Maneuver Detection via a Confidence Dominance Maneuver Indicator
Xingyu Zhou, Roberto Armellin, Laura Pirovano, Dong Qiao and Xiangyu Li · 2025
Read Maneuver Detection via a Confidence Dominance Maneuver Indicator by Xingyu Zhou, Roberto Armellin, Laura Pirovano, Dong Qiao and Xiangyu Li (2025) 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-2507-08234.txt.
Introduction
Xingyu Zhou, Roberto Armellin, Laura Pirovano, Dong Qiao, and Xiangyu Li spot spacecraft maneuvers in this two thousand twenty-five paper. To know when a satellite fires its thrusters is key for safety and for traffic that can be called ahead. Their way sets two confidence levels off against each other: one tied to the orbital state, one to how likely the observations are if no maneuver is taken. When the confidence of the observations tops the state's, the indicator flags a maneuver.
To work out the maximum likelihood took a recursive polynomial optimization method built on convex optimization and polynomial approximation. A joined version does away with hand-picking that level and reports a maneuver likelihood. It was set off against an optimal control distance metric and two detectors. The joined way hit detection accuracy up to ninety-nine point three three percent, at least ten points above rivals.
Catalog record
| CREATOR | Xingyu Zhou, Roberto Armellin, Laura Pirovano, Dong Qiao and Xiangyu Li |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 88 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-08234
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Maneuver Detection via a Confidence Dominance Maneuver Indicator Xingyu Zhou, Roberto Armellin, Laura Pirovano, Dong Qiao and Xiangyu Li Abstract Accurate and efficient maneuver detection is critical for ensuring the safety and predictability of spacecraft trajectories. This paper presents a novel maneuver detection approach based on comparing the confidence levels associated with the orbital state estimation and the observation likelihood. First, a confidence-dominance maneuver indicator (CDMI) is proposed by setting a confidence level for the state estimation and computing the maximum likelihood of the observation and its confidence level. The CDMI then flag a maneuver when the observation’s confidence level exceeds that of the state estimation, indicating that the observation is unlikely under the no-maneuver hypothesis while maintaining consistency with the prior state estimation confidence. To efficiently compute the maximum likelihood of the observation and obtain the CDMI, a recursive polynomial optimization method is developed, taking advantage of convex optimization and polynomial approximation. In addition, an integrated CDMI approach is developed to eliminate the need to manually select the state confidence level. The integrated CDMI approach maintains high detection accuracy while simultaneously providing an indication of maneuver likelihood, thereby enhancing robustness and practical applicability. The performance of the proposed CDMI-based maneuver detection approaches is evaluated against an optimal control distance metric and two mixture-based approaches. The simulation results demonstrate that the proposed integrated CDMI approach can achieve up to 99.33% detection accuracy, at least 10% higher than the competing methods, while substantially reducing computational costs. I Introduction Since space activities grow increasingly complex, the need for reliable maneuver detection techniques has become more critical [ 1 , 2 , 3 , 4 , 5 ] . Spacecraft trajectories are influenced not only by natural disturbances such as gravitational forces, solar radiation pressure, and atmospheric drag but also by intentional maneuvers for orbit adjustments, station-keeping, or evasive actions. If a spacecraft does not transmit its maneuver intentions, its trajectory becomes unpredictable [ 6 , 7 , 8 , 9 , 10 , 11 ] . While estimation is central to maintaining catalog accuracy and predicting conjunctions, it is accurate maneuver detection that enables these capabilities in the first place. Maneuver detection serves as a prerequisite for maintaining up-to-date orbital catalogs [ 12 , 13 , 14 ] , correlating new tracking data with known objects [ 15 , 16 , 17 , 18 ] , and forecasting future trajectories to assess potential conjunctions [ 19 , 20 , 21 , 22 ] . These functions are essential to ensuring timely collision avoidance and safeguarding space operations [ 23 , 24 , 25 , 26 ] . The maneuver detection problems can be broadly divided into two types: maneuver detection of two uncorrelated tracks (UCTs) [ 14 ] and maneuver detection of one UCT and a certain