Co-State Based Data Fusion and Risk Aware Filtering for Spacecraft Navigation and Hazard Prediction
Co-State Based Data Fusion and Risk Aware Filtering for Spacecraft Navigation and Hazard Prediction
Authors: Surya Ratna Prakash D, Soumyendu Raha Date: 2026-04-22 Paper ID: openalex:2604.20485
Summary
This paper introduces a co-state-based framework for spacecraft navigation and hazard prediction, leveraging differential algebraic Lagrange multipliers to monitor system consistency. The architecture employs these co-states to learn a continuous-time Markov generator, which facilitates probabilistic risk assessment via mean first-passage time and mode transitions. By unifying geometric projection and stochastic inference, the approach provides interpretable, data-driven early-warning signals for failure detection without relying on predefined fault models. Evaluation on lunar powered-descent telemetry demonstrates that the method detects internal model inconsistencies significantly earlier than traditional statistical methods like the EKF.
Key Contributions
- Introduces a co-state fusion framework that enforces measurement dynamics compatibility at the differential level as a proxy for geometric inconsistency.
- Enables intrinsic probabilistic risk forecasting by learning a continuous-time Markov generator from co-state and innovation trajectories.
- Demonstrates superior early-warning capabilities compared to Extended Kalman Filters (EKF) on lunar powered-descent telemetry without requiring labeled failure data.
Open Questions & Future Work
Key Concepts
- co-state-based-fusion-framework: A fusion framework that uses co-state Lagrange multipliers to enforce measurement dynamics compatibility for diagnostic monitoring and anomaly detection.
Archivist Review
The co-state fusion framework is approved as it offers a novel, physically interpretable approach to anomaly detection that avoids the reliance on labeled failure data. The open question regarding risk-aware control integration is approved for its significance in advancing diagnostic systems toward autonomous decision-making. The real-time optimization question was rejected as it pertains to general deployment engineering rather than a core scientific bottleneck.
Approved Concepts
- Co-state based fusion framework: Provides a novel, physically grounded mechanism for detecting inconsistencies in dynamical systems by using Lagrange multipliers as diagnostic signals.
Approved Open Questions
- Active Risk-Aware Control Integration: Bridging the gap between diagnostic monitoring and autonomous control is essential for deploying safety-critical navigation systems in uncertain environments.
Rejected Candidates
- [open_question] Real-Time Embedded Implementation Optimization (
embedded-navigation-real-time-optimization) - generic: The challenge is a standard engineering implementation requirement for almost all embedded navigation algorithms rather than a fundamental research bottleneck specific to the proposed method.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2604.20485
- paper_id
- 2604.20485
- paper_source
- openalex
- domain
- time-series
- tags
- time-seriesstochastic-inferenceanomaly-detectionnavigation
- architectures
-
- datasets
-
- concept_slugs
- co-state-based-fusion-framework
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-04-25T05:38:24Z