The big growth and diffusion of robots in everyday life has led the scientific community to an increasing interest in the Human-Robot Interaction (HRI). HRI consists of the mutual and collaborative interaction of three elements: robot, human and external environment. Due to interaction complexity, it is necessary making the communication as safe, stable and transparent as possible. Indeed, the main problem related to HRI, is that the elements involved have no reciprocal knowledge, interacting with uncertain and potentially hostile dynamic characteristics. In this scenario, the concept of awareness is introduced. It means the knowledge, not only of the physical features, but most of all of the intentions of the counterpart, with the aim of estimating them and sometime of predicting them. Although, it seems easy for the humans to interpret the robot behavior, the task is much more complicated for the robots. Regarding this issue, in the last few years, the scientific community has been oriented towards the search for control parameters which could consent robot to modulate and adapt their dynamic characteristics during the interaction with humans. The studies of the the human arm dynamic models pointed out that the human stiffness is one of the best parameter for this purpose. Indeed, it is a very accurate indicator of human neural-motor regulation strategies, since it is the first parameter modulated by Central Nervous System, both kinematically and dynamically, as adaptive response to environmental stimuli. Moreover, it plays a fundamental role in the maintenance of postular and motor stability. This thesis work has the goal to estimate the cartesian stiffness, computed at the human arm end-point (the hand) as response to external perturbations. Therefore, the project consists in the development of an experimental acquisition set-up to record force and displacement data, and, subsequently, in the implementation of an estimation algorithm to identify stiffness. The acquisition takes place using a collaborative robot to generate stochastic perturbations in space, while the user is asked to hold the robot manipulator and maintain both a constant arm configuration and muscle co-contraction level. Trials are performed for different kinematic configurations and for different levels of co-contraction, in order to correlate the stiffness variation with respect to these parameters. The hand displacement from equilibrium point and the respective restoring force are measured at the interface between human and robot. The robot used in the interaction is KUKA light weight 4+, particularly suitable in collaborative tasks, thanks to its adaptability, light weight and safety. Moreover, kinematics and the electromyographic signals are detected to correlate their variations to the stiffness. For this aim, two MYO armband gesture controls are used, thanks to their ability to simultaneously acquire the two characteristics. The implemented algorithm has the purpose of estimating the end-point stiffness matrix, starting from the measured forces and displacements. It consists of a frequency estimate that exploits a dual identification of the forces/displacements system (non-parametric and parametric). The results obtained are analyzed, representing the end-point stiffness as an ellipse/ellipsoid. Based on the results obtained, the relationship between stiffness ellipses/ellipsoids and the kinematic and dynamic modulation of the human arm is investigated. In order to validate the frequency-based stiffness estimation, the obtained data are compared with the literature about another traditional pertubation-based method, which uses known displacement. The project has been developed among the Medical Robotic Section (MRSLab) of NearLab at Politecnico di Milano (Department of Electronics, Information and Bioengineering).
La grande crescita e diffusione dei robot nella realtà quotidiana ha portato la comunità scientifica ad un sempre maggiore interesse verso la Human-Robot Interaction. L’HRI consiste nell’interazione mutualistica e collaborativa di tre elementi: robot, uomo e ambiente esterno. Data la delicatezza e la complessità di tale interazione, diventa molto importante rendere la comunicazione il più possibile sicura, stabile e trasparente. Infatti, uno dei principali problemi legati all’HRI, è che gli elementi coinvolti non hanno alcuna conoscenza reciproca, trovandosi ad interagire con caratteristiche dinamiche incerte e potenzialmente ostili per il raggiungimento di un obiettivo comune. In questo scenario, assume un’importanza fondamentale il concetto di awarness, inteso come presa di coscienza reciproca, non solo delle specifiche fisiche, ma sopratutto delle intenzioni della controparte, in modo tale da poterle stimare ed in alcuni casi anche prevederle. Se da un lato appare più semplice per la controparte umana interpretare il comportamento del robot, per quest’ultimo il compito è molto più complesso. A tal proposito, la comunità scientifica, negli ultimi anni, si è orientata verso la ricerca di parametri di robot control adattabili sempre meglio ai meccanismi motori umani, in modo da conferire al robot caratteristiche complianti e modulabili durante l’interazione con gli umani. Dagli studi del modello dinamico del braccio umano è emerso che la proprietà di stiffness è quella che meglio si presta a questo scopo. Essa, infatti, è un indicatore molto accurato delle strategie di moto-regolazione neurale negli esseri umani, essendo il primo parametro che il Sistema Nervoso Centrale modula come risposta adattiva agli stimoli ambientali. Inoltre, essa gioca un ruolo fondamentale nel mantenimento della stabilità postulare e motoria, conferendo alla struttura scheletrica il sostegno necessario per non collassare su se stessa. Questo lavoro di tesi ha come obiettivo principale la stima della stiffness cartesiana, sviluppata all’end-point umano (la mano) in risposta a perturbazioni esterne. Il progetto, pertanto, consisterà nello sviluppo di un ambiente sperimentale per l’acquisizione del parametro in questione e successivamente nell’implementazione di un algoritmo di stima per identificare la stiffness investigata. L’acquisizione avverrà utilizzando