Distributed Pose Graph Optimization for Collaborative SLAM Using Fast ADMM

Collaborative Simultaneous Localization and Mapping (CSLAM) is a technology that enables multiple robots to estimate their own trajectories and collaboratively construct a global map in a large-scale unknown environment by sharing local information. The back-end of this technology involves solving a non-convex Pose Graph Optimization (PGO), which is usually formulated as <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$SE(d)$</tex>-synchronization. To address the performance degradation of centralized solvers when dealing with large-scale pose graphs, we employ a multi-level graph partitioning method to split the large-scale pose graph, resulting in Distributed Pose Graph Optimization (DPGO) with constraints. Additionally, this paper proposes a Fast Alternating Direction Method of Multipliers (ADMM) algorithm to achieve high-quality solutions more efficiently. Finally, we test the proposed method on various standard datasets and compare it with traditional ADMM. The experimental results demonstrate that the Fast ADMM accelerates solution convergence.