Accurate estimation of the position and orientation of a spacecraft during proximity operations—such as rendezvous, docking, on-orbit servicing (OOS), and active debris removal (ADR)—is critical to ensuring mission success and safety. Tradi- tional visual navigation methods based on hand-engineered fea- ture matching often struggle with robustness and generalization, while existing deep learning approaches face limitations due to heuristic hyperparameter tuning and limited training data. In this work, a novel convolutional neural network (CNN)-based architecture for monocular pose estimation of non-cooperative spacecraft is proposed, specifically designed to improve robust- ness across diverse operational scenarios. The model is trained on a high-fidelity synthetic dataset comprising approximately 25,000 images, simulating realistic proximity conditions with variations in lighting, background textures, and spacecraft geometries. To assess its performance, an extensive benchmarking study is conducted against representative State-of-the-Art methods using standardized evaluation metrics and controlled test conditions. The results demonstrate the competitive performance of the proposed method and provide critical insights into the factors affecting pose estimation accuracy in realistic spaceborne appli- cations
Deep Learning-Optimized Monocular Navigation for Autonomous Rendezvous and Proximity Maneuvers in Small Satellite Missions / Lovaglio, Lucrezia; Stesina, Fabrizio. - (2025), pp. 459-464. (Intervento presentato al convegno 2025 IEEE 12th International Workshop on Metrology for AeroSpace tenutosi a Napoli (Ita) nel 18-20 June, 2025) [10.1109/MetroAeroSpace64938.2025.11114522].
Deep Learning-Optimized Monocular Navigation for Autonomous Rendezvous and Proximity Maneuvers in Small Satellite Missions
Lovaglio, Lucrezia;Stesina, Fabrizio
2025
Abstract
Accurate estimation of the position and orientation of a spacecraft during proximity operations—such as rendezvous, docking, on-orbit servicing (OOS), and active debris removal (ADR)—is critical to ensuring mission success and safety. Tradi- tional visual navigation methods based on hand-engineered fea- ture matching often struggle with robustness and generalization, while existing deep learning approaches face limitations due to heuristic hyperparameter tuning and limited training data. In this work, a novel convolutional neural network (CNN)-based architecture for monocular pose estimation of non-cooperative spacecraft is proposed, specifically designed to improve robust- ness across diverse operational scenarios. The model is trained on a high-fidelity synthetic dataset comprising approximately 25,000 images, simulating realistic proximity conditions with variations in lighting, background textures, and spacecraft geometries. To assess its performance, an extensive benchmarking study is conducted against representative State-of-the-Art methods using standardized evaluation metrics and controlled test conditions. The results demonstrate the competitive performance of the proposed method and provide critical insights into the factors affecting pose estimation accuracy in realistic spaceborne appli- cationsFile | Dimensione | Formato | |
---|---|---|---|
IEEE_mfa_Loaglio_final.pdf
accesso aperto
Tipologia:
2. Post-print / Author's Accepted Manuscript
Licenza:
Pubblico - Tutti i diritti riservati
Dimensione
1.58 MB
Formato
Adobe PDF
|
1.58 MB | Adobe PDF | Visualizza/Apri |
Deep_Learning-Optimized_Monocular_Navigation_for_Autonomous_Rendezvous_and_Proximity_Maneuvers_in_Small_Satellite_Missions.pdf
accesso riservato
Tipologia:
2a Post-print versione editoriale / Version of Record
Licenza:
Non Pubblico - Accesso privato/ristretto
Dimensione
1.8 MB
Formato
Adobe PDF
|
1.8 MB | Adobe PDF | Visualizza/Apri Richiedi una copia |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/11583/3002697