Swarm-SLAM : Sparse Decentralized Collaborative Simultaneous Localization and Mapping Framework for Multi-Robot Systems (2301.06230v3)

Abstract: Collaborative Simultaneous Localization And Mapping (C-SLAM) is a vital component for successful multi-robot operations in environments without an external positioning system, such as indoors, underground or underwater. In this paper, we introduce Swarm-SLAM, an open-source C-SLAM system that is designed to be scalable, flexible, decentralized, and sparse, which are all key properties in swarm robotics. Our system supports inertial, lidar, stereo, and RGB-D sensing, and it includes a novel inter-robot loop closure prioritization technique that reduces communication and accelerates convergence. We evaluated our ROS-2 implementation on five different datasets, and in a real-world experiment with three robots communicating through an ad-hoc network. Our code is publicly available: https://github.com/MISTLab/Swarm-SLAM

The paper "Swarm-SLAM : Sparse Decentralized Collaborative Simultaneous Localization and Mapping Framework for Multi-Robot Systems" (Lajoie et al., 2023 ) introduces Swarm-SLAM, a C-SLAM system designed with properties of scalability, flexibility, decentralization, and sparsity for swarm robotics. It supports inertial, lidar, stereo, and RGB-D sensing modalities. The system incorporates an inter-robot loop closure prioritization technique to decrease communication overhead and accelerate convergence. The paper validates Swarm-SLAM using a ROS 2 implementation across several datasets and a real-world experiment involving three robots communicating via an ad-hoc network. The source code is available on GitHub.

Core Contributions of Swarm-SLAM

Swarm-SLAM's primary contributions center on inter-robot loop closure detection, decentralized framework architecture, and open-source integration, which are detailed below:

Inter-Robot Loop Closure Detection

The system employs a budgeted prioritization approach based on algebraic connectivity to curtail communication needs. This method contrasts with exhaustive matching, which can quickly become a bottleneck in multi-robot systems due to communication constraints.

Decentralized Framework

The decentralized architecture adeptly handles sporadic inter-robot communications by optimizing neighbor management and pose graph computations. Leader election mechanisms facilitate localized optimization without reliance on a central server, enhancing robustness and scalability.

Open-Source Integration

Swarm-SLAM's compatibility with ROS 2 and its capacity to function over ad-hoc networks lowers the barrier to entry for researchers and practitioners. This open-source approach fosters community-driven development and wider adoption in diverse robotic applications.

Technical Underpinnings

Swarm-SLAM follows a conventional SLAM pipeline, dividing tasks into front-end and back-end processing.

Front-End Innovations

• Loop Closure Detection: The front-end integrates ScanContext for lidar-based loop closure and CosPlace for image-based global descriptors.
• Graph Sparsification: Preemptive graph sparsification on inter-robot matches, uses spectral sparsification techniques, which balances computational load and accuracy.

Back-End Processing

• Decentralized Computing: A temporary leader is selected among robots within communication range to facilitate localized optimization.
• Optimization: The back-end employs a robust Graduated Non-Convexity (GNC) solver, with strategic anchor selection to ensure global reference frame consistency. This approach is crucial for maintaining map coherence across the swarm.

Empirical Evaluation

Swarm-SLAM's performance was assessed through simulations and real-world experiments.

Dataset Performance

• Datasets: Evaluated on the KITTI, GrAco, M2DGR, and S3E datasets.
• Metrics: Swarm-SLAM achieved lower Absolute Translation Errors (ATE) with reduced communication overhead compared to distributed methods like DGS+PCM and D-GNC.

Real-World Validation

• Setup: A multi-robot experiment was conducted in an indoor parking lot.
• Results: The system demonstrated real-time performance under communication constraints, showing its suitability for resource-constrained platforms.

Implications and Future Trajectories

Swarm-SLAM offers advancements in multi-robot SLAM, especially for environments lacking centralized infrastructure. The framework's decentralized design and communication efficiency make it suitable for autonomous exploration and mapping tasks. Future work may include collaborative calibration methods and uncertainty estimation to reduce inter-robot localization errors, further enhancing the robustness and applicability of C-SLAM systems.

In summary, Swarm-SLAM introduces a C-SLAM framework designed for decentralized robotic systems, emphasizing communication efficiency and adaptability. Through innovations in loop closure detection, decentralized computing, and rigorous experimental validation, this work contributes to the ongoing development of multi-robot SLAM technologies.