Generative AI for Affective Vibration: Human-Centered Evaluation of LLM and VAE Models

Haptic feedback plays a significant role in expressing emotions, however, there is a lack of research on haptics compared with visual and audio channels. In this paper we investigate AI and machine learning methodologies for generating affective vibrotactile feedback. Two generative AI (GenAI) approaches to vibration generation were examined using a custom dataset of vibrationemotion pairings: a Variational Autoencoder (VAE) approach and a f ine-tuned large language model (LLM) approach. A quantitative user study involving 15 people validated the GenAIs’ capabilities to generate vibrations conveying a range of levels of emotion valence or arousal. Subjective interviews were conducted afterwards which provided valuable insights for multimodal interaction design and future research topics of affective haptics.