number of measurements after that UCT [ 27 , 28 ] . In this paper, a track refers to a known orbital state obtained by an orbit determination (OD) process based on a set of measurements, typically represented by a mean state vector and an associated covariance matrix. For the second type, the measurements given ( e.g. , only one angle measurement) are inadequate for an independent OD due to the limited observation capacity. Thus, the second track is not available. Conducting a two-line element (TLE) analysis is one of the most widely used ways to detect unknown maneuvers between two UCTs [ 29 , 30 , 31 ] . An alternative approach can be the optimal control distance metric (OCDM) proposed by Holzinger and Scheeres [ 32 , 33 , 34 ] . The OCDM formulates the maneuver detection problem as an uncertainty two-point boundary-value problem (UTPBVP) by identifying the energy-optimal control to connect two UCTs (the UTCs serve as boundary conditions), which can further estimate the maneuver, enhancing its applicability to object correlation and maneuver characterization [ 34 , 35 ] . Subsequent studies applied OCDM to detect anomalies caused by unknown dynamics [ 36 ] ( e.g. , solar radiation pressure and atmospheric drag) and maneuver detection in cislunar periodic orbits [ 37 ] . More recent advancements include combining OCDM with binary hypothesis testing [ 38 ] , solving for the OCDM using sequential convex optimization [ 14 ] , and simplifying OCDM by approximating continuous maneuvers with impulsive ones [ 39 ] . The second type of problem is more intractable since limited measurements are available, resulting in much larger uncertainty levels. A basic implementation for this type of maneuver detection involves propagating the last estimated state (within the UCT) forward to obtain predicted observations and then comparing them with sensors’ observed measurements [ 40 , 41 , 42 , 43 ] . The residual between the predicted and observed ones is then evaluated—if it exceeds a pre-defined threshold, a maneuver is detected. This approach is simple and easy to implement, but it heavily depends on the OD quality of the UCT and the measurement noise level. To improve maneuver detection accuracy, Kelecy and Jah utilized the difference between the solutions obtained by the extended Kalman filter (EKF) and the batch least squares (BLS) as a metric for the detection of continuous maneuvers, leveraging the fact that BLS may converge to incorrect solutions under unknown maneuvers [ 44 ] . Li et al. employed the Mahalanobis Distance (MD) as a metric based on measurement residual and covariance from the EKF process [ 45 ] . The OCDM can also be adopted here by replacing the boundary condition in the UTPBVP with the measurement constraints, which finally becomes a measurement residual boundary-value problem (MRBVP) [ 34 ] . Recently, Montilla et al. made significant progress in this field by proposing two mixture-based approaches (MBAs). The MBAs employ Gaussian mixture representation to approximate the uncertainties [ 28 ] , which is particularly useful when crucial uncertainties are quantified, for example, after a long-term propagation under highly nonlinear dynamics. The MBAs have promising maneuver detection accuracy, but at the cost of not being efficient. A key limitation of most of the aforementioned methods is their reliance on manually defined thresholds, which can be difficult to tune and may not generalize well across scenarios. To overcome this, recent years have seen the emergence of data-driven maneuver detection approaches, which leverage artificial intelligence techniques to learn decision boundaries directly from data. These methods typically frame maneuver detection as a classification problem—similar to facial or pattern recognition—by identifying latent patterns in measurement sequences. For example, Gaussian binary classification has been applied to detect maneuver-induced trajectory deviations [ 46 ] , and convolutional neural networks have been used to identify anomalies in TLE data [ 47 ] . While these methods often achieve high detection accuracy, many are restricted to fixed-length input sequences. The adoption of long short-term memory (LSTM) networks has improved flexibility by accommodating variable-length data and has shown promising preliminary results in maneuver detection tasks [ 5 , 48 ] . Nevertheless, such methods remain in early stages of development and face notable challenges, including limited availability of real-world labeled data and low interpretability, which hinder their practical reliability in operational systems [ 49 ] . This paper aims to handle the second type