[4K1-GS-9] Human interface:

[4K1-GS-9-05] diffusion model based on objective personality for interaction motion generation

〇Takahiro Shibasaki¹, Yutaka Nakamura², Yuya Okadome^1,2 (1. Tokyo University of Science, 2. Riken R-IH)

Keywords:diffusion model, interaction, personality

In a dyadic conversation, the behaviors of one participant, such as nodding and smiling, are influenced by those of the conversation partner. The velocity and magnitude of behaviors during conversation are also affected by the personality traits of each participant. In this study, we propose the personality-aware interaction diffusion model (PIDM) that generates interaction behaviors based on the masking features of participants and their personalities. PIDM generates the interaction behavior using masks of a binary matrix for conditioned generation. PIDM is applied to the motion generation during a dyadic conversation, and the differences in the generated results according to the personality are investigated. The results suggested that PIDM can generate various behaviors by changing one parameter of the Big Five, e.g., extroversion.