Learning Motions and Responses from Large-Scale Data on a Cloud-Robotics Platform

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K16074
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Intelligent robotics
• Research Institution: National Institute of Information and Communications Technology
• Principal Investigator: Sugiura Komei 国立研究開発法人情報通信研究機構, 先進的音声翻訳研究開発推進センター先進的音声技術研究室, 主任研究員 (60470473)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: 知能ロボティクス / マルチモーダル言語理解 / クラウドロボティクス / 模倣学習 / 機械学習

Outline of Final Research Achievements

In this study, we aim to develop a learning method of non-verbal knowledge such as motion and multimodal dialogue with a robot using a cloud robotics platform. Time series prediction method based on Deep Neural Network which introduced Dynamic Pre-training was proposed and its effectiveness was validated by using a standard motion data. In addition to improving the cloud robotics platform Rospeex, we have built a domestic service robot that has over 10,000 multimodal concepts. We also developed a multimodal language understanding method Latent Classifier GAN (LAC-GAN) that can understand user commands according to the situation.

Free Research Field

知能ロボティクス