Cooperative Mapping and Target Search Over an Unknown Occupancy Graph Using Mutual Information

Abstract

A cooperative mapping and target-search algorithm is presented for detecting a single moving ground target in an urban environment that is initially unknown to a team of autonomous quadrotors equipped with noisy, range-limited sensors. The target moves according to a biased random-walk model, and search agents (quadrotors) build a target state graph that encodes past and present target positions. A track-before-detect algorithm assimilates target measurements into the log-likelihood ratio and anisotropic kriging interpolation predicts the location of occupancy nodes in unexplored regions. Mutual information evaluated at each location in the search area defines a sampling-priority surface that is partitioned by a weighted Voronoi algorithm into candidate waypoint tasks. Tasks are assigned to each agent by iteratively solving a utility-maximizing assignment problem. Numerical simulations show that the proposed approach compares favorably to non-adaptive lawnmower and random coverage strategies. The proposed strategy is also demonstrated experimentally through an outdoor flight test using two real and two virtual quadrotors.