Delay-Adaptive Federated Filtering with Online Model Calibration Dataset

Precise orbit determination for multi-spacecraft deep-space missions faces challenges including long communication delays, sparse tracking, dynamic model uncertainties, and inefficient data fusion. Presenting a hybrid estimation architecture, this study integrates onboard autonomous navigation with ground-based batch processing of delayed measurements. The framework makes three key contributions: (1) a delay-aware fusion paradigm that dynamically weights space- and ground-based observations according to real-time Earth–Mars latency (4–22 min); (2) a model-informed online calibration framework that jointly estimates and compensates dominant dynamic error sources, reducing model uncertainty by 60%; (3) a lightweight hierarchical architecture that balances accuracy and efficiency for resource-constrained “one-master-multiple-slave” formations. Validated through Tianwen-1 mission-data replay and simulated Mars sample-return scenarios, the method achieves absolute and relative orbit determination accuracies of 14.2 cm and 9.8 cm, respectively—an improvement of >50% over traditional centralized filters and a 30% enhancement over existing federated approaches. It maintains 20.3 cm accuracy during 10-minute ground-link outages and shows robustness to initial errors >1000 m and significant model uncertainties. This study presents a robust framework applicable to future multi-agent deep-space missions such as Mars sample return, asteroid reconnaissance, and cislunar navigation constellations. This dataset is used to validate the orbit determination accuracy of the TDA-FKF algorithm, containing simulation data, HIL test data, and model calibration data.