Tesi etd-02152022-144423
Link copiato negli appunti
Tipo di tesi
Dottorato
Autore
SORRENTINO, ALESSANDRA
URN
etd-02152022-144423
Titolo
From humans to robots: leveraging social intelligence to improve HRI
Settore scientifico disciplinare
ING-IND/34
Corso di studi
Istituto di Biorobotica - BIOROBOTICS
Commissione
relatore Prof. CAVALLO, FILIPPO
Parole chiave
- empathetic module
- Neuro-behavioural model
- social aware navigatiion
- Socially Assistive Robots
Data inizio appello
28/03/2022;
Disponibilità
parziale
Riassunto analitico
Population aging represents the 21st century’s dominant demographic phenomenon in Europe. This phenomenon will dramatically increase the pressure on the health care systems due to the increasing number of people who will face cognitive and mobility problems related to aging, and the decreasing number of available professional caregivers. In this framework, socially assistive robotics (SAR) emerges as an alternative solution for copying this phenomenon. Concerning Information Communication Technology(ICT) solutions commonly adopted in this context (e.g. wearable and environmental sensors, tablet), socially assistive robots could create a time-extended engaging relationship with the final users. Additionally, SARs capability could represent a support tool for professional caregivers in the administration of some cognitive and physical protocols, and in the detection of progressive deterioration of the cognitive and physical state of the patients.
To be treated as social companions and establish trustworthy interactions, socially assistive robots should be endowed with social intelligence. This term usually refers to a set of cognitive processes occurring in the human brain to module the interaction with other human beings (e.g., Theory of Mind and emotional perspective-taking). The rationale behind this work is that socially assistive robots should incorporate this human-like capability to enhance the interaction with the final user.
To this aim, the first research goal relies on the design and the development of a brain-inspired behavioral model which could be incorporated into any robotic platform. The architecture of the proposed model resembles the functionality of the human brain areas which are commonly clustered under the concept of social brain (i.e., thalamus, sensory cortex, associative cortex, prefrontal cortex, and semantic cortex). In the development of this model, two main functions emerge. First, the robot should be able to detect and assess the current profile of the user. To this aim, we designed and developed a perceptual module that can assess the user's attitude during the interaction, by simply detecting the verbal and no-verbal behavior of the user. As described in Chapter \ref{c4}, the module was tested on two different datasets in which two groups of participants interacted with Buddy and Pepper robots, respectively. Especially in the second application scenario, where the behavior of the robot was manually adjusted to match the context of the interaction, the results confirmed that the behavior of the robot fostered the emotional elicitation in the user reactions. Thus, the second function of the behavioral model consists of adopting the user information to shape the robot's behavior accordingly. To this aim, we included in the proposed model two robot's behaviors, namely social aware navigation and the empathetic response, which modify the delivery of the services based on the user behavior.
The second research goal of this thesis focused more on the adaptation behavior of the robot in a social context. Namely, the aim was to investigate if increasing the human-likeness in the robot's behaviors, the final users would better understand the robot's intentions. Namely, we expect that the reflection of human-like behaviors implicitly explains the behavior of the robot to the people interacting with it due to the resemblance with their attitudes. To be endowed with social intelligence, the robot should understand the mental state of the user, and its intentions should be also easily understandable by the user. There must be present a mutual understanding of the behaviors. In this direction, we explored the degree of human-likeness that the robot could achieve in performing the navigation and emotional feedback tasks. In the first case, we shaped the proxemics kept by the robot in the navigation task so that to mimic two human-like personality traits (i.e., Extrovert and introvert). In the empathetic case, we asked a group of 109 human users to directly teach the robot the one-to-many association between emotion and facial expressions. This strategy allowed us to detect whether a robot expressing the learned facial expressions stimulated the mindreading phenomenon in the interacting users. In both cases, the participant perceived a difference in the ``human-like'' behaviors against the non-social behaviors, preferring the first ones.
