Demonstrating and Learning Multimodal Socio-communicative Behaviors for HRI: Building Interactive Models from Immersive Teleoperation Data
Résumé
The main aim of artificial-intelligence (AI) is to provide machines with intelligence. Machine learning is now widely used to extract such intelligence from data. Collecting and modeling mul-timodal interactive data is thus a major issue for fostering AI for HRI. We first discuss the egg-and-chicken problem of collecting ground-truth HRI data without actually disposing of robots with mature social skills. Particular issues raised by the current multimodal end-to-end mapping frameworks are also commented. We then analyze the benefits and challenges raised by using immersive tele-operation for endowing humanoid robots with such skills. We finally argue for establishing stronger gateways between HRI and Augmented/Virtual Reality research domains.
Domaines
Robotique [cs.RO]
Origine : Fichiers produits par l'(les) auteur(s)
Loading...