Self-Organizing Motion Primitives for Robots Utilizing Deep Learning

2019 Fiscal Year Final Research Report

• Project/Area Number: 15H01710
• Research Category: Grant-in-Aid for Scientific Research (A)
• Allocation Type: Single-year Grants
• Section: 一般
• Research Field: Intelligent robotics
• Research Institution: Waseda University
• Principal Investigator: Ogata Tetsuya 早稲田大学, 理工学術院, 教授 (00318768)
• Co-Investigator(Kenkyū-buntansha): 有江 浩明 早稲田大学, 次世代ロボット研究機構, その他(招聘研究員) (20424814)
• Project Period (FY): 2015-04-01 – 2020-03-31
• Keywords: 深層学習 / 予測学習 / マルチモーダル / 動作プリミティブ / RTミドルウェア

Outline of Final Research Achievements

In recent years, deep learning has been used in a variety of domains, but its application is limited to cyber data processing. It is not sufficiently easy to work in the real world. On the other hand, there is a strong expectation for life support using robots. In recent years, the possibility of general-purpose robots has been attracting attention. In this study, we have developed a novel robot learning model using deep learning. Also we have improved the robot performance for manipulating various materials and reduced the development cost with robot operation system, RT-Middleware.
Specifically, based on the findings of cognitive developmental robotics research which relates to infant developmental learning including imitation learning, predictive coding, etc. we developed multiple robots which are modeling the sensory-motor information (experience) by deep learning.

Free Research Field

知能ロボティクス，認知発達ロボティクス

Academic Significance and Societal Importance of the Research Achievements

本研究の成果は「手離れ」がよく，その後，多様な企業との共同研究に展開している．
例えば，デンソーウェーブ，ベッコフオートメーション，そして研究代表者が技術顧問を務めるエクサウィザーズが開発した「マルチモーダルAIロボット」は，タオルの折畳み，サラダの盛り付けなどを実現している（2017)．また日立製作所では，ドアへの接近，ドア開け，通り抜けという全身動作を，複数の深層学習モデルを用いて学習することに成功した(2018)．デンソーウェーブとエクサウィザーズは，小型ロボットCOBOTTAを利用した粉体秤量を実現した(2018)．
今後，このような多くのロボット応用に発展していくことが期待される．