Space Trajectory Design Using Artificial Intelligence

最近的研究表明，人工智能可以用来帮助设计空间轨迹。机器学习不需要解决复杂且耗时的优化问题，而是可以被训练成即时提供传输成本的估计。这可用于初步任务设计，或在需要快速计算大量轨迹的多目标任务中使用。目前，这依赖于一个预先计算的最优转移数据库，可以用来训练人工神经网络。这个博士项目将深入研究机器学习和强化学习，用于空间轨迹设计，主要目标是创建一个框架，在该框架中，代理将迭代地自动学习解的最优性，从而根本不需要执行任何优化。它还将扩大到包括选择可能的任务目标(物体或轨道)和航天器本身的系统设计--轨道的设计往往取决于推进系统，反之亦然，推进系统的选择取决于轨道。综合飞行任务系统设计将通过机器学习来解决。应用包括多体飞行任务，其中需要评估数千到数百万种可能的轨迹，例如星际多小行星交会地点，以及多个活跃的碎片清除任务。然而，快速轨迹优化也用于任务设计的初步阶段，其中卫星系统(特别是推进系统)不会冻结，应与轨迹本身一起进行选择和优化。理想的候选人应具有计算机科学、人工智能、机器学习背景，对数学建模和空间系统有浓厚兴趣。或者，应聘者可以有空间系统工程背景，对人工智能有浓厚的兴趣(最好是有经验)。

Recent research has shown that artificial intelligence can be used to aid the design of space trajectories. Instead of solving a complex and time-consuming optimisation problem, machine learning can be trained to provide an estimate of the cost of a transfer instantaneously. This can be used in preliminary mission design, or in multi-target missions, where the fast computation of a high number of trajectories is necessary. Currently this relies on a database of pre-computed optimal transfers, that can be used to train an artificial neural network.This PhD project will delve into machine learning and reinforcement learning for space trajectory design, with the main aim to create a framework where an agent would automatically learn about optimality of solutions iteratively, taking away the need to perform any optimisation at all. It will also expand to include the selection of possible mission targets (bodies or orbits), and systems design of the spacecraft itself - very often, the design of the trajectory relies on propulsion system, and vice-versa, the selection of propulsion system relies on the trajectory. Integrated mission-system design will be tackled with machine learning.Applications include multi-body missions, where thousands to millions of possible trajectories have to be evaluated, such as interplanetary multi-asteroid rendezvouses, and multiple active debris removal missions. However, fast trajectory optimisation is also used in the preliminary phases of the mission design, where the satellite systems (and propulsion system in particular) are not frozen, and shall be selected and optimised together with the trajectory itself.The ideal candidate will have a background in computer science, artificial intelligence, machine learning, with a strong interest for mathematical modelling and space systems. Alternatively, the candidate can have a background in space systems engineering, with a strong interest (and preferably experience) in artificial intelligence.