State Forecasting in an Estimation Framework with Surrogate Sensor Modeling

Sriram Narayanan, Mohamed Naveed Gul Mohamed, Ishan Paranjape, Indranil Nayak, Suman Chakravorty, and Mrinal Kumar · 2026

Read State Forecasting in an Estimation Framework with Surrogate Sensor Modeling by Sriram Narayanan, Mohamed Naveed Gul Mohamed, Ishan Paranjape, Indranil Nayak, Suman Chakravorty, and Mrinal Kumar (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-19442.txt.

Introduction

Narayanan and colleagues address state forecasting with surrogate sensor models for tracking resident space objects. Data-driven methods like system identification and machine learning have transformed analysis of complex physical systems through the inverse problem approach. Yet extracting useful information from partial measurements remains a major hurdle, especially in aerospace, where data are scarce and sensor limits constrain tracking. The paper develops surrogate sensor models to fill gaps left by limited range and bearing data.

Catalog record

CREATORSriram Narayanan, Mohamed Naveed Gul Mohamed, Ishan Paranjape, Indranil Nayak, Suman Chakravorty, and Mrinal Kumar
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE61318
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.19442

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

State Forecasting in an Estimation Framework with Surrogate Sensor Modeling Sriram Narayanan, Mohamed Naveed Gul Mohamed, Ishan Paranjape, Indranil Nayak, Suman Chakravorty, and Mrinal Kumar Abstract In recent years, computational power and data availability breakthroughs have revolutionized our ability to analyze complex physical systems through the inverse problem approach. Data-driven techniques like system identification and machine learning play an important role in this field, allowing us to gain insights into previously inaccessible phenomena. However, a major hurdle remains: How can meaningful information from partial measurements be extracted? In the aerospace domain, the challenge of state estimation is particularly pronounced due to the limited availability of observational data and the constraints imposed by sensor capabilities for tracking resident space objects (RSOs). To address these limitations, advanced compensation methodologies are required. Currently, range and bearing measurements obtained from radar and optical systems constitute the primary observational tools in the space situational awareness (SSA) community. In this work, we propose a novel framework that integrates a simplified reference dynamics model with a data-driven surrogate measurement model. This fusion process leverages the strengths of both models to estimate complex dynamical behaviors under conditions of partial observability. Extensive numerical experiments were conducted across multiple datasets to validate the proposed framework. The results demonstrate its efficacy in accurately reconstructing system dynamics from incomplete measurement data. Furthermore, to ensure the robustness of the framework, an initial consistency analysis of the surrogate modeling approach is presented. By addressing the current challenges and refining the integration of data-driven techniques with traditional physics-based modeling, this framework aims to advance state estimation methodologies in the aerospace sector. \PaperNumber 25-577 1 Introduction Accurate and efficient orbit prediction is increasingly crucial for improved SSA ( SSA ) [ 2 ] . The ability to precisely forecast the trajectory of RSO ( RSO ) is essential for conjunction analysis and collision avoidance. Traditional orbit prediction methods rely primarily on physics-based models, often lacking robustness and effectiveness in a dynamic space environment. These challenges arise due to limited information on space conditions, RSO characteristics, and the intent of maneuvering objects. Recent advancements in ML ( ML ) provide promising solutions to address these limitations and have been explored in aerospace applications [ 1 , 13 ] . For example, Girimonte and Izzo reviewed the challenges of applying artificial intelligence in space systems, including distributed intelligence, enhanced situational awareness, and decision support for spacecraft design [ 4 ] . However, a fundamental question remains: can ML or SID ( SID ) methods be used to improve forecasting reliably and consistently? There is work in the literature where Gaussian process models have been combined with conventional estimation methods such as the EKF ( EKF ) to generate more accurate state estimates [ 12 ] . In our work, we use the DMD ( DMD ) algorithm described in Ref. [ 8 ] to build a surrogate model and fuse the predictions in a standard EKF framework. Since the role of the surrogate is to forecast measurements alone, there is no need to apply any physical or pseudo-physical