An Automatic Behavior Toolkit for a Virtual Character

Abstract

Approaches that apply programming by demonstration (PbD) to automatically generate the behaviors of virtual characters have been actively studied. One of PbD directly delivers the knowledge of a predecessor to the virtual character. Therefore, a virtual character learns the behaviors to be executed by observing the behaviors of a predecessor. All consecutive actions are derived from the actions collected as behaviors. The behaviors to be executed are selected from defined behaviors using the Maximin Selection algorithm. However, these approaches collect a large amount of data in real time. Therefore, the amount of data significantly increases, and their analysis becomes difficult. This paper proposes a toolkit that employs PbD to automatically generate the behaviors of virtual characters based on those of a predecessor. Furthermore, an approach to manage and analyze the collected data is described. On the basis of the results of an experiment, it was verified that the proposed toolkit could generate a script of the behaviors of virtual characters for driving in a car simulation.

This research was supported by HUNIC(Hub University for industrial collaboration) at Dongguk University. This paper summarized the results of the “Development of a Supervised Learning Framework for Eldercare Robot Contents” project.