M.Sc Thesis

M.Sc StudentSisso Itay
SubjectInfo-Gap Approach to Multi Uninhabited Aerial Vehicles
Control under Severe Uncertainty
DepartmentDepartment of Aerospace Engineering
Supervisors PROF. Tal Shima
Full Thesis textFull thesis text - English Version


A robust satisficing approach based on info-gap theory is suggested as a solution for a spatial search planning problem with imprecise probabilistic data. In the presented problem, a group of agents (uninhabited aerial vehicles (UAVs)) are searching predefined patches of land for stationary ground targets, given an a priori probability map of the targets' locations. This prior probabilistic information is assumed to be severely uncertain and may contain large errors. An analysis of a simplified 2-cell and one searching UAV case shows that in some situations one might prefer a different strategy than the expected utility maximizing one, in terms of robustness to uncertainty. Numerical results, obtained using a deterministic numeric method, confirm the theoretical predictions for more complex cases, containing 4 cells, while ensuring the optimality of the examined solutions due to the completeness nature of the deterministic search algorithm used. Finally, numeric analysis of robust satisficing solutions, obtained using a stochastic search method (genetic algorithm), on a group of 50 randomly generated cases, each one containing 15 cells and 3 searching UAVs, reveals an interesting behavior of a consolidation of effort in specific cells. The consolidation focused on cells with a high prior probability of target existence, and became more distinct when lowering the failure criterion. The phenomenon was explained based on the uncertainty model which was used in the solutions. Moreover, the high order solutions implies the potential of robust satisficing in more realistic scenarios. As the robustness to uncertainty comes at the expense of the expected utility, one must choose its decisions carefully. However, it is shown throughout the work that, in various circumstances, one obtains results which are significantly superior to the expected utility maximizing strategy in terms of robustness, while sacrificing almost no expected utility.