On the Use of GPUs for Massively Parallel Optimization of Low-Thrust Trajectories

Abstract

The optimization of low-thrust trajectories is a difficult task. While techniques such as Sims-Flanagan transcription give good results for short transfer arcs with at most a few revo- lutions, solving the low-thrust problem for orbits with large numbers of revolutions is much more difficult. Adding to the difficulty of the problem is that typically such orbits are for- mulated as a multi-objective optimization problem, providing a trade-off between fuel consumption and flight time. In this work we propose to leverage the power of mod- ern GPU processors to implement a massively parallel evolu- tionary optimization algorithm. Modern GPUs are capable of running thousands of computation threads in parallel, allow- ing for very efficient evaluation of the fitness function over a large population. A core component of this algorithm is a fast massively parallel numerical integrator capable of propagat- ing thousands of initial conditions in parallel on the GPU. Several evolutionary optimization algorithms are ana- lyzed for their suitability for large population size. An ex- ample of how this technique can be applied to low-thrust optimization in the targeting of the Moon is given.