Unmanned Aerial Vehicle Trajectory Planning with Direct Methods

Open Access
Geiger, Brian Raymond
Graduate Program:
Aerospace Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
April 14, 2009
Committee Members:
  • Joseph Francis Horn, Dissertation Advisor
  • Joseph Francis Horn, Committee Chair
  • Lyle Norman Long, Committee Member
  • Jack W Langelaan, Committee Member
  • Alok Sinha, Committee Member
  • direct collocation
  • uav
  • path planning
  • trajectory optimization
  • neural network
  • flight test
A real-time method for trajectory optimization to maximize surveillance time of a fixed or moving ground target by one or more unmanned aerial vehicles (UAVs) is presented. The method accounts for performance limits of the aircraft, intrinsic properties of the camera, and external disturbances such as wind. Direct collocation with nonlinear programming is used to implement the method in simulation and onboard the Penn State/Applied Research Lab's testbed UAV. Flight test results compare well with simulation. Both stationary targets and moving targets, such as a low flying UAV, were successfully tracked in ight test. In addition, a new method using a neural network approximation is presented that removes the need for collocation and numerical derivative calculation. Neural networks are used to approximate the objective and dynamics functions in the optimization problem which allows for reduced computation requirements. The approximation reduces the size of the resulting nonlinear programming problem compared to direct collocation or pseudospectral methods. This method is shown to be faster than direct collocation and psuedospectral methods using numerical or automatic derivative techniques. The neural network approximation is also shown to be faster than analytical derivatives but by a lesser factor. Comparative results are presented showing similar accuracy for all methods. The method is modular and enables application to problems of the same class without network retraining.