An Asynchronous Parallel Algorithm Framework for Decentralized Pose Graph Optimization

This paper proposes an asynchronous parallel algorithm framework for decentralized pose graph optimization (PGO) in collaborative simultaneous localization and mapping. Our framework combines and generalizes the superiority of the existing works on PGO. Specifically, we decompose the objective function of decentralized PGO into the sum of multiple sub-problems to be solved in a fully distributed fashion, and reformulate the sub-problems with the rank-restricted relaxation strategy to reduce the size of the search space and provide a computational certifiable. Moreover, the proposed framework is Gauss-Seidel like without compromising convergence and accelerates the optimization inference with an asynchronous parallel scheme. Finally, we experimentally evaluate the performance of our framework embedded with different optimizers on bench-mark PGO datasets under bounded communication delays. The experiment results demonstrate that, compared to the existing methods, our framework has a faster converge without additional parameter turning and a strong resilience for communication delays.