EAPSI: A Machine Learning Approach to Lunar Spacecraft Trajectory Optimization

这项研究将在日本宇宙航空研究开发机构（JAXA）与川胜康弘博士合作，为即将到来的月球小型航天器使命EQUULEUS研究创新的航天器轨道优化方法，该任务将于2018年底由NASA的太空发射系统（SLS）火箭发射。这些发现可能使曾经被认为是不可能的太空任务成为可能，从而为科学收集带来更多奇异和令人兴奋的机会。这一使命还将通过对地球等离子体的成像和测量其分布，使科学家进一步了解地球周围的辐射环境，这可能为保护人类和电子设备在长途太空旅行中免受辐射损害提供重要的见解。这项研究探索了变革性的概念，结合了机器学习和轨迹优化，这两个主题结合在一起，在很大程度上尚未探索。与日本宇宙航空研究开发机构的合作是进一步开展这项研究的独特机会，因为它是公认的轨道设计领导者，并且在小推力和低能量航天器方面拥有丰富的使命经验。航天器将通过地月轨道将自身插入地月系统L2拉格朗日点附近的稳定轨道，通过低能路径利用地月系统有效势能的拓扑稳定性。通过传统的轨迹优化方法（即直接方法和间接方法）将生成大量的最佳控制轨迹数据集。一旦生成状态-控制对的数据集，将在该数据集上训练人工神经网络（ANN）。通过训练，人工神经网络开发了一个可以实时实施的空间控制策略。航天器在任何时候都会感知到它的环境并相应地采取行动（即节流推进器）。这种控制方法类似于生物在自然界中的行为。就像一只简单的家蝇能够找到食物来源，实时做出决定一样，航天器也应该能够实现同样的目标。该奖项是东亚和太平洋夏季研究所计划的一部分，由NSF和日本科学促进会共同资助，资助一名美国研究生的夏季研究。

This research will investigate innovative spacecraft trajectory optimization methods at the Japanese Aerospace Exploration Agency (JAXA) in collaboration with Dr. Yasuhiro Kawakatsu for the upcoming lunar small-spacecraft mission, EQUULEUS, which will be launched aboard NASA's Space Launch System (SLS) rocket at the end of 2018. The findings may enable space missions that were once thought to be impossible, leading to more exotic and exciting opportunities for science collection. This mission will also further scientists understanding of the radiation environment surrounding Earth by imaging its plasmasphere and measuring its distribution, which may provide important insight for protecting both humans and electronics from radiation damage during long space journeys. This research explores transformative concepts, combining machine learning and trajectory optimization, two subjects which, in combination, have been largely unexplored. Collaboration with JAXA is a unique opportunity to further this research, as it is a recognized trajectory design leader and has extensive mission experience with low-thrust and low-energy spacecraft. The spacecraft will insert itself into a stable orbit about the L2 Lagrange point of the Earth-Moon system through a cislunar trajectory, exploiting the topological stability of the Earth-Moon system's effective potential through low-energy pathways. A large data set of optimal control trajectories will be generated through conventional trajectory optimization methods (i.e. direct methods and indirect methods). Once the data set of state-control pairs is generated, an artificial neural network (ANN) will be trained on the data set. Through training, the ANN develops a spatial control policy that can be implemented in real-time. The spacecraft, at any moment in time, will perceive its environment and take actions (i.e. throttle its thrusters) accordingly. This control method is analogous to how organisms behave in nature. Just as a simple house fly is able to navigate to its food source, making decisions in real-time, a spacecraft should be able to do the same when trying to achieve its objective.This award, under the East Asia and Pacific Summer Institutes program, supports summer research by a U.S. graduate student and is jointly funded by NSF and the Japan Society for the Promotion of Science.