The last research goal of this thesis investigated the influence of permanent and semi-permanent aspects of the user profile in the interaction. For this analysis, the presented model was integrated with two no-human-like robotic platforms and tested with external users, both in a controlled environment both in a real assistive setting (i.e. elderly care center). The results collected among the experimental scenarios suggest that each user characteristic influences the interaction. As an example, the extroversion trait of the participant was highly correlated with the time spent in the interaction. Similarly, the positive attitude of the participants was positively correlated to calm behavior during the interaction which was reflected in a higher evaluation of the perceived trust in the robotic platform.
In conclusion, the results described in this thesis tend to confirm our main hypothesis that endowing a robotic platform with human-like social capability has more impact on the interaction, than the human-like aesthetic of the robot. Thus, endowing socially assistive robots with social intelligence represents a key aspect for turning robots from a machine into trustworthy companions.
To be treated as social companions and establish trustworthy interactions, socially assistive robots should be endowed with social intelligence. This term usually refers to a set of cognitive processes occurring in the human brain to module the interaction with other human beings (e.g., Theory of Mind and emotional perspective-taking). The rationale behind this work is that socially assistive robots should incorporate this human-like capability to enhance the interaction with the final user.
To this aim, the first research goal relies on the design and the development of a brain-inspired behavioral model which could be incorporated into any robotic platform. The architecture of the proposed model resembles the functionality of the human brain areas which are commonly clustered under the concept of social brain (i.e., thalamus, sensory cortex, associative cortex, prefrontal cortex, and semantic cortex). In the development of this model, two main functions emerge. First, the robot should be able to detect and assess the current profile of the user. To this aim, we designed and developed a perceptual module that can assess the user's attitude during the interaction, by simply detecting the verbal and no-verbal behavior of the user. As described in Chapter \ref{c4}, the module was tested on two different datasets in which two groups of participants interacted with Buddy and Pepper robots, respectively. Especially in the second application scenario, where the behavior of the robot was manually adjusted to match the context of the interaction, the results confirmed that the behavior of the robot fostered the emotional elicitation in the user reactions. Thus, the second function of the behavioral model consists of adopting the user information to shape the robot's behavior accordingly. To this aim, we included in the proposed model two robot's behaviors, namely social aware navigation and the empathetic response, which modify the delivery of the services based on the user behavior.
The second research goal of this thesis focused more on the adaptation behavior of the robot in a social context. Namely, the aim was to investigate if increasing the human-likeness in the robot's behaviors, the final users would better understand the robot's intentions. Namely, we expect that the reflection of human-like behaviors implicitly explains the behavior of the robot to the people interacting with it due to the resemblance with their attitudes. To be endowed with social intelligence, the robot should understand the mental state of the user, and its intentions should be also easily understandable by the user. There must be present a mutual understanding of the behaviors. In this direction, we explored the degree of human-likeness that the robot could achieve in performing the navigation and emotional feedback tasks. In the first case, we shaped the proxemics kept by the robot in the navigation task so that to mimic two human-like personality traits (i.e., Extrovert and introvert). In the empathetic case, we asked a group of 109 human users to directly teach the robot the one-to-many association between emotion and facial expressions. This strategy allowed us to detect whether a robot expressing the learned facial expressions stimulated the mindreading phenomenon in the interacting users. In both cases, the participant perceived a difference in the ``human-like'' behaviors against the non-social behaviors, preferring the first ones.
The last research goal of this thesis investigated the influence of permanent and semi-permanent aspects of the user profile in the interaction. For this analysis, the presented model was integrated with two no-human-like robotic platforms and tested with external users, both in a controlled environment both in a real assistive setting (i.e. elderly care center). The results collected among the experimental scenarios suggest that each user characteristic influences the interaction. As an example, the extroversion trait of the participant was highly correlated with the time spent in the interaction. Similarly, the positive attitude of the participants was positively correlated to calm behavior during the interaction which was reflected in a higher evaluation of the perceived trust in the robotic platform.
In conclusion, the results described in this thesis tend to confirm our main hypothesis that endowing a robotic platform with human-like social capability has more impact on the interaction, than the human-like aesthetic of the robot. Thus, endowing socially assistive robots with social intelligence represents a key aspect for turning robots from a machine into trustworthy companions.
File
| Nome file | Dimensione |
|---|---|
Ci sono 1 file riservati su richiesta dell'autore. |
|