Dr. Yutaka Kondo mainly conducts this research.

We propose a method to generate android’s body gesture by real-time reconfiguring a large-scaled motion database which captured humans’ motions. Realization of human-robot interaction (HRI) is crucial for the future direction of human-robot symbiosis. Especially, natural body gesture for visual information, as well as speech dialog for audio information, is very important for human-like robots (e.g., humanoids and androids). First, we propose to morph motions with some semantic meaning based on Bags-of-words based motion retrieval and Match-Web. Next, we propose to interpolate several types of motions based on Motion Graph and sampling-based path planning. We propose to connect necessary components by Key-Value Store. We actually evaluated the effectiveness of the proposed methods by helps of over 1,700 subjects who attend the event. As a future direction, we also describe how to enlarge motion database and how to obtain human natural activities.

Reference

Yutaka Kondo, Kentaro Takemura, Jun Takamatsu and Tsukasa Ogasawara, “A Gesture-Centric Android System for Multi-Party Human-Robot Interaction,” Journal of Human-Robot Interaction, Vol. 2, No. 1, pp. 133-151, 2013.
近藤豊, 竹村憲太郎, 高松淳, 小笠原司, “データベースとオンラインプランニングを統合した高速応答可能なジェスチャ生成”, 日本ロボット学会誌, Vol. 30, No. 9, pp. 899-906, 2012.

Materials

Slide (related to this research topic)