Synthesis of motion representation based on geometric operation in the stochastic space of motions and language

2019 Fiscal Year Final Research Report

• Project/Area Number: 17K20000
• Research Category: Grant-in-Aid for Challenging Research (Exploratory)
• Allocation Type: Multi-year Fund
• Research Field: Human informatics and related fields
• Research Institution: Osaka University
• Principal Investigator: Takano Wataru 大阪大学, 数理・データ科学教育研究センター, 特任教授(常勤) (30512090)
• Project Period (FY): 2017-06-30 – 2020-03-31
• Keywords: ロボット / 自然言語 / 機械学習

Outline of Final Research Achievements

Language is absolutely essential for a robot to be implemented into our daily life that lies on the highly sophisticated intelligence based on symbols and language. In this research, we have presented a stochastic mathematical model with multiple hidden layers to represent the association between human motions and their descriptive language. The motions and language set to the input and output layer, respectively. The input layer are connected to the output layer via multiple hidden layers, and the connection is specified the probabilistic parameters, such as probability of hidden state being generated from an input. We have derived an algorithm to optimize these parameters. It implies than the association between the motion and language can be extracted as the distribution of hidden variables in their space. This model makes it possible to convert motion data into their descriptive sentence.

Free Research Field

ロボティクス

Academic Significance and Societal Importance of the Research Achievements

本研究成果では，運動と言語を結ぶ多層統計モデルの学習法を提案し，行動を言語表現に変換することを実現した．確率パラメータの性質を利用した最適化計算は．新たな学習アルゴリズムであり，学術的意義がある．また，開発した技術は，行動を言語として理解するロボットの基盤となり，研究室，介護施設，病院などの人間行動を言語化する応用研究に発展している．ロボットや人工システムが日常生活に浸透することを加速させる社会的意義がある．