2. Environment Setup

2.1 System Requirements

• Ubuntu 20.04+
• Python 3.11+
• CUDA 11.0+

2.2 Environment Installation

conda create -n coma_slam python=3.11
conda activate coma_slam

pip install -r requirements.txt

3. Supported Datasets

Download the MultiAgentDataset and organize the dataset according to the following structure:

CoMA-SLAM/
├── ReplicaMultiagent/
│   ├── Office-0/
│   ├── Apart-0/
└── AriaMultiagent/
    ├── room0/
    └── room1/

4. Quick Start

4.1 Running SLAM System

4.1.1 Multi-GPU Parallel Run

To utilize multiple GPUs, open multiple terminals and run the same command with different GPU IDs:

# Terminal 1 (GPU 0)
conda activate coma_slam
python run_slam.py configs/ReplicaMultiagent/office_0.yaml \
    --input_path ReplicaMultiagent/Office-0/ \
    --output_path experience/2dgs_office_0/ \
    --gpu 0

# Terminal 2 (GPU 1)  
python run_slam.py configs/ReplicaMultiagent/office_0.yaml \
    --input_path ReplicaMultiagent/Office-0/ \
    --output_path experience/2dgs_office_0/ \
    --gpu 1

4.2 Post-processing

After SLAM processing is complete, run the post-processing script:

python post_slam.py configs/ReplicaMultiagent/office_0.yaml \
    --input_path ReplicaMultiagent/Office-0/ \
    --output_path experience/2dgs_office_0/

5. Configuration Files

System configuration files are located in the configs/ directory:

configs/
├── ReplicaMultiagent/
│   ├── office_0.yaml
│   ├── apart_0.yaml
│   ├── apart_1.yaml
│   └── apart_2.yaml
└── AriaMultiagent/
    ├── room0.yaml
    └── room1.yaml

Citation

If you find this work useful, please consider citing:

@article{Chen_Su_Wang_Han_Xia_Bu_Li_Hu_Meng_Wang_2026,
  title={CoMA-SLAM: Collaborative Multi-Agent Gaussian SLAM with Geometric Consistency},
  volume={40},
  url={https://ojs.aaai.org/index.php/AAAI/article/view/37283},
  DOI={10.1609/aaai.v40i4.37283},
  number={4},
  journal={Proceedings of the AAAI Conference on Artificial Intelligence},
  author={Chen, Lin and Su, Yongxin and Wang, Jvboxi and Han, Pengcheng and Xia, Zhenyu and Bu, Shuhui and Li, Kun and Hu, Boni and Meng, Shengqi and Wang, Guangming},
  year={2026},
  month={Mar.},
  pages={2922--2929}
}