No.14! Tsundere Intaration ~Evaluation of Tsundere AR agent using measured behavior data aimed at behavior change~
Tsundere Intaration ~Evaluation of Tsundere AR agent using measured behavior data aimed at behavior change~
Tainaka Keishi
The Virtual Reality Society of Japan
Abstract
We propose an interaction system using Tsundere AR agent based on the measurement of Tsundere behavior data that is actually used commercially. We verified the effectiveness of the proposed method on the willingness of continuing the work
page link
Yatani Format
Motivation
They proposed an interaction that combines cool behavior for punishment and kind behavior for reward. But they haven't verified its effectiveness.
What issues or problems did they try to solve?
Operant conditioning is only possible if there is a good relationship between the person who encourages the behavior and the person who is encouraged. They would like to establish good relationships and have the person who is encouraged continue to work with motivation through Tsundere interaction.
What was lacking in existing research?
The feedback from the avatar is only audio and appearance changes, and more complex ways such as avatar behavior.
Method
What system did they create? Why did they think that the system design was well justified?
The users who use the System wear Holo Lense 2 through which they watch AR agent. Researchers measured Tsundere behavior data (utterances, audio and action).
They empirically set the situation 1.Before Work, 2.After Work, 3. Before Breaktime, 4. After Breaktime, 5. Being Tired and 6. Can’t Concentrate.
Staff of maid cafes set the sentences for Tsuntsun, Tsundere and Deredere. Example sentences are shown below. Tsuntsun “Humm. Well done!”(ふん、よくやったじゃない) Tsundere “How long is this taking! Bu..., but you did your best in your own way.”(どれだけ時間がかかっているのよ！で、でも、君なりに頑張ったのね。) Deredere “Thank you for your hard work! You worked hard all day and my master is the best in the world! ” (お疲れ様でした。今日もいっぱい頑張って、ご主人様は世界一優秀です)
Actual maid cafe staff play a role based on each utterance and measured posed data using mic and motion tracking device (Azure Kinect DK). The maid cafe staff wear clothes for every use in work and they set a person in front of the staff.
What algorithm did they create? Why did they consider that the algorithm design was good enough?
TDI : Tsundere Interaction (cool behavior and kind behavior) DDI : Deredere Interaction (only kind behavior) NDI : Non dere Interaction (No behavior)
They have the test subjects do tasks performed by a single person who is unlikely to feel approval or responsibility from those around him or her.
What kind of surveys did they do? How did they justify their study design?
Corrent study!
What experiments did they conduct? How did they justify their experimental design?
Test subjects choiced an agent which they liked the best. They performed tasks twice in a row. Test subjects got the behavior from the agent before The task, 5 minutes later from start and after the task Answered questions about achievement, recognition, interest, growth, responsibility, motivation, boredom.
Insight
What results did they obtain? Under what conditions did it work, and under what conditions did it not work?
They obtained good results about achievement, recognition, interest, growth, responsibility, motivation in TDI and DDI, and NDI did not work.
What are the new findings? What are the findings that could be used in other applications or systems?
DDI and TDI did not differ a lot.
Test subjects said good feedback such as “I’m glad it was slow and steady in TDI ”, “I got motivation from the last deredere”. In contrast, some said “I want to be cheered at first instead of being punished”, “I was irritated at the first Tsun utterances”.
