The use of social robots in education, healthcare, and rehabilitation contexts is increasingly widespread thanks to the enormous benefits and positive impacts they have proven to bring to these fields. In such application domains, the robot often fulfills the role of an expert tutor, generally with the ultimate goal of facilitating and encouraging the learning process, tackling the problem of low motivation and limited commitment to the task. In this perspective, it is crucial to understand how the complex and implicit social mechanisms driving human-human interactions evolve to ultimately address the problem of how to replicate a comparable level of social intelligence on a robot. The targeted application areas require adaptive, reliable, and autonomous solutions for achieving effective interactions over time, even in unstructured contexts. The main challenge for HRI in dealing with the complexity and unpredictability of such contexts is to design solutions that allow the robot to learn during deployment by exploiting its embodiment in the real world and interactions with its human partners. This thesis identifies and practically addresses some key questions and challenges in pursuit of one final goal: to contribute to the development of adaptive and autonomous social robot tutors that can interact with both individuals and groups in a personalized and effective way. We summarize these challenges in 4 key points: i) understand the impact of a social embodied agent providing physical assistance in motor learning tasks; ii) find solutions to let the robot autonomously organize its experience and learn to recognize the partners, also in unstructured scenarios (e.g., multiparty interactions); iii) provide solutions able to foster adaptation and personalization at different temporal levels (i.e., adaptation in real-time and across multiple encounters, to pursuit the final goal of long-term personalized solutions) and iv) the design of tutoring and behavioral robot strategies in groups. Our findings strengthened the evidence of the benefits of social robots on users’ motivation, bringing new evidence about the importance of considering the sphere of social/affective communication in the context of motor learning and physical training. As a fundamental first step towards autonomous adaptation and personalization in long-term tutoring, we developed a novel architecture to address person recognition in HRI that concurrently addresses the following aspects: (i) multimodality; (ii) automation of the recognition pipeline; (iii) robustness to multiparty interactions; (iv) open-set, incremental learning. Furthermore, we exploited this architecture in a tutoring context, integrating it with specific functionalities to personalize the tutoring experience through real-time feedback and behaviors based on past experience. Finally, we provide a first insight into how to manage the design of more complex social behaviors for robot assisting groups by investigating a participatory design and users-in-the-loop approach. The presented solutions and outcomes lay important foundations for developing adaptive and autonomous social robots capable of exploiting their first-hand experience to learn valuable information about their human partners and thus customize their interaction accordingly.
Personalizing social robot tutoring for individual and group interactions
BELGIOVINE, GIULIA
2022-06-16
Abstract
The use of social robots in education, healthcare, and rehabilitation contexts is increasingly widespread thanks to the enormous benefits and positive impacts they have proven to bring to these fields. In such application domains, the robot often fulfills the role of an expert tutor, generally with the ultimate goal of facilitating and encouraging the learning process, tackling the problem of low motivation and limited commitment to the task. In this perspective, it is crucial to understand how the complex and implicit social mechanisms driving human-human interactions evolve to ultimately address the problem of how to replicate a comparable level of social intelligence on a robot. The targeted application areas require adaptive, reliable, and autonomous solutions for achieving effective interactions over time, even in unstructured contexts. The main challenge for HRI in dealing with the complexity and unpredictability of such contexts is to design solutions that allow the robot to learn during deployment by exploiting its embodiment in the real world and interactions with its human partners. This thesis identifies and practically addresses some key questions and challenges in pursuit of one final goal: to contribute to the development of adaptive and autonomous social robot tutors that can interact with both individuals and groups in a personalized and effective way. We summarize these challenges in 4 key points: i) understand the impact of a social embodied agent providing physical assistance in motor learning tasks; ii) find solutions to let the robot autonomously organize its experience and learn to recognize the partners, also in unstructured scenarios (e.g., multiparty interactions); iii) provide solutions able to foster adaptation and personalization at different temporal levels (i.e., adaptation in real-time and across multiple encounters, to pursuit the final goal of long-term personalized solutions) and iv) the design of tutoring and behavioral robot strategies in groups. Our findings strengthened the evidence of the benefits of social robots on users’ motivation, bringing new evidence about the importance of considering the sphere of social/affective communication in the context of motor learning and physical training. As a fundamental first step towards autonomous adaptation and personalization in long-term tutoring, we developed a novel architecture to address person recognition in HRI that concurrently addresses the following aspects: (i) multimodality; (ii) automation of the recognition pipeline; (iii) robustness to multiparty interactions; (iv) open-set, incremental learning. Furthermore, we exploited this architecture in a tutoring context, integrating it with specific functionalities to personalize the tutoring experience through real-time feedback and behaviors based on past experience. Finally, we provide a first insight into how to manage the design of more complex social behaviors for robot assisting groups by investigating a participatory design and users-in-the-loop approach. The presented solutions and outcomes lay important foundations for developing adaptive and autonomous social robots capable of exploiting their first-hand experience to learn valuable information about their human partners and thus customize their interaction accordingly.File | Dimensione | Formato | |
---|---|---|---|
phdunige_4625579_1.pdf
Open Access dal 17/12/2022
Descrizione: Initial part of the thesis (including coverpage, abstract, list of contents, chapters from 1 to 5)
Tipologia:
Tesi di dottorato
Dimensione
17.83 MB
Formato
Adobe PDF
|
17.83 MB | Adobe PDF | Visualizza/Apri |
phdunige_4625579_2.pdf
Open Access dal 17/12/2022
Descrizione: Chapter 6 of the thesis
Tipologia:
Tesi di dottorato
Dimensione
8.15 MB
Formato
Adobe PDF
|
8.15 MB | Adobe PDF | Visualizza/Apri |
phdunige_4625579_3.pdf
Open Access dal 17/12/2022
Descrizione: Chapters 7 and 8, list of publications, references
Tipologia:
Tesi di dottorato
Dimensione
11.6 MB
Formato
Adobe PDF
|
11.6 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.