Mostrar el registro sencillo del ítem
Engagement Estimation in Child-Robot Interaction via Transfer Learning from a Pre-trained Facial Emotion Recognition Model
| dc.contributor.author | García Jiménez, Gonzalo Andrés | |
| dc.contributor.author | Laycock, Rohan | |
| dc.contributor.author | Pérez, Guillermo | |
| dc.contributor.author | Amores, José Gabriel | |
| dc.contributor.author | Álvarez, Gloria | |
| dc.contributor.author | Castro, Manuel | |
| dc.contributor.author | Gomez, Randy | |
| dc.date | 2026 | |
| dc.date.accessioned | 2026-04-30T07:52:36Z | |
| dc.date.available | 2026-04-30T07:52:36Z | |
| dc.identifier.citation | García, G. A., Laycock, R., Pérez, G., Amores, J. G., Álvarez, G., Castro, M., & Gomez, R. (2025, September). Engagement Estimation in Child-Robot Interaction via Transfer Learning from a Pre-trained Facial Emotion Recognition Model. In International Conference on Social Robotics (pp. 237-252). Singapore: Springer Nature Singapore. | es_ES |
| dc.identifier.issn | 978-981-95-2397-9 | |
| dc.identifier.issn | 978-981-95-2398-6 | |
| dc.identifier.uri | https://reunir.unir.net/handle/123456789/19787 | |
| dc.description | Haru4Kids is a system developed to emulate the family-oriented robot Haru, enabling child-robot interaction (CRI) within home environments. During a two-week trial with six families, we collected interaction data, including images, which were later labelled by human annotators into four engagement levels. In this study, we present an artificial intelligence model that estimates a child’s engagement level during CRI using pictures captured at one frame per second. Our model leverages transfer learning, starting with a pre-trained ResNet50-based Facial Emotion Recognition (FER) model, which initially achieved an F1 score of 0.28. Incorporating a Support Vector Classifier raised this to 0.48. Further fine-tuning the FER model yielded minimal gains, but applying Low-Rank Adaptation (LoRA) significantly improved performance, achieving an accuracy of 0.76, an F1 score of 0.69 in cross-validation and 0.56 in the much stricter validation leave-one-group-out. These results highlight the effectiveness of adapting pre-trained emotion models for engagement estimation in CRI. The ultimate goal of this work is to equip robots with some degree of artificial empathy —defined here as the ability to sense and adapt to user affective and attentional states— to support more effective and natural interactions. While desirable for any human-interacting robot, this capability is especially critical --yet often absent- in social robots designed for children, a particularly underexplored user group in human-robot interaction research. | es_ES |
| dc.description.abstract | [Resumen no disponible] | es_ES |
| dc.language.iso | eng | es_ES |
| dc.publisher | Springer Nature Singapore | es_ES |
| dc.relation.uri | https://link.springer.com/chapter/10.1007/978-981-95-2398-6_17 | es_ES |
| dc.rights | restrictedAccess | es_ES |
| dc.subject | child-robot interaction in the wild | es_ES |
| dc.subject | engagement estimation | es_ES |
| dc.subject | affective computing | es_ES |
| dc.subject | emotion recognition | es_ES |
| dc.subject | deep learning | es_ES |
| dc.subject | transfer learning | es_ES |
| dc.subject | Low-Rank Adaptation (LoRA) | es_ES |
| dc.title | Engagement Estimation in Child-Robot Interaction via Transfer Learning from a Pre-trained Facial Emotion Recognition Model | es_ES |
| dc.type | bookPart | es_ES |
| reunir.tag | ~OPU | es_ES |
| dc.identifier.doi | https://doi.org/10.1007/978-981-95-2398-6_17 |
Ficheros en el ítem
| Ficheros | Tamaño | Formato | Ver |
|---|---|---|---|
|
No hay ficheros asociados a este ítem. |
|||
