Automated detection, localization, and registration of smart devices with multiple robots

In this paper we examine the problem of detecting and localizing an unknown number of objects of interest using a small team of mobile robots. Such problems frequently arise in infrastructure inspection or smart building applications, where the number of objects of interest is not known a priori, though the type of object being sought out is known. This task is difficult because the data association, i.e., the matching of measurements to targets, is unknown and the sensors are unreliable, i.e., they experience false positive and false negative detections and are unable to uniquely identify and label individual targets. We utilize the Probability Hypothesis Density (PHD) filter to perform multi-target estimation, completely avoiding the need for any explicit data association while simultaneously estimating the number of objects and their locations. The team selects actions, generated over a range of length scales, that maximize the expected information gain given the current estimate of the object locations. This information gain is computed as the mutual information between the set of targets and the binary event of receiving no detections. This hedges against uninformative actions in a computationally tractable manner. We present a series of simulated experiments, validating the performance of the proposed framework using multiple sensor modalities.