meaning to its output to use it with the EKF . It can simply be used with the measurement function. Orbit information fusion in SSA has particular challenges compared to other fields: the orbit prediction duration is usually much longer than, for instance, a robotic control problem and can vary dramatically depending on the availability of observation stations. Therefore, orbit forecasting is carried out by carefully fusing the surrogate predictions with estimates from a placeholder dynamical system. Building upon the methodologies discussed in Refs. [ 9 ] [ 10 ] , this work introduces a framework for forecasting with partial observations. The previously proposed iterative scheme employed a two-step process where the surrogate model and the updated state estimates were inferred from measurement data. In this work, we refine this approach by decoupling the process: the data-driven model is used solely for learning the measurements, and we introduce a reference dynamics model as a placeholder to describe the system’s approximate behavior. The reference model used is typically a simpler representation and does not account for the perturbative forces that RSO s experience on orbit. We examine a few different scenarios and provide insight into when the framework works best and when it does not. In this improved framework, the surrogate model extrapolates measurements for instances when direct observations are unavailable. Concurrently, the reference dynamics model offers approximate state estimates at that time instant. These two components are then integrated within a conventional fusion framework derived from estimation theory to produce a more accurate and refined state estimate. This decoupled approach not only simplifies the modeling process but also enhances the robustness and accuracy of the forecasting system. The proposed framework leverages the strengths of both surrogate models and reference dynamics models to address the challenges posed by partial observations. The surrogate model, trained on available measurement data, is adept at predicting measurements for time instances where direct observations are missing. This capability is crucial in scenarios where sensor data is intermittent or incomplete. The fusion process combines the surrogate model’s measurement predictions with the reference dynamics model’s state estimates using techniques from estimation theory, such as the EKF . This approach not only enhances the accuracy of the state estimates but also improves the robustness of the forecasting system against uncertainties and noise in the measurement data. While this work focuses on the EKF , the framework is designed to be adaptable to other filtering algorithms, ensuring flexibility and applicability across various scenarios. The work is structured as follows: Sec. 2 introduces the proposed approach, detailing the iterative scheme, the roles of the surrogate and reference dynamics models, and the fusion process. Sec. 3 presents the results of applying the proposed framework to various case studies, including the simple pendulum and various orbital mechanics scenarios, covering applications in the Cislunar regime as well. Sec. 4 provides a consistency analysis of the proposed method, and finally, Sec. 5 summarizes the key contributions and implications of this work. 2 Proposed Framework for Forecasting via Filtering \HyColor@XZeroOneThreeFour \pc @goptd@color \pc@hyenc@color pdfcommentcolor \HyColor@XZeroOneThreeFour \pc @goptd@fontcolor \pc@hyenc@fontcolor pdfcommentcolor \HyColor@XZeroOneThreeFour \pc @goptd@icolor \pc@hyenc@icolor pdfcommentcolor \pdfmark [ ]pdfmark=/ANN,Subtype=/Widget,Raw=/TU (This figure illustrates the iterative scheme for forecasting with partial observations in an estimation framework. It includes colored zones for measurement collection, surrogate model training, initial condition determination, placeholder dynamics propagation, and state estimation fusion. \textCR ( \pc @goptd@deadline)) /T (tooltip zref@0) /C [ ] /FT/Btn /F 768 /Ff 65536 /H/N /BS ¡¡ /W 0 ¿¿ Figure 1: Iterative Scheme for Forecasting with Partial Observations in an Estimation Framework. In this section, we detail the proposed iterative framework for forecasting with partial observations. Fig. ( 1 ) illustrates this approach. The flowchart is divided into different zones, each represented by a distinct color for clarity. The blue zone represents the batch of measurements \mathbf{z}(1:k) , which are accrued and used to train the surrogate data-driven model. Several different types of measurements are considered, ranging from the state vectors directly to nonlinear transformations of them, such as range, azimuth, and elevations in the context of SSA , represented by \mathbf{z}(k)=\mathbf{h}(\mathbf{x}(k)) , where \mathbf{h} is the measurement function. In this work,

Read the complete text