The demand for accessible and continuously updated technical documentation is growing across various industries, particularly in maintenance and manufacturing. Traditional manuals often fail to keep pace with evolving procedures, making updates cumbersome and inefficient. This thesis presents a digital assistant framework designed to interact with experts in real-time, capturing procedural knowledge and structuring it into dynamic and adaptable manuals. The system operates through guided interactions where experts describe and demonstrate procedures while the assistant extracts key information via structured dialogues, multimodal data processing, and contextual refinement. Leveraging voice input, image capture, and motion tracking, the assistant organizes tasks into a structured format, ensuring the precise documentation of actions, tools, conditions, and best practices. The collected data is processed by a Python-based server and integrated into a Unity-based interactive interface, allowing users to access, edit, and refine documentation dynamically. Another important feature of the system is the possibility to train users on a specific, pre-gathered procedure. Unlike static manuals, this approach enables real-time updates, allowing procedural knowledge to evolve alongside industrial practices. The framework’s modularity ensures adaptability across multiple sectors beyond maintenance applications. As industries increasingly rely on structured data for automation and operational efficiency, this system presents a scalable and innovative solution for the future of technical documentation.
La necessità di documentazione tecnica accessibile e costantemente aggiornata è in crescita in diversi settori, in particolare nella manutenzione e nella produzione industriale. I manuali tradizionali spesso non riescono a tenere il passo con l’evoluzione delle procedure, rendendo gli aggiornamenti complessi e inefficaci. Questa tesi propone una struttura per la creazione di un assistente digitale volto ad interagire in tempo reale con gli esperti, catturando la loro conoscenza procedurale e strutturandola in manuali dinamici e adattabili. Il sistema opera attraverso un’interazione guidata in cui gli esperti descrivono e dimostrano le procedure, mentre l’assistente estrae le informazioni chiave mediante dialoghi strutturati, elaborazione multimodale dei dati e raffinamento contestuale. Grazie all’uso di input vocali, acquisizione di immagini e tracciamento dei movimenti, l’assistente organizza le diverse parti delle procedure in un formato strutturato, garantendo la documentazione accurata di azioni, strumenti, requisiti e consigli . I dati raccolti vengono elaborati da un server basato su Python e integrati in un’interfaccia interattiva sviluppata in Unity, consentendo agli utenti di accedere, modificare e aggiornare la documentazione in modo dinamico. A differenza dei manuali statici, questo approccio permette aggiornamenti in tempo reale, consentendo alla conoscenza procedurale di evolversi insieme alle pratiche industriali. La natura modulare dell’architettura ne garantisce l’adattabilità a diversi settori oltre l’ambito della manutenzione. Con l’industria sempre più orientata all’uso di dati strutturati per l’automazione e l’efficienza operativa, questo sistema propone una soluzione innovativa e scalabile per il futuro della documentazione tecnica.
An AI-powered digital assistant framework for automated procedure documentation
Cappello, Luca
2023/2024
Abstract
The demand for accessible and continuously updated technical documentation is growing across various industries, particularly in maintenance and manufacturing. Traditional manuals often fail to keep pace with evolving procedures, making updates cumbersome and inefficient. This thesis presents a digital assistant framework designed to interact with experts in real-time, capturing procedural knowledge and structuring it into dynamic and adaptable manuals. The system operates through guided interactions where experts describe and demonstrate procedures while the assistant extracts key information via structured dialogues, multimodal data processing, and contextual refinement. Leveraging voice input, image capture, and motion tracking, the assistant organizes tasks into a structured format, ensuring the precise documentation of actions, tools, conditions, and best practices. The collected data is processed by a Python-based server and integrated into a Unity-based interactive interface, allowing users to access, edit, and refine documentation dynamically. Another important feature of the system is the possibility to train users on a specific, pre-gathered procedure. Unlike static manuals, this approach enables real-time updates, allowing procedural knowledge to evolve alongside industrial practices. The framework’s modularity ensures adaptability across multiple sectors beyond maintenance applications. As industries increasingly rely on structured data for automation and operational efficiency, this system presents a scalable and innovative solution for the future of technical documentation.| File | Dimensione | Formato | |
|---|---|---|---|
|
2025_04_Cappello_Tesi.pdf
accessibile in internet per tutti
Dimensione
6.73 MB
Formato
Adobe PDF
|
6.73 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/10589/234331