Future space missions increasingly require autonomous navigation and control capabilities to operate in uncertain and partially known dynamical environments. In such scenarios, the availability of accurate analytical models cannot always be guaranteed, motivating the development of adaptive approaches capable of learning system dynamics directly from data. This thesis presents an adaptive framework for orbital station keeping based on the integration of Sparse Identification of Nonlinear Dynamics (SINDy) and Nonlinear Model Predictive Control (NMPC). The proposed methodology combines a physics-informed model discovery procedure with an online model adaptation mechanism, enabling the spacecraft to continuously refine its internal representation of the surrounding environment during mission operations. A Batch-SINDy approach is first employed to identify the dominant perturbative contributions from commissioning data, while a Physics-Informed Recursive-SINDy formulation is used to update the model online as new information becomes available. The identified model is then integrated into an NMPC controller to generate optimal control actions while accounting for the evolving dynamics. The obtained results demonstrate the feasibility of combining sparse model discovery, online adaptation, and predictive control within a unified framework. The proposed approach improves the capability of the spacecraft to operate autonomously under uncertain dynamical conditions while maintaining a physically interpretable representation of the environment.
Le future missioni spaziali richiedono sempre più capacità di navigazione e controllo autonome per operare in ambienti dinamici incerti e solo parzialmente conosciuti. In tali scenari, la disponibilità di modelli analitici accurati non può essere sempre garantita, rendendo necessario lo sviluppo di approcci adattativi in grado di apprendere la dinamica del sistema direttamente dai dati. Questa tesi presenta un framework adattativo per il mantenimento orbitale (station keeping) basato sull'integrazione di Sparse Identification of Nonlinear Dynamics (SINDy) e Nonlinear Model Predictive Control (NMPC). La metodologia proposta combina una procedura di identificazione fisicamente informata con un meccanismo di aggiornamento online del modello, consentendo al veicolo spaziale di affinare continuamente la propria rappresentazione dell'ambiente circostante durante le operazioni di missione. In una prima fase, Batch-SINDy viene utilizzato per identificare le principali perturbazioni dinamiche a partire dai dati raccolti durante il commissioning, mentre una formulazione Physics-Informed Recursive-SINDy permette di aggiornare il modello durante il volo man mano che nuove informazioni diventano disponibili. Il modello identificato viene successivamente integrato all'interno di un controllore predittivo non lineare per generare azioni di controllo ottimali tenendo conto dell'evoluzione della dinamica del sistema. I risultati ottenuti dimostrano la fattibilità dell'integrazione tra identificazione sparsa, adattamento online e controllo predittivo in un unico framework, migliorando l'autonomia operativa del veicolo spaziale e preservando al contempo l'interpretabilità fisica del modello dinamico identificato.
Adaptive model predictive control for autonomous orbital station keeping via sparse identification of nonlinear dynamics
BERGAMASCHI, DANIELE
2025/2026
Abstract
Future space missions increasingly require autonomous navigation and control capabilities to operate in uncertain and partially known dynamical environments. In such scenarios, the availability of accurate analytical models cannot always be guaranteed, motivating the development of adaptive approaches capable of learning system dynamics directly from data. This thesis presents an adaptive framework for orbital station keeping based on the integration of Sparse Identification of Nonlinear Dynamics (SINDy) and Nonlinear Model Predictive Control (NMPC). The proposed methodology combines a physics-informed model discovery procedure with an online model adaptation mechanism, enabling the spacecraft to continuously refine its internal representation of the surrounding environment during mission operations. A Batch-SINDy approach is first employed to identify the dominant perturbative contributions from commissioning data, while a Physics-Informed Recursive-SINDy formulation is used to update the model online as new information becomes available. The identified model is then integrated into an NMPC controller to generate optimal control actions while accounting for the evolving dynamics. The obtained results demonstrate the feasibility of combining sparse model discovery, online adaptation, and predictive control within a unified framework. The proposed approach improves the capability of the spacecraft to operate autonomously under uncertain dynamical conditions while maintaining a physically interpretable representation of the environment.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_07_Bergamaschi_Tesi.pdf
accessibile in internet per tutti
Descrizione: Testo Tesi
Dimensione
5.24 MB
Formato
Adobe PDF
|
5.24 MB | Adobe PDF | Visualizza/Apri |
|
2026_07_Bergamaschi_Executive_Summary.pdf
accessibile in internet per tutti
Descrizione: Executive Summary
Dimensione
774 kB
Formato
Adobe PDF
|
774 kB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/261098