Communication-Robust Asynchronous Distributed LiDAR Collaborative Smoothing and Mapping

In large-scale, complex, and communication-constrained environments, efficient multi-robot cooperation places stringent demands on Simultaneous Localization and Mapping (SLAM) systems. Existing Collaborative LiDAR SLAM (C-LSLAM) approaches achieve high localization accuracy but rely on high-bandwidth, low-latency communication, limiting real-Time performance, reliability, and scalability. We present Multi-Proxy, an asynchronous, distributed, and decentralized C-LSLAM framework using a progressive loop closure detection strategy. The lightweight descriptor reduces inter-robot communication while maintaining rich collaborative constraints across large variations in viewpoint. Unlike conventional synchronous distributed optimization, Multi-Proxy's back-end employs an asynchronous Alternating Direction Method of Multipliers (ADMM), enabling efficient Asynchronous Distributed Pose Graph Optimization (ADPGO) without waiting for synchronous estimates, ensuring robust real-Time performance under communication packet loss or latency. Designed as a collaborative localization plugin, Multi-Proxy integrates seamlessly with existing C-LSLAM systems, enhancing flexibility and scalability. Experiments show that Multi-Proxy outperforms State-Of-The-Art (SOTA) C-LSLAM methods in Absolute Trajectory Error (ATE), Mean Map Entropy (MME), Average Wasserstein Distance (AWD), and Chamfer Distance (CD), while reducing communication bandwidth by over 80%.