un robot collaborativo per generare perturbazioni stocastiche nello spazio, mentre all’utente sarà chiesto di impugnare il manipolo del robot mantenendo una configurazione cinematica ed un livello di co-contrazione muscolare costanti. I trials verranno effettuati per diverse configurazioni cinematiche e diversi livelli di forza, al fine di correlare le variazioni di stiffness con questi parametri. Gli spostamenti della mano dallo stato di equilibrio e le rispettive forze di risposta saranno misurate all’interfaccia tra uomo e robot. Il robot impiegato nell’interazione è un KUKA light weight 4+, particolarmente indicato grazie alla sua adattabilità, ridotto peso e sicurezza nei tasks cooperativi. Sono stati poi acquisiti sia la cinematica del braccio che il segnale elettromiografico, al fine di valutare l’end-point stiffness rispetto alle loro variazioni. Per fare ciò sono stati utilizzati due MYO armband gesture control, grazie alla loro capacità di acquisire contemporaneamente le due caratteristiche di interesse. L’algoritmo implementato ha lo scopo di stimare la matrice di end-point stiffness a partire dalle forze e dai displacement misurati. Esso consiste in una stima in frequenza che sfrutta una duplice identificazione del sistema forze/spostamenti, non parametrica, prima, e successivamente parametrica. I risultati ottenuti saranno analizzati sia numericamente che geometricamente, sfruttando la rappresentazione della end-point stiffness come ellisse. Sulla base dei risultati ottenuti, sarà investigata la relazione tra ellissi di stiffness e le variazioni cinematiche e dinamiche del braccio umano. Per validare la stima effettuata, i dati ottenuti saranno confrontati con la letteratura relativa ad un altro metodo di stima che sfrutta spostamenti noti come pertubazioni all’end-effector. Il progetto è stato sviluppato presso la sezione di Medical Robotic del NearLab al Politecnico di Milano (Dipartimento di Elettronica, Informazione e Bioingegneria).
End-point stiffness estimation of the human arm in human robot interaction
CUCUGLIATO, LUISA
2017/2018
Abstract
The big growth and diffusion of robots in everyday life has led the scientific community to an increasing interest in the Human-Robot Interaction (HRI). HRI consists of the mutual and collaborative interaction of three elements: robot, human and external environment. Due to interaction complexity, it is necessary making the communication as safe, stable and transparent as possible. Indeed, the main problem related to HRI, is that the elements involved have no reciprocal knowledge, interacting with uncertain and potentially hostile dynamic characteristics. In this scenario, the concept of awareness is introduced. It means the knowledge, not only of the physical features, but most of all of the intentions of the counterpart, with the aim of estimating them and sometime of predicting them. Although, it seems easy for the humans to interpret the robot behavior, the task is much more complicated for the robots. Regarding this issue, in the last few years, the scientific community has been oriented towards the search for control parameters which could consent robot to modulate and adapt their dynamic characteristics during the interaction with humans. The studies of the the human arm dynamic models pointed out that the human stiffness is one of the best parameter for this purpose. Indeed, it is a very accurate indicator of human neural-motor regulation strategies, since it is the first parameter modulated by Central Nervous System, both kinematically and dynamically, as adaptive response to environmental stimuli. Moreover, it plays a fundamental role in the maintenance of postular and motor stability. This thesis work has the goal to estimate the cartesian stiffness, computed at the human arm end-point (the hand) as response to external perturbations. Therefore, the project consists in the development of an experimental acquisition set-up to record force and displacement data, and, subsequently, in the implementation of an estimation algorithm to identify stiffness. The acquisition takes place using a collaborative robot to generate stochastic perturbations in space, while the user is asked to hold the robot manipulator and maintain both a constant arm configuration and muscle co-contraction level. Trials are performed for different kinematic configurations and for different levels of co-contraction, in order to correlate the stiffness variation with respect to these parameters. The hand displacement from equilibrium point and the respective restoring force are measured at the interface between human and robot. The robot used in the interaction is KUKA light weight 4+, particularly suitable in collaborative tasks, thanks to its adaptability, light weight and safety. Moreover, kinematics and the electromyographic signals are detected to correlate their variations to the stiffness. For this aim, two MYO armband gesture controls are used, thanks to their ability to simultaneously acquire the two characteristics. The implemented algorithm has the purpose of estimating the end-point stiffness matrix, starting from the measured forces and displacements. It consists of a frequency estimate that exploits a dual identification of the forces/displacements system (non-parametric and parametric). The results obtained are analyzed, representing the end-point stiffness as an ellipse/ellipsoid. Based on the results obtained, the relationship between stiffness ellipses/ellipsoids and the kinematic and dynamic modulation of the human arm is investigated. In order to validate the frequency-based stiffness estimation, the obtained data are compared with the literature about another traditional pertubation-based method, which uses known displacement. The project has been developed among the Medical Robotic Section (MRSLab) of NearLab at Politecnico di Milano (Department of Electronics, Information and Bioengineering).| File | Dimensione | Formato | |
|---|---|---|---|
|
2018_12_Cucugliato.pdf
accessibile in internet solo dagli utenti autorizzati
Descrizione: Testo della tesi
Dimensione
9.43 MB
Formato
Adobe PDF
|
9.43 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/10589/145144