The steep increase in the number of spacecrafts launched is allowing to perform much more science and exploration than before, but this comes at the cost of new challenges that have to be faced. Indeed, the high operational costs, together with the communication delay in interplanetary missions, prevent from taking advantage of the usual ground-based techniques to navigate and orient the spacecraft. Thus, it is required to design new strategies to allow the spacecraft to perform these tasks autonomously. The goal of this thesis is to develop a new autonomous navigation method that partially relies on artificial intelligence. In particular, the case in which a CubeSat is orbiting around the Moon with a camera that always captures the entire Moon's disk is considered, which is ideal for optical navigation. This work proposes a novel autonomous horizon-based procedure that exploits artificial intelligence to extract, from an image, the edge of the Moon's disk, which in turn can be used to retrieve the position vector of the spacecraft. The performances of this novel strategy have been analyzed and, when compared to the traditional approaches for optical navigation, they showed a higher accuracy. The artificial intelligence network employed is taken from the literature and is called PiDiNet, while the images used both to test the methods and to train the Convolutional Neural Network have been rendered with Unreal Engine 5. Lastly, a preliminary deployability analysis has been performed to estimate the computational time required to process an image using PiDiNet by interpolating data found in the literature about the performances of a Myriad 2 processor.
L'importante aumento nel numero di satelliti lanciati sta permettendo di condurre molte più ricerche ed esplorazioni rispetto al passato, ma ciò comporta anche nuove sfide che devono essere affrontate. Infatti gli alti costi operativi, uniti al ritardo nelle comunicazioni nelle missioni interplanetarie, impediscono di avvalersi delle tecniche usuali basate sull’utilizzo di strumenti a terra per orientare il satellite e controllarne la navigazione. Pertanto è necessario progettare nuove strategie per consentire al satellite di eseguire autonomamente queste operazioni. Lo scopo di questa tesi è sviluppare un nuovo metodo di navigazione autonoma basato in parte sull'intelligenza artificiale. In particolare, si considera il caso in cui un CubeSat orbita attorno alla Luna con una telecamera che cattura costantemente l'intero disco lunare, condizione ideale per la navigazione ottica. Questo lavoro propone una nuova procedura autonoma che sfrutta l'intelligenza artificiale per estrarre da un'immagine il bordo del disco lunare, che a sua volta può essere utilizzato per calcolare il vettore posizione del satellite. Le prestazioni di questa nuova strategia sono state analizzate e, quando confrontate con gli approcci tradizionali alla navigazione ottica, hanno rivelato una maggiore accuratezza. La rete di intelligenza artificiale impiegata è tratta dalla letteratura e si chiama PiDiNet, mentre le immagini utilizzate sia per testare i metodi che per allenare la rete neurale convoluzionale sono state generate utilizzando Unreal Engine 5. Infine è stata eseguita un'analisi preliminare per stimare il tempo computazionale necessario per elaborare un'immagine utilizzando PiDiNet attraverso l'interpolazione di dati trovati in letteratura riguardanti le prestazioni di un processore Myriad 2.
Mid-range horizon-based optical navigation around the Moon using artificial intelligence
Piazza, Fabio
2022/2023
Abstract
The steep increase in the number of spacecrafts launched is allowing to perform much more science and exploration than before, but this comes at the cost of new challenges that have to be faced. Indeed, the high operational costs, together with the communication delay in interplanetary missions, prevent from taking advantage of the usual ground-based techniques to navigate and orient the spacecraft. Thus, it is required to design new strategies to allow the spacecraft to perform these tasks autonomously. The goal of this thesis is to develop a new autonomous navigation method that partially relies on artificial intelligence. In particular, the case in which a CubeSat is orbiting around the Moon with a camera that always captures the entire Moon's disk is considered, which is ideal for optical navigation. This work proposes a novel autonomous horizon-based procedure that exploits artificial intelligence to extract, from an image, the edge of the Moon's disk, which in turn can be used to retrieve the position vector of the spacecraft. The performances of this novel strategy have been analyzed and, when compared to the traditional approaches for optical navigation, they showed a higher accuracy. The artificial intelligence network employed is taken from the literature and is called PiDiNet, while the images used both to test the methods and to train the Convolutional Neural Network have been rendered with Unreal Engine 5. Lastly, a preliminary deployability analysis has been performed to estimate the computational time required to process an image using PiDiNet by interpolating data found in the literature about the performances of a Myriad 2 processor.File | Dimensione | Formato | |
---|---|---|---|
2023_12_Piazza_Executive_Summary.pdf
accessibile in internet per tutti
Descrizione: Executive summary
Dimensione
717.34 kB
Formato
Adobe PDF
|
717.34 kB | Adobe PDF | Visualizza/Apri |
2023_12_Piazza_Thesis.pdf
accessibile in internet per tutti
Descrizione: Master thesis
Dimensione
13.09 MB
Formato
Adobe PDF
|
13.09 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/10589/215541