Spacecraft relative navigation is a crucial and yet challenging capability for missions ranging from science and exploration to on-orbit-servicing and active debris removal. Traditional approaches vastly rely on the target’s cooperation or on the ground acquisition of knowledge of the target’s characteristics, then exploited on-board. Today, the achievement of an autonomous navigation is becoming a primary task, enabling new complex mission objectives that require low latencies and thus can not completely rely on ground commands, as is typically done. To accomplish this task, future missions will embark sensors suites composed by multiple Electro-Optical instruments. In fact, in the last decade, the technological development of space sensors as uncooled microbolometers, multispectral cameras and laser scanners, made them portable also on small platforms and apt to be employed for navigation purposes, as opposed to complex scientific instruments. Differentiating the input measurements to navigation not only would raise robustness and accuracy, which are keys for autonomy, but would also increase flexibility, allowing to widen the operational ranges. This research work focuses on the exploitation of Electro-Optical sensors for enhancing autonomous relative navigation in proximity of an uncooperative target. Different types of sensors are examined, probing the advantages and limitations of their stand-alone employment and proposing methods to engage them in synergy for the navigation. Monocular cameras working in the visible band are considered at first, being the most common solution adopted for relative navigation. To explore benefits and boundaries of such approach, particularly challenging applications are considered, as of use to understand when other sensors may come into place. As a result, multispectral navigation is proposed to overcome visibility and illumination-related issues proper of the visible band, identifying from analyses the Long Wave Infrared Band as interesting for a multispectral navigation. Secondly, the navigation performance achievable with currently available thermal infrared cameras is studied. The operative conditions in which thermal-based navigation is advantageous with respect to visible imaging are defined and a multi-modal navigation is proposed. Moreover, visible-infrared images fusion is implemented and compared with the two stand-alone techniques of visible and infrared images. Finally, classical and Deep Learning methods based on LiDAR scanners are developed for pose estimation, offering structured information of the target geometry. The possibility of an integration with a classical vision-based approach is also proposed. The research work successfully finds key elements and guidelines for an exploitation of the different kind of sensors in various scenarios. The proposed methods can increase accuracy, robustness or enable operations for new mission concepts. The Thesis also poses its accent on the verification and testing of the proposed navigation techniques. In particular, the topic of generating representative synthetic data and of their validation with real or experimental data is argued for the three sensors. The lack of prior tools is mostly challenging for thermal cameras, therefore an innovative pipeline for generating synthetic thermal images is proposed and implemented. Experimental activities including calibration of sensors in GNC facilities are carried-out. Such activities allowed tests of the navigation algorithms, comparing the performance achieved with respect to synthetic data. This Thesis offers an overview and initial framework for the concurrent exploitation of visible cameras, thermal cameras and LiDAR sensors, making a step forward to overcome major challenges of proximity relative navigation with uncooperative objects.
La navigazione relativa dei veicoli spaziali è una capacità cruciale, ma impegnativa, per missioni che vanno dalla scienza all'esplorazione, dalla manutenzione in orbita alla rimozione attiva dei detriti. Gli approcci tradizionali si basano in larga misura sulla cooperazione del target o sull'acquisizione a terra di conoscenze sulle caratteristiche del target, poi sfruttate a bordo. Oggi, il raggiungimento di una navigazione autonoma sta diventando un compito primario, consentendo di raggiungere nuovi obiettivi di missione complessi che richiedono basse latenze e quindi non possono affidarsi completamente ai comandi di terra, come avviene di solito. Per realizzare questo compito, le missioni future imbarcheranno suite di sensori composte da più strumenti elettro-ottici. Nell'ultimo decennio, infatti, lo sviluppo tecnologico di sensori spaziali come microbolometri non raffreddati, telecamere multispettrali e laser scanner, li ha resi trasportabili anche su piccole piattaforme e adatti a essere impiegati per scopi di navigazione, in contrapposizione a strumenti scientifici complessi. La differenziazione delle misure in ingresso alla navigazione non solo aumenterebbe la robustezza e l'accuratezza, che sono fondamentali per l'autonomia, ma aumenterebbe anche la flessibilità, consentendo di ampliare i campi operativi. Questo lavoro di ricerca si concentra sullo sfruttamento dei sensori elettro-ottici per migliorare la navigazione autonoma relativa in prossimità di un bersaglio non cooperativo. Vengono esaminati diversi tipi di sensori, sondando i vantaggi e i limiti del loro impiego e proponendo metodi per utilizzarli in sinergia per la navigazione. In primis, vengono prese in considerazione le telecamere monoculari che operano nella banda del visibile, essendo la soluzione più comunemente adottata per la navigazione relativa. Per esplorare i vantaggi e i limiti di questo approccio, si considerano applicazioni particolarmente impegnative, utili per capire quando altri sensori possono entrare in gioco. Di conseguenza, viene proposta la navigazione multispettrale per superare i problemi di visibilità e illuminazione propri della banda visibile, individuando dalle analisi la banda dell'infrarosso a onde lunghe come interessante per una navigazione multispettrale. In secondo luogo, vengono studiate le prestazioni di navigazione ottenibili con le termocamere a infrarossi attualmente disponibili. Vengono definite le condizioni operative in cui la navigazione basata sulle immagini termiche è vantaggiosa rispetto a quella visibile e viene proposta una navigazione multimodale. Inoltre, è stata implementata la fusione di immagini visibili e infrarosse e confrontata con le due tecniche autonome di immagini visibili e infrarosse. Infine, vengono sviluppati metodi classici e di Deep Learning basati su scanner LiDAR per la stima della posa, che offrono informazioni strutturate sulla geometria del target. Viene anche proposta la possibilità di un'integrazione con un approccio classico basato sulla visione. Il lavoro di ricerca è riuscito a individuare gli elementi chiave e le linee guida per lo sfruttamento dei diversi tipi di sensori in vari scenari. I metodi proposti possono aumentare l'accuratezza, la robustezza o consentire operazioni per nuovi concetti di missione. La tesi pone inoltre l'accento sulla verifica e sul collaudo delle tecniche di navigazione proposte. In particolare, viene affrontato il tema della generazione di dati sintetici rappresentativi e della loro validazione con dati reali o sperimentali per i tre sensori. La mancanza di programmi di generazione dati esistenti rappresenta una sfida soprattutto per le termocamere, pertanto viene proposta e implementata una pipeline innovativa per la generazione di immagini termiche sintetiche. Sono state condotte attività sperimentali, tra cui la calibrazione dei sensori nelle strutture del GNC. Tali attività hanno permesso di testare gli algoritmi di navigazione, confrontando le prestazioni ottenute con i dati sintetici. Questa tesi offre una panoramica e un quadro iniziale per lo sfruttamento simultaneo di telecamere visibili, telecamere termiche e sensori LiDAR, facendo un passo avanti per superare le principali sfide della navigazione relativa di prossimità con oggetti non cooperativi.
Spacecraft relative navigation with electro-optical sensors around uncooperative targets
Piccinin, Margherita
2022/2023
Abstract
Spacecraft relative navigation is a crucial and yet challenging capability for missions ranging from science and exploration to on-orbit-servicing and active debris removal. Traditional approaches vastly rely on the target’s cooperation or on the ground acquisition of knowledge of the target’s characteristics, then exploited on-board. Today, the achievement of an autonomous navigation is becoming a primary task, enabling new complex mission objectives that require low latencies and thus can not completely rely on ground commands, as is typically done. To accomplish this task, future missions will embark sensors suites composed by multiple Electro-Optical instruments. In fact, in the last decade, the technological development of space sensors as uncooled microbolometers, multispectral cameras and laser scanners, made them portable also on small platforms and apt to be employed for navigation purposes, as opposed to complex scientific instruments. Differentiating the input measurements to navigation not only would raise robustness and accuracy, which are keys for autonomy, but would also increase flexibility, allowing to widen the operational ranges. This research work focuses on the exploitation of Electro-Optical sensors for enhancing autonomous relative navigation in proximity of an uncooperative target. Different types of sensors are examined, probing the advantages and limitations of their stand-alone employment and proposing methods to engage them in synergy for the navigation. Monocular cameras working in the visible band are considered at first, being the most common solution adopted for relative navigation. To explore benefits and boundaries of such approach, particularly challenging applications are considered, as of use to understand when other sensors may come into place. As a result, multispectral navigation is proposed to overcome visibility and illumination-related issues proper of the visible band, identifying from analyses the Long Wave Infrared Band as interesting for a multispectral navigation. Secondly, the navigation performance achievable with currently available thermal infrared cameras is studied. The operative conditions in which thermal-based navigation is advantageous with respect to visible imaging are defined and a multi-modal navigation is proposed. Moreover, visible-infrared images fusion is implemented and compared with the two stand-alone techniques of visible and infrared images. Finally, classical and Deep Learning methods based on LiDAR scanners are developed for pose estimation, offering structured information of the target geometry. The possibility of an integration with a classical vision-based approach is also proposed. The research work successfully finds key elements and guidelines for an exploitation of the different kind of sensors in various scenarios. The proposed methods can increase accuracy, robustness or enable operations for new mission concepts. The Thesis also poses its accent on the verification and testing of the proposed navigation techniques. In particular, the topic of generating representative synthetic data and of their validation with real or experimental data is argued for the three sensors. The lack of prior tools is mostly challenging for thermal cameras, therefore an innovative pipeline for generating synthetic thermal images is proposed and implemented. Experimental activities including calibration of sensors in GNC facilities are carried-out. Such activities allowed tests of the navigation algorithms, comparing the performance achieved with respect to synthetic data. This Thesis offers an overview and initial framework for the concurrent exploitation of visible cameras, thermal cameras and LiDAR sensors, making a step forward to overcome major challenges of proximity relative navigation with uncooperative objects.| File | Dimensione | Formato | |
|---|---|---|---|
|
phdthesis_piccinin_20230201.pdf
accessibile in internet per tutti
Descrizione: PhD Thesis - Piccinin Margherita
Dimensione
85.28 MB
Formato
Adobe PDF
|
85.28 MB | 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/10589/196597