Remote VDB-Mapping: A Level-Based Data Reduction Framework for Distributed Mapping

Abstract

In recent times, the distribution of mobile walking robots has increased sharply due to their high availability. To utilize the increased mobility of these robots, autonomous navigation algorithms require fast and accurate volumetric maps, generated by the sensors of the robot. Usually these maps have to be transferred between different systems, like for example in a multi robot setup or if a operator needs access to the data on a remote control room. Since the memory footprint of the data scales indefinitely with the resolution and size of the environment, these maps can become quite large. This is especially the case in volumetric 3D maps due to the added third dimension. Sending the whole map over a network would quickly overburden a wireless network. In this paper we present a distributed mapping approach, which is able to efficiently share identical maps between different machines. Further this approach can be be used to combine data from various sources (e.g. collaborative robot teams) into a single map. This is archived by only exchanging memory efficient reduced update grids between the systems. We present a level based approach which is able to further reduce the memory footprint of the update grids depending on different use-cases. The algorithm can easily integrate various systems and was evaluated using an ANYmal robot operated over a remote base station. The presented experiments show, that this approach is able to significantly reduce the amount network traffic required in a remote mapping scenario.