We apply the theory of partially observable Markov decision processes (POMDPs) to the design of guidance algorithms for controlling the motion of unmanned aerial vehicles (UAVs) with on-board sensors for tracking multiple ground targets. While POMDPs are intractable to optimize exactly, principled approximation methods can be devised based on Bellmans principle. We introduce a new approximation method called nominal belief-state optimization (NBO). We show that NBO, combined with other application-specific approximations and techniques within the POMDP framework, produces a practical design that coordinates the UAVs to achieve good long-term mean-squared-error tracking performance in the presence of occlusions and dynamic constraints. Although the POMDP/NBO combination exemplifies increased tracking performance, this performance gain can be hindered by computational complexity. Implementing computationally intense subroutines intrinsic to the POMDP/NBO approach in highly parallel graphics processing units (GPUs) will allow the realization of our approach on complex systems in near real time.
Benefit: Improved UAV surveillance technique, Optimal sensor resource management, High Performance GPU library
Keywords: Non-Linear Control, Stocastic Dynamic Programming, Performance Tuning, Gpgpu
