Egilea: Unai Zabala Cristobal

Izenburua: Steps toward the plausibility of social robots

Zuzendariak: Elena Lazkano / Igor Rodríguez

Eguna: 2025eko abenduaren 16an
Ordua: 11:00h
Lekua: Ada Lovelace aretoa

Abstract:

"When thinking about social robots, capabilities often depicted in science fiction narratives come to mind. However, although significant advances have been made in robotics and human–robot interaction (HRI), current systems remain limited in their capacity to involve in naturalistic and socially meaningful interactions with humans. While many existing platforms can generate lifelike speech or gestures under controlled conditions, these behaviors frequently lack the coherence, adaptability, and emotional richness necessary for robust real-world social engagement.

To this end, three milestones have been defined. The first is to develop a natural talking body gesture generation system and define metrics for what "natural" means (M1). This milestone directly addresses the lack of association between verbal and non-verbal communication, aiming to produce gestures that are temporally and semantically aligned with speech. Establishing quantitative and qualitative metrics will help evaluate and refine what constitutes "natural" movement in human-robot interaction.

The second, enrich the generation system with multimodal features to be able to produce emotional behavior (M2). By integrating emotional expressivity across modalities, such as facial expressions, posture, and prosody, this milestone addresses the current limitations in conveying emotional states through behavior that feels authentic and context-sensitive.

And the third, integrate the evolved system in the storytelling scenario, a possible application of social robots (M3). Storytelling provides a rich, socially meaningful context in which to test adaptability, coherence, and user participation. This milestone confronts the challenge of transferring expressive behavior from isolated responses to a continuous, emotionally engaging interaction, revealing how well the system performs in more realistic, open-ended environments.

Within this context, the RSAIT research group, under which this thesis is situated has been actively contributing to the development of social behaviors for humanoid robots. In particular, the focus has been on the generation of naturalistic gestural behaviors that accompany verbal communication, as well as the implementation of expressive behaviors that enhance perceived social presence. The present study builds upon this foundation by further exploring these behavioral modules, leveraging recent technological and methodological advancements in the field.

In sum, this thesis addresses fundamental challenges and highlights the limitations in the generation and evaluation of socially meaningful robot behavior. By approaching gesture, emotion, and storytelling as interconnected facets of social communication, it offers an integrated framework for advancing HRI research. The findings are intended to inform future studies and practical implementations, with the goal of making humanoid robots more relatable, expressive, and socially competent.

The results show that the system successfully achieved the objectives of M1, establishing a strong foundation for M2 and M3. Integrating semantically related gestures and expressive behaviors improved user engagement, supported by a probabilistic model for gesture selection and sentiment-based modulation of expressiveness. While generally effective, occasional context ambiguities and polarity misclassifications required manual adjustments. In standalone and collaborative storytelling, the system demonstrated robust performance, highlighting its potential for fully automated, contextually adaptive expressive behavior generation."