Optimization of Multiple-Rendezvous Low-Thrust Missions on General-Purpose Graphics Processing Units

A massively parallel method for the identification of optimal sequences of targets in multiple-rendezvous low-thrust missions is presented. Given a list of possible targets, a global search of sequences compatible with the mission requirements is performed. To estimate the feasibility of each transfer, a heuristic model based on Lambert's transfers is evaluated in parallel for each target, making use of commonly available general-purpose graphics processing units such as the Nvidia Tesla cards. The resulting sequences are ranked by user-specified criteria such as length or fuel consumption. The resulting preliminary sequences are then optimized to a full low-thrust trajectory using classical methods for each leg. The performance of the method is discussed as a function of various parameters of the algorithm. The efficiency of the general-purpose graphics processing unit implementation is demonstrated by comparing it with a traditional CPU-based branch-and-bound method. Finally, the algorithm is used to compute asteroid sequences used in a solution submitted to the seventh edition of the Global Trajectory Optimization Competition.