On the Validation of Adversarial Threats to Cooperative Unmanned Aerial Systems for Search and Rescue Missions

We study the effects of adversarial threats on the localization stack of cooperative Unmanned Aerial Systems (UAS) that perform cooperative search and rescue missions (CooSAR). We propose a methodology for incorporating different types of attacks in the localization stack of the UAS to spoof the sensors and actuators of the vehicles with two goals: fail the mission (not locate the target) and fly off path. For this purpose, we resort to a progressive-spiral-in (PSI) search path with theoretical success guarantees to cover the search area in finite time given at least a certain minimum number of aerial vehicles. Then, we formulate a control strategy for each UAS and implement two types of spoofing attacks: i) attack to the sensors and actuators, and ii) attack to the residuals of the GNSS localization measurements. We evaluate the effects of the attacks by measuring the decrease in effectiveness in finding the target and the reduction in area coverage when the UAS are subject to the attack. The attack is applied to one UAS at a time (the one closest to the target), and we find that even with this simple approach, the probability of finding the target is greatly reduced up to 36%.