2A1-T03 Agent with Intensified Intelligence and Inclusive Strategy for Kendo: "I"s : 2nd Report: Attack Motion Identification(Cooperation between Human and Machine)
Yasufumi Tanaka, Kazuhiro Kosuge · The Proceedings of JSME annual Conference on Robotics and Mechatronics (Robomec) · 2014
In this article, we propose a method of human motion identification for Kendo agent. There are three basic attack motion, "Kote-Uchi", "Men-Uchi" and "Dou-Uchi", in Kendo. Although we proposed attack motion prediction method for kendo agent, the agent can not predict his opponent's motion without selection of motion model, which is generated as Gaussian Mixture Model (GMM). Therefore the agent should choose the suitable model to predict the motion. In order to choose a suitable model, current opponent's motion is identified from current his opponent's state and GMMs. The motion identification method is evaluated by carrying out some experiments.