The article presents a novel algorithm for Distributed Pose Graph Optimization (DPGO) in collaborative localization and mapping (CSLAM) for multi-robot systems. The proposed method combines multi-level graph partitioning with an accelerated Riemannian optimization method to balance the optimization problem and improve performance. The algorithm outperformed previous DPGO protocols in simulations. The authors also evaluate the effects of different graph partitioning approaches on the correlation of the inter-subgraphs. The Highest scheme was found to have the best partitioning performance.
Publication date: 5 Jan 2024
Project Page: https://github.com/tjcunhao/distributed-pose-graph
Paper: https://arxiv.org/pdf/2401.01657
