In recent years, there has been an increase in the availability and variety of haptic devices. These, however, tend to be expensive, especially if they are of the admittance type. The main obstacle preventing the use of an impedance type device in admittance type tasks, is the unavailability of the measures of the forces a user is applying to it. Low cost impedance haptic devices, then, provide an opportunity. If these cheaper devices are made able to estimate these forces, they could be used, depending on the application, in place of significantly more expensive ones, potentially providing significant savings. The main goal of this thesis is to realize a complete mathematical model of a device that, together with machine learning techniques, can be used to reconstruct the forces acting on the apparatus with the greatest possible accuracy. To this aim, complete analytical expressions of the kinematics and dynamics of the device are obtained. The accuracy of the model is validated through the comparison of force and torque values predicted by the model with those measured on the apparatus. Neural networks and support vector regression are then used to improve the accuracy of the model, in order to reproduce the nonlinear characteristics of the device that are not captured by the mathematical formulation. This results in a scheme that can perform force estimation based on the combination of both methods. Finally, the possibility of using neural networks to provide assistive forces in a generic teleoperation task is briefly investigated
Negli ultimi anni si è assistito a un aumento della varietà e disponibilità di dispositivi aptici. Questi strumenti, però, sono tipicamente costosi, specialmente se progettati per l’utilizzo nella modalità ad ammettenza. Il principale ostacolo che impedisce l’impiego di apparati di tipo ad impedenza nell’eseguire gli stessi compiti di quelli ad ammettenza è la possibilità di misurare le forze esercitate da un operatore. La maggiore disponibilità di dispositivi ad impedenza, dunque, rappresenta un’opportunità; se si rendessero questi strumenti più economici in grado di effettuare una stima di queste forze, potrebbero essere usati al posto di apparati più costosi, portando potenzialmente a risparmi significativi. L’obiettivo principale di questa tesi consiste nel realizzare un modello matematico completo di un dispositivo che, unitamente a tecniche di apprendimento macchina, possa essere utilizzato per ricostruire le forze agenti sullo strumento con la maggiore accuratezza possibile. A questo scopo, vengono ottenute espressioni analitiche complete della cinematica e dinamica del dispositivo. L’accuratezza del modello è valutata confrontandone i valori di forze e coppie previste con quelle agenti sullo strumento. Reti neurali e support vector regression sono poi usate per migliorarne l’accuratezza, e riprodurre le caratteristiche non lineari dell’apparato non catturate dalla formulazione matematica. Si ottiene così uno schema che esegue stima di forza tramite una combinazione di entrambi i metodi. Infine, viene brevemente investigata la possibilità di utilizzare reti neurali per assistere un operatore in un generico compito di teleoperazione tramite la generazione di forze ausiliarie.
Model-based accurate force estimation on a low cost haptic device
MARCHI, MATTEO
2015/2016
Abstract
In recent years, there has been an increase in the availability and variety of haptic devices. These, however, tend to be expensive, especially if they are of the admittance type. The main obstacle preventing the use of an impedance type device in admittance type tasks, is the unavailability of the measures of the forces a user is applying to it. Low cost impedance haptic devices, then, provide an opportunity. If these cheaper devices are made able to estimate these forces, they could be used, depending on the application, in place of significantly more expensive ones, potentially providing significant savings. The main goal of this thesis is to realize a complete mathematical model of a device that, together with machine learning techniques, can be used to reconstruct the forces acting on the apparatus with the greatest possible accuracy. To this aim, complete analytical expressions of the kinematics and dynamics of the device are obtained. The accuracy of the model is validated through the comparison of force and torque values predicted by the model with those measured on the apparatus. Neural networks and support vector regression are then used to improve the accuracy of the model, in order to reproduce the nonlinear characteristics of the device that are not captured by the mathematical formulation. This results in a scheme that can perform force estimation based on the combination of both methods. Finally, the possibility of using neural networks to provide assistive forces in a generic teleoperation task is briefly investigatedFile | Dimensione | Formato | |
---|---|---|---|
2017_4_Marchi.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Thesis
Dimensione
2.96 MB
Formato
Adobe PDF
|
2.96 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/10589/133131