Cocaud, Cedric2013-11-072013-11-0720072007Source: Masters Abstracts International, Volume: 46-03, page: 1705.http://hdl.handle.net/10393/27508http://dx.doi.org/10.20381/ruor-18746The feasibility and survivability of Unmanned Air Vehicles (UAV's) in field applications have been demonstrated during many of the recent conflicts. UAV's have the potential to reduce costs and personnel while performing certain missions such as surveillance and reconnaissance with higher efficiency and lower risks. Since it is now recognized that a single UAV cannot achieve all tasks required during a mission, the option of using fleets of UAV's with complementary capabilities is now explored. However, controlling and coordinating such fleets will be almost impossible when considering the overwhelming number of degrees of freedom and their associated constraints, if individual UAV do not possess a high level of autonomy. Addressing the need for intelligent controllers, this work presents a new approach for on-line trajectory planning of a single UAV. The approach uses a genetic algorithm and integrates constraints of different types including specific times of arrival, vectors of approach close to target points, multiple objectives, weather conditions and behavioural guidelines with respect to speed, altitude and obstacle distance. The UAV's controller is divided into three modules. The first one is a trajectory generator that uses an analytical model to simulate the dynamics of UAV's. It is used to compute new trajectories in 3D as well as to assess the feasibility of existing ones. The second module is a knowledge base that uses a hybrid octree structure to represent terrain in 3D. The octree compresses digital elevation data by setting the resolution of the UAV's internal map to different levels based on the homogeneity of the terrain. The third module is a Genetic Algorithm that finds and optimizes candidate trajectories which fulfill all objectives of a mission while taking in account a variable number of constraints with varying preponderance. Simulations have shown that the intelligent controller could find viable trajectories within 10 to 20 seconds for missions with a single objective, demonstrating the feasibility of using this control scheme for real-time path planning for UAV's. Dynamically feasible trajectories satisfying all constraints could also be found for missions with up to three objectives. However, real time capabilities were severely reduced for such complex missions, the average computing time varying from 200 to 450 seconds. In all cases, all resulting trajectories were dynamically feasible, and could be readily used by the autopilot.168 p.enEngineering, Mechanical.Thesis