Multi-Agent Mapping and Navigation of Unknown GPS-Denied Environments Using a Relative Navigation Framework

When generating 3D maps with unmanned aerial vehicles (UAVs) in GPS-denied environments, it is important to correctly handle path planning, estimation, and mapping techniques. Because multirotor UAVs are limited in flight time, using multiple UAVs to map an environment collaboratively can significantly improve the mapping efficiency. This paper addresses the following key issues required to enable mapping with multiple agents: Combining a reactive path planner with an obstacle avoidance algorithm to handle navigation in complex environments. Estimating the relative and global states of a UAV separately with a relative navigation framework to allow for loop closures in the mapping process without causing the estimation to diverge. Adapting a graph-based simultaneous localization and mapping (graph-SLAM) technique for multiple UAVs flying simultaneously and merging their maps in realtime. We were able to use these strategies to generate dense maps in complex GPS-denied environments with multiple UAVs.