Multi-Agent Monocular SLAM

This article describes the development of an optimization method for multi-agent monocular SLAM systems. These systems allow autonomous robots to create a map of an unknown environment and to simultaneously localize themselves within it. The proposed multi-agent system combines measurements made by independent agents to increase the accuracy of the estimated poses of the agents and the created map. Our method is based on the single-agent monocular ORB-SLAM2 framework, and we develop a complete multi-agent optimization post-processing algorithm, effectively refining all camera trajectories and map points. Our experiments on the EuRoC machine hall dataset show that we can successfully combine the information of multiple SLAM agents to increase the accuracy of the estimated trajectories.