The field of collaborative robotics is rapidly evolving, offering numerous solutions to enhance industrial efficiency and improve human working conditions. Ensuring operator safety is paramount for collaborative robots, but to be truly effective, these robots must also adapt to human actions and intentions. This thesis presents an innovative task assignment strategy for human-robot collaboration, utilizing a deep-learning-based approach. The primary objective is to optimize decision-making during human-robot interactions by minimizing downtime caused by safety-related stops or slowdowns. To achieve this goal, the approach employs machine learning, specifically neural networks, to predict the speed scaling factor dictated by safety measures, which determines the robot’s operating speed based on human proximity. These predictions inform the task selection process, enabling the robot to make real-time, informed decisions. The neural network model is based on a Feedforward network that takes as inputs the positions and tasks of both the robot and the operator, and outputs the average speed scaling factor over a specified future time window. Then, two decision-making algorithms are introduced: the “faster-first” method and the “search-based” method. Both algorithms dynamically consider task availability and safety constraints. The faster-first method selects the next robot task with the lowest predicted slowdown in a future time window. On the other hand, the search-based method makes its decision based on multiple simulations of different combinations of future task selections. Extensive simulations and experimental validations of the proposed algorithms were conducted in various collaborative scenarios. The results demonstrate that the algorithms enhance task completion time and safety performance, with promising applications in real-world manufacturing environments.
Il campo della robotica collaborativa è in rapida evoluzione e offre numerose soluzioni per aumentare l'efficienza industriale e migliorare le condizioni di lavoro dell'uomo. Garantire la sicurezza dell'operatore è fondamentale per i robot collaborativi, ma per essere veramente efficaci, questi robot devono anche adattarsi alle azioni e alle intenzioni dell'uomo. Questa tesi presenta una strategia innovativa di assegnazione dei compiti per la collaborazione uomo-robot, utilizzando un approccio basato sul deep learning. L'obiettivo primario è quello di ottimizzare il processo decisionale durante le interazioni uomo-robot, riducendo al minimo i tempi di inattività causati da arresti o rallentamenti per motivi di sicurezza. Per raggiungere questo obiettivo, l'approccio impiega il machine learning, in particolare le reti neurali, per prevedere il fattore di scala della velocità dettato dalle misure di sicurezza, che determina la velocità operativa del robot in base alla vicinanza dell'uomo. Queste previsioni informano il processo di selezione dei compiti, consentendo al robot di prendere decisioni informate in tempo reale. Il modello di rete neurale si basa su una rete Feedforward che prende come input le posizioni e i compiti del robot e dell'operatore e restituisce il fattore di scala della velocità media in una finestra temporale futura specificata. Vengono poi introdotti due algoritmi decisionali: il metodo “faster-first” e il metodo “search-based”. Entrambi gli algoritmi considerano dinamicamente la disponibilità del compito e i vincoli di sicurezza. Il metodo “faster-first” seleziona il prossimo compito del robot con il minor rallentamento previsto in una finestra temporale futura. Il metodo “search-based”, invece, prende le sue decisioni sulla base di simulazioni multiple di diverse combinazioni di selezioni di compiti futuri. Sono state condotte ampie simulazioni e convalide sperimentali degli algoritmi proposti in vari scenari collaborativi. I risultati dimostrano che gli algoritmi migliorano il tempo di completamento dei task e le prestazioni di sicurezza, con applicazioni promettenti in ambienti industriali reali.
Safe and efficient online task assignment for human-robot collaboration: a deep-learning approach
Spanó, Alessio
2023/2024
Abstract
The field of collaborative robotics is rapidly evolving, offering numerous solutions to enhance industrial efficiency and improve human working conditions. Ensuring operator safety is paramount for collaborative robots, but to be truly effective, these robots must also adapt to human actions and intentions. This thesis presents an innovative task assignment strategy for human-robot collaboration, utilizing a deep-learning-based approach. The primary objective is to optimize decision-making during human-robot interactions by minimizing downtime caused by safety-related stops or slowdowns. To achieve this goal, the approach employs machine learning, specifically neural networks, to predict the speed scaling factor dictated by safety measures, which determines the robot’s operating speed based on human proximity. These predictions inform the task selection process, enabling the robot to make real-time, informed decisions. The neural network model is based on a Feedforward network that takes as inputs the positions and tasks of both the robot and the operator, and outputs the average speed scaling factor over a specified future time window. Then, two decision-making algorithms are introduced: the “faster-first” method and the “search-based” method. Both algorithms dynamically consider task availability and safety constraints. The faster-first method selects the next robot task with the lowest predicted slowdown in a future time window. On the other hand, the search-based method makes its decision based on multiple simulations of different combinations of future task selections. Extensive simulations and experimental validations of the proposed algorithms were conducted in various collaborative scenarios. The results demonstrate that the algorithms enhance task completion time and safety performance, with promising applications in real-world manufacturing environments.File | Dimensione | Formato | |
---|---|---|---|
2024_10_Spano_Tesi.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Testo della tesi
Dimensione
62.78 MB
Formato
Adobe PDF
|
62.78 MB | Adobe PDF | Visualizza/Apri |
2024_10_Spano_Executive_Summary.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Executive summary
Dimensione
3.93 MB
Formato
Adobe PDF
|
3.93 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/10589/227058