Rapid Indirect Trajectory Optimization on Highly Parallel Computing Architectures

摘要

A graphics-processing-units-accelerated indirect trajectory optimization methodology that uses the multiple shooting method and continuation is developed using the CUDA platform. The algorithm is designed to exploit the parallelism inherent in the indirect shooting method while maximizing computational efficiency. The resulting rapid optimal control framework enables the construction of high quality optimal trajectories that satisfy constraints and the necessary conditions of optimality. The performance of the framework is highlighted by construction of maximum terminal velocity trajectories for a hypothetical long-range weapon system. Various hypothetical mission scenarios that enforce different combinations of initial, terminal, interior point, and path constraints that demonstrate the rapid construction of complex trajectories are used to compare performance of the graphics-processing-units-accelerated solver to MATLAB (R)'s bvp4c. Trajectory problems of this kind were previously considered impractical to solve using indirect methods. The graphics-processing-units-accelerated solver is found to be two to four times faster than MATLAB (R)'s bvp4c for a small-dimensional system, even while running on graphics processing units hardware that is five years behind the state of the art.