Presentation 2022-10-07
Generation of natural agent behavior based on machine learning
Nobuaki Miyazawa, Atsushi Nakazawa,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) As avatars (agents) are beginning to be used in various situations, there is a demand for agents that perform more human-like actions. In this study, we trained various machine-learning models to perform human-like actions in conversations, generated actions using the models, and evaluated the effectiveness of the models. Specifically, we conducted experiments in which two subjects talked with each other and obtained time-series data about the subjects' gazing states and utterances. Using this dataset, we trained models based on HMM and Transformer. Next, we generated behavior of agents in conversations using the learned model, reproduced conversation scenes using a game engine, and subjectively evaluated the naturalness and reality of the generated results. The results showed that the model based on the Transformer can flexibly select actions.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Human-Agent Interaction / machine-learning / behavioral decision model / anthropomorphic agents
Paper # MVE2022-30
Date of Issue 2022-09-29 (MVE)

Conference Information
Committee MVE / VRSJ-SIG-MR / IPSJ-EC / HI-SIG-DeMO / VRSJ-SIG-CS
Conference Date 2022/10/6(2days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair Kiyoshi Kiyokawa(NAIST)
Vice Chair Sumaru Niida(KDDI Research)
Secretary Sumaru Niida(NAIST) / (DNP) / (Univ. of ToKyo) / (NTT)
Assistant Hidehiko Shishido(Univ. of Tsukuba) / Atsushi Nakazawa(Kyoto Univ.) / Naoya Tojo(KDDI Research) / Naoki Hagiyama(NTT)

Paper Information
Registration To Technical Committee on Media Experience and Virtual Environment / SIG-MR / Special Interest Group on Entertainment Computing / Special Interest Group on De-vice Media Oriented UI / SIG-CS
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Generation of natural agent behavior based on machine learning
Sub Title (in English)
Keyword(1) Human-Agent Interaction
Keyword(2) machine-learning
Keyword(3) behavioral decision model
Keyword(4) anthropomorphic agents
1st Author's Name Nobuaki Miyazawa
1st Author's Affiliation Kyoto University(Kyoto Univ)
2nd Author's Name Atsushi Nakazawa
2nd Author's Affiliation Kyoto University(Kyoto Univ)
Date 2022-10-07
Paper # MVE2022-30
Volume (vol) vol.122
Number (no) MVE-200
Page pp.pp.69-73(MVE),
#Pages 5
Date of Issue 2022-09-29 (MVE)