Human-Robot Collaborative Site Inspection Under Resource Constraints

This paper is on human-robot collaborative site inspection and target classification. We consider the realistic case that human visual performance is imperfect (depending on the sensory input quality), and that the robot has constraints in communication with human (e. g., limited chances for query, poor channel quality). The robot has limited onboard motion and communication energy and operates in realistic channel environments experiencing path loss, shadowing, and multipath. We then show how to co-optimize motion, sensing, and human queries. Given a probabilistic assessment of human visual performance and a probabilistic channel prediction, we pose the co-optimization as multiple-choice multidimensional knapsack problems. We then propose a linear program-based efficient near-optimal solution, mathematically characterize the optimality gap, showing it to be very small, and mathematically characterize properties of the optimum solution. We then comprehensively validated the proposed approach with extensive real human data (from Amazon MTurk) and real channel data (from downtown San Francisco), confirming that the proposed approach significantly outperforms benchmark methodologies.