Cooperative Decision-Making Under Uncertainties for Multi-Target Surveillance with Multiples UAVs

Abstract

Surveillance is an interesting application for Unmanned Aerial Vehicles (UAVs). If a team of UAVs is considered, the objective is usually to act cooperatively to gather as much information as possible from a set of moving targets in the surveillance area. This is a decision-making problem with severe uncertainties involved: relying on imperfect sensors and models, UAVs need to select targets to monitor and determine the best actions to track them. Partially Observable Markov Decision Processes (POMDPs) are quite adequate for optimal decision-making under uncertainties, but they lack scalability in multi-UAV scenarios, becoming tractable only for toy problems. In this paper, we take a step forward to apply POMDP methods in real situations, where the team needs to adapt to the circumstances during the mission and foster cooperation among the team-members. We propose to split the original problem into simpler behaviors that can be modeled by scalable POMDPs. Then, those behaviors are auctioned during the mission among the UAVs, which follow different policies depending on the behavior assigned. We evaluate the performance of our approach with extensive simulations and propose an implementation with real quadcopters in a testbed scenario.

http://ift.tt/2htjtcO