Adaptive Satellite State Estimation Using Liquid Neural Networks
- Paper ID
93893
- author
- company
Instituto Tecnológico Autónomo de México
- country
Mexico
- year
2025
- abstract
The increasing deployment of satellites for earth and marine observation has increased the demand for accurate orbital prediction models. This task can be achieved by physical models with different degrees of accuracy, however these models present limitations in the face of active propulsion maneuvers and unmodeled disturbances. This work proposes a novel solution based on Liquid Neural Networks (LLN) to predict and adjust satellite orbits using historical observations of their position. LLNs have the ability to adapt to new information after training, so the model will be able to adjust to new measurements in real time when needed, allowing a more robust and adaptive orbit prediction in real operational environments.