Cooperative Cloud SLAM on Matrix Lie Groups

Abstract

In this paper we present a Cooperative Cloud SLAM on Matrix Lie Groups ($\text {C}^2\text {LEARS}$), which enables efficient and accurate execution of simultaneous localization and environment mapping, while relying on integration of data from multiple agents. Such fused information is then used to increase mapping accuracy of every agent itself. In particular, the agents perform only computationally simpler tasks including local map building and single trajectory optimization. At the same time, the efficient execution is ensured by performing complex tasks of global map building and multiple trajectory optimization on a standalone cloud server. The front-end part of $\text {C}^2\text {LEARS}$ is based on a planar SLAM solution, while the back-end is implemented using the exactly sparse delayed state filter on matrix Lie groups (LG-ESDSF). The main advantages of the front-end employing planar surfaces to represent the environment are significantly lower memory requirements and possibility of the efficient map exchange between agents. The back-end relying on the LG-ESDSF allows for efficient trajectory optimization utilizing sparsity of the information form and exploiting higher accuracy supported by representing the state on Lie groups. We demonstrate $\text {C}^2\text {LEARS}$ on a real-world experiment recorded on the ground floor of our faculty building.