Abstract's details
Towards a Standardised Satellite Maneuver Log Dataset: Leveraging IDS Data for Space Domain Awareness Benchmarking and Machine Learning Applications
Event: 2026 IDS Workshop
Session: Advances in DORIS: Data, Antennas, and Modeling
Presentation type: Oral
A persistent challenge in space domain awareness (SDA) is the apparent absence of a standardised ground-truth dataset against which satellite maneuver detection and characterisation algorithms can be evaluated. Understanding a satellite's pattern of life - from routine station-keeping to conjunction avoidance and space weather responses - is central to catalogue maintenance and conjunction assessment. Yet, algorithm development is constrained by the lack of a large, shared, verified reference of maneuver events. This presentation reports on an ongoing effort to compile such a dataset from multiple operator and agency sources, and leverages the historical maneuver logs maintained by the International DORIS Service (IDS) as a well-structured reference to guide the multi-source compilation and standardisation.
The IDS maneuver logs contain high-quality, timestamped records for listed satellites such as those in the Sentinel series, spanning decades of operations. While these represent the most consistent and well-structured source identified to date, the broader effort also incorporates maneuver records obtained from other agencies and mission operators. The aim is to merge these heterogeneous sources into a unified catalogue, using the IDS format and coverage as the reference standard given its consistency, temporal depth, and traceability to the POD chain. Determining the optimal format for the unified catalogue will be an important part of this effort, with the IDS approach providing a robust starting point due to its structured, well-documented, and temporally extensive records. Initial data from other sources have already been acquired to inform the compilation process.
The compiled dataset is being developed to serve three core functions. First, as ground truth for algorithm validation: in a current calibration project by UCL in collaboration with the UK Space Agency, a maneuver detection pipeline has been applied to both publicly available Two-Line Element (TLE) orbital element data and improved calibrated position data produced by the project, with the maneuver logs providing a definitive reference against which detection performance is measured using standard classification metrics. Second, and of particular importance, as a resource for Machine Learning (ML) integration: verified maneuver records enable automatic labelling of thrust periods in historical orbit data, which is essential for training supervised and semi-supervised models, for excluding corrupted samples from training sets, and for segmenting orbit arcs into maneuver-free intervals suited to learning orbital error patterns. The scarcity of labelled maneuver data is recognised as a bottleneck for ML-based SDA methods, and a curated, multi-satellite dataset of this kind would directly address that gap. The presentation will include examples of how classical and ML pipelines benefit from access to verified maneuver records, and a comparison of detection performance with and without the use of historical log data as a training and validation resource. Third, as a cross-method benchmark: the logs enable a controlled comparison of different detection approaches and input data sources on equal terms, isolating the contribution of each component.
A key objective is to map the compiled data to international standards, particularly the CCSDS Orbit Data Messages Blue Book (CCSDS 502.0-B-3), which defines maneuver parameter blocks in both the OPM and OCM formats. The presentation will discuss the current state of the compiled database, the planned catalogue structure, and ongoing work to align fields such as delta-v magnitude, thrust direction, and maneuver type across sources. The presentation will also discuss challenges, limitations, and practical considerations encountered in compiling and aligning multi-source maneuver data, providing transparency and guidance for future improvements.
This contribution is positioned at the interface between the geodesy and SDA communities, and feedback is sought on feasibility, interest, and potential collaboration in developing this resource as a shared asset for both fields.
Keywords: Space Domain Awareness, Dataset, Satellites, Maneuvers, Machine Learning
Type of participation: In person (however open to remote participation if required)
Back to the list of abstractThe IDS maneuver logs contain high-quality, timestamped records for listed satellites such as those in the Sentinel series, spanning decades of operations. While these represent the most consistent and well-structured source identified to date, the broader effort also incorporates maneuver records obtained from other agencies and mission operators. The aim is to merge these heterogeneous sources into a unified catalogue, using the IDS format and coverage as the reference standard given its consistency, temporal depth, and traceability to the POD chain. Determining the optimal format for the unified catalogue will be an important part of this effort, with the IDS approach providing a robust starting point due to its structured, well-documented, and temporally extensive records. Initial data from other sources have already been acquired to inform the compilation process.
The compiled dataset is being developed to serve three core functions. First, as ground truth for algorithm validation: in a current calibration project by UCL in collaboration with the UK Space Agency, a maneuver detection pipeline has been applied to both publicly available Two-Line Element (TLE) orbital element data and improved calibrated position data produced by the project, with the maneuver logs providing a definitive reference against which detection performance is measured using standard classification metrics. Second, and of particular importance, as a resource for Machine Learning (ML) integration: verified maneuver records enable automatic labelling of thrust periods in historical orbit data, which is essential for training supervised and semi-supervised models, for excluding corrupted samples from training sets, and for segmenting orbit arcs into maneuver-free intervals suited to learning orbital error patterns. The scarcity of labelled maneuver data is recognised as a bottleneck for ML-based SDA methods, and a curated, multi-satellite dataset of this kind would directly address that gap. The presentation will include examples of how classical and ML pipelines benefit from access to verified maneuver records, and a comparison of detection performance with and without the use of historical log data as a training and validation resource. Third, as a cross-method benchmark: the logs enable a controlled comparison of different detection approaches and input data sources on equal terms, isolating the contribution of each component.
A key objective is to map the compiled data to international standards, particularly the CCSDS Orbit Data Messages Blue Book (CCSDS 502.0-B-3), which defines maneuver parameter blocks in both the OPM and OCM formats. The presentation will discuss the current state of the compiled database, the planned catalogue structure, and ongoing work to align fields such as delta-v magnitude, thrust direction, and maneuver type across sources. The presentation will also discuss challenges, limitations, and practical considerations encountered in compiling and aligning multi-source maneuver data, providing transparency and guidance for future improvements.
This contribution is positioned at the interface between the geodesy and SDA communities, and feedback is sought on feasibility, interest, and potential collaboration in developing this resource as a shared asset for both fields.
Keywords: Space Domain Awareness, Dataset, Satellites, Maneuvers, Machine Learning
Type of participation: In person (however open to remote participation if required)