Vector maps: A lightweight and accurate map format for multi-robot systems

Abstract

SLAM algorithms produce accurate maps that allow localization with typically centimetric precision. However, such a map is materialized as a large Occupancy Grid. Beside the high memory footprint, Occupancy Grid Maps lead to high CPU consumption for path planning. The situation is even worse in the context of multi-robot exploration. Indeed, to achieve coordination, robots have to share their local maps and merge ones provided by their teammates. These drawbacks of Occupancy Grid Maps can be mitigated by the use of topological maps. However, topological maps do not allow accurate obstacle delimitations needed for autonomous robots exploration. So, robots still have to handle with Occupancy Grid Maps. We argue that Vector-based Maps which materialize obstacles using collections of vectors is a more interesting alternative. Vector Maps both provide accurate metric information likewise Occupancy Grid Maps, and represent data as a graph that can be processed for path planning and maps merging as efficiently as with topological maps. Conclusions are backed by several metrics computed with several terrains that differ in size, form factor, and obstacle density.