Factor Graph-Based Dense Mapping for Mobile Robot Teams Using VDB-Submaps

A large number of works exist in the field of mobile robot based simultaneous localization and mapping. While the original SLAM problem has been considered solved for years, there still exist various environments, use cases, or robot configurations which require new approaches in order to successfully perform the task. This work addresses how a group of mobile robots can collaboratively create a dense 3D map that is globally consistent and accounts for uncertainties in measurement data and estimates. The main challenge is a compact representation of the robot-local submaps in order to minimize the data flow as well as a fast and accurate merging scheme to create a consistent global map. We leverage OpenVDB as underlying data structure to efficiently create submaps which are then fused in a factor graph-based backend. We extensively test and evaluate the framework and show that it is capable of creating dense 3D maps of challenging environments in real-time.