1. Symbol Emergence in Robotics

Why people can use language and communicate each other? How can we develop a robot which communicates with people in our daily life?

We consider a symbolic system as an “emergent property” of human society in which people interact physically and socially. Symbols, including language, are not a static gift given by an unknown being, but dynamic phenomena. We take a constructive approach towards symbol emergent system. Recently, we call the academic field “symbol emergence in robotics”  where we try to develop a developmental autonomous robot which can obtain language in bottom-up manner.

1-1. Imitation learning and language acquisition using double articulation structure

Human children obtain various behaviors by observing the daily activities of their parents. In addition, children learn many words and phrases through listening to their parents’ discourse. Is it possible for us to build a robot that can obtain various behaviors and words incrementally? Human children learn languages naturally in their daily lives. However, this is still a very difficult challenge for current robotics research. We have recently been focusing on the double articulation structure that human motion and spoken language implicitly have.  We are developing a developmental robot that can learn languages and behaviors through natural interactions.


1.Tadahiro Taniguchi, Keita Hamahata, and Naoto Iwahashi, Unsupervised Segmentation of Human Motion Data Using Sticky HDP-HMM and MDL-based Chunking Method for Imitation Learning, Advanced Robotics, Vol.25 (17), 2143--2172. (2011)  [PDF]

2. Tadahiro Taniguchi and Shogo Nagasaka, Double Articulation Analyzer for Unsegmented Human Motion using Pitman-Yor Language model and Infinite Hidden Markov Model, 2011 IEEE/SICE International Symposium on System Integration.(2011)  [PDF]


1-2. Multimodal Categorization and Concept Formation


The subjective world of an autonomous robot is totally different from that of humans. Therefore, a symbolic system that is suitable for an autonomous robot might be different from that of humans. Multimodal categorization and concept formation is a series of research in which we try to create a robot that can obtain its concepts and categories on its own. We hope that such a constructive approach will unveil the mystery of the language acquisition process of human infants and the brain mechanism for language acquisition.


1. Takaya Araki, Tomoaki Nakamura, Takayuki Nagai, Shogo Nagasaka, Tadahiro Taniguchi, and Naoto Iwahashi
Online Learning of Concepts and Words Using Multimodal LDA and Hierarchical Pitman-Yor Language Model
IEEE/RSJ International Conference on Intelligent Robots and Systems 2012 (IROS 2012), 1623-1630 .(2012)
2. Tadahiro Taniguchi and Tetsuo Sawaragi,
Incremental Acquisition of Behaviors and Signs based on Reinforcement Learning Schema Model and STDP
Advanced Robotics, Vol.21 (10), pp. 1177-1199 .(2007)  [PDF]

1-3. Simultaneous localization, mapping, and language acquisition

sigverse A robot usually cannot directly and clearly perceive its location. This is usually considered a problem in robotics. However, human seem to abstractly perceive their location to some extent. In other words, humans recognize their location in a more symbolic way, e.g., “in front of the table”, “in the toilet”, and “on the sofa”. We are trying to combine the language acquisition process of such words for location with the localization and mapping process. We will develop a simultaneous localization, mapping, and language acquisition technique that will enable a robot to have a more integrated concept of location.


1. Akira Taniguchi, Haruki Yoshizaki, Tetsunarui Inamura, and Tadahiro Taniguchi, Research on Simul-
taneous Estimation of Self-Location and Location Concepts, Transactions of the Institute of Systems,
Control and Information Engineers, accepted. (in Japanese)


2. Machine Learning

We develop various machine learning techniques and apply them to several human-in-the-loop systems.

2-1. Driving Support System using Double Articulation Analyzer


We don’t randomly drive a car, we drive it in some organized way. Humans drive cars while referring to some driving contexts and output some well-organized sequence of driving behaviors to adequately move along a street. We are trying to develop an unsupervised machine learning technique, called the double articulation analyzer, which can extract meaningful segments of driving behavior. In addition, we intend to develop several advanced driving assistance systems in the future for creating an exciting symbiosis between humans and machines.

** This research is mainly a collaborative research with DENSO CO.

1. Tadahiro Taniguchi, Shogo Nagasaka, Kentarou Hitomi, Naiwala P. Chandrasiri, and Takashi Bando,
Semiotic Prediction of Driving Behavior using Unsupervised Double Articulation Analyzer
2012 IEEE Intelligent Vehicles Symposium, 849 - 854 .(2012)  [PDF]
2. Kazuhito Takenaka, Takashi Bando, Shogo Nagasaka, Tadahiro Taniguchi, and Kentarou Hitomi,
Contextual Scene Segmentation of Driving Behavior based on Double Articulation Analyzer
IEEE/RSJ International Conference on Intelligent Robots and Systems 2012 (IROS 2012), 4847-4852 .(2012)
3. Kazuhito Takenaka, Takashi Bando, Shogo Nagasaka, and Tadahiro Taniguchi
Drive Video Summarization based on Double Articulation Structure of Driving Behