Active perception for 3D range-only simultaneous localization and mapping with UAVs

Abstract

Most localization approaches do not take into account the possibility of controlling the robot to improve the perception, instead, the robot is just commanded with a predefined path. Active sensing strategies may lead to more efficient exploration and mapping approaches. The robot can adapt its trajectory, avoiding for instance non-observable motions or following those paths which are most informative. In the case of Range-only Simultaneous Localization and Mapping (RO-SLAM), this means selecting the path which is likely to produce the highest reduction in the uncertainty on the landmarks' positions. This paper proposes an active sensing approach able to command the UAV to more informative areas from the RO-SLAM point of view, allowing a faster filter convergence and better mapping accuracy.