Autonomous exploration and simultaneous object search using aerial robots

Abstract

In this paper, a strategy for autonomous exploration of unknown environments and simultaneous object search using aerial robots is proposed. An aerial robot is initially armed with the capacity to autonomously explore unknown environments, detect objects that belong to certain semantic classes of interest and relate these detections to the respective location on the map. Provided this information, a sampling-based semantically-enhanced exploration path planning algorithm is designed such that in its every iteration it identifies a finite-depth collision free path that maximizes a gain related to exploring new space, as well as a gain related to the resolution of the observation of the previously mapped parts of the environment that relate to detected objects of interest. Through this ability to explore the unknown space, while accounting for the resolution of observation of viewpoints perceiving objects of interest, online 3D mapping that is focused on the subsets of the map most relevant with object search missions is achieved. The proposed algorithm is evaluated, verified and demonstrated in both simulation, as well as experimental studies using an autonomous aerial robot capable of GPS-denied visual-inertial localization and mapping operation.