RII Track-4:@NSF: Surrogate-based Optimal Atmospheric Entry Guidance using High-fidelity Simulation Data

研究基础设施改进Track-4 EPSCoR研究人员(RII Track-4)项目将为爱荷华州立大学的一名助理教授提供奖学金，并为一名研究生提供培训。这项工作将与NASA艾姆斯研究中心的研究人员合作进行。对于行星探测，航天器必须通过进入大气层和动力下降阶段，才能安全减速并准确着陆。追溯到阿波罗任务时代，大气进入制导(AEG)控制航天器的大气阻力来实现这些目标。研究人员一直在研究最优的AEG，通过以最小的速度轨迹终止进入阶段，在随后的动力下降过程中最大限度地节省燃料。这些最优的AEG方法依赖于与实际进入环境存在不确定差异的理想动态模型。因此，开发和利用了更先进和更复杂的计算建模和仿真技术来最大限度地减少这些差异。尽管蒙特卡罗模拟有很多优点，但其复杂性的增加使得它不能量化建模的不确定性。此外，进入飞行器的车载计算机功能不够强大，无法运行复杂模型的最佳AEG。为了解决这些局限性，本研究旨在创建一种基于代理的最优制导系统，该系统基于来自复杂模拟的高保真数据进行训练。所提出的制导方法通过减少计算负担、节省航天器燃料并实现建模不确定性量化来提高空间探索的安全性和效率。显然，需要一种新的最优AEG来降低计算成本并实现建模不确定性量化。为了满足这一需求，将开发一个基于代理的AEG系统，该系统使用来自高级进入系统建模(ESM)的高保真仿真数据进行训练。对于开发来说，准备精确和计算高效的训练数据以有效地封装大气进入的核心是至关重要的。拟议的研究将确定影响AEG性能的主要变量，并使用NASA的进入模拟工具生成所需的训练数据。还将探索用于训练的各种替代模型，如高斯过程回归和广义加法模型。最终目标是使用训练有素的代理人建立车载最优AEG框架。这个框架可以结合各种反馈控制算法，以帮助在地球、火星、金星和土卫六上执行行星进入任务。虽然以前的研究主要集中于将代理应用于子部件建模，如空气密度、流体和气动热动力学，但这种方法的目标是应用于最优制导，并将加快在嵌入式平台上实现的计算速度。为了减少计算量和训练时间，本研究提出了一种简化进入制导轮廓的方法，该方法还可以降低训练数据的维度。该项目的成功将为将拟议的基于代理的技术扩展到其他空间应用铺平道路，例如航天器轨道或姿态制导，并为将传统的空间制导、导航和控制(GNC)方法扩展到基于数据的学习技术做出重大贡献。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

This Research Infrastructure Improvement Track-4 EPSCoR Research Fellows (RII Track-4) project would provide a fellowship to an Assistant professor and training for a graduate student at Iowa State University. This work would be conducted in collaboration with researchers at the NASA Ames Research Center. For planetary exploration, spacecraft must pass through atmospheric entry and powered descent stages to safely decelerate and accurately land. Dating back to the Apollo mission era, Atmospheric Entry Guidance (AEG) controls the atmospheric drag of a spacecraft to achieve these objectives. Researchers have been working on optimal AEG to maximize fuel savings during the subsequent powered descent by terminating the entry phase with a minimum velocity trajectory. These optimal AEG methods have relied on ideal dynamic models with uncertain differences from the actual entry environment. Therefore, more advanced and complex computational modeling and simulation technologies have been developed and utilized to minimize these discrepancies. Despite the advantages of Monte Carlo simulation, the increased complexity makes it an impractical method to quantify modeling uncertainty. In addition, the entry vehicle's onboard computer is not powerful enough to run optimal AEG with a complex model. To address these limitations, this research aims to create a surrogate-based optimal guidance system, trained on high-fidelity data from complex simulations. The proposed guidance method enhances safety and efficiency in space exploration by reducing computational burden, saving spacecraft fuel, and enabling modeling uncertainty quantification.The need for a new optimal AEG that reduces computational costs and enables modeling uncertainty quantification is evident. To satisfy this need, a surrogate-based AEG system, trained using high-fidelity simulation data from advanced Entry System Modeling (ESM), will be developed. For the development, preparing precise and computationally efficient training data that effectively encapsulates the core of atmospheric entry is crucial. The proposed research will identify the dominant variables influencing AEG performance and generate the required training data using NASA's entry simulation tool. Various surrogate models for training, such as Gaussian Process Regression and Generalized Additive Model, will also be explored. The ultimate objective is establishing an onboard optimal AEG framework using a trained surrogate. This framework can incorporate various feedback control algorithms to aid in planetary entry missions on Earth, Mars, Venus, and Titan. While prior research has focused on applying surrogates for subcomponent modeling, such as air density and fluid and aerothermal dynamics, this approach targets application to optimal guidance and will accelerate calculation speed for implementation on embedded platforms. To reduce computations and training time, this research proposes a simplification method for the entry guidance profile that can also reduce the dimension of the training data. The success of this project will pave the way for extending the proposed surrogate-based technique to other space applications, such as spacecraft orbit or attitude guidance, and contribute significantly to extending the traditional space Guidance, Navigation, and Control (GNC) approach to data-based learning techniques.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.