A Collaborative Graph-based SLAM Framework Using a Computationally Effective Measurement Algebra

Abstract

Simultaneous localization and mapping (SLAM) is an essential task for autonomous rover navigation in an unknown environment, especially if no absolute location information is available. This paper presents a computationally lightweight framework to enable agents with limited processing power to carry out the SLAM cooperatively and without absolute onboard localization sensors in a 2D environment. The proposed solution is built on a graph-based map representation, where nodes (resp. edges) represent landmarks (resp. odometry-based relative measurements), a measurement algebra with embedded uncertainty, and a compact database format that could be stored on a server in a centralized manner. The operations required by the agents to insert a new landmark in the graph, update landmark positions and combine measurements as a loop is closed in the graph are detailed. The resulting framework was tested in a laboratory environment and on a public dataset with encouraging results; hence our method can be used for cost-effective indoor mobile agents with limited computational resources and onboard sensors to achieve a mapping while keeping track of the agent's position. The method can also be easily generalized for a 3D scenario.