[3Xin4] Poster session 1

[3Xin4-59] Verification Study of Stochasitic Sequence Prediction Approach using GPDM for Modeling User Skill in VR Kendama

〇Yusuke Goutsu¹, Tetsunari Inamura^1,2 (1.National Institute of Informatics, 2.The Graduate University for Advanced Studies, SOKENDAI)

Keywords:Virtual reality, User model, Difficulty adjustment, Stochastic approach, Gaussian process

This paper presents a task difficulty adjustment method that allows the user to reach desired success rate instantly using VR technology. We propose a methodology based on a Gaussian process dynamical model (GPDM) to model the user's skill from a small number of past performance observations, and predict future performance at a targeted difficulty level under consideration of model uncertainty. As a task to be performed within a VR environment, we focus on Kendama (a cup-and-ball sports game), in which the cup size is changeable to adjust the difficulty level. In the experiment, we evaluated the personalized skill model with participants who performed the VR Kendama. Our results indicate that the GPDM-based approach accurately reflects the users' skills, and the predicted success rate when changing the difficulty level is close to the actual success rate even with a small number of trials. This instant difficulty adjustment can therefore help users to receive a pleasant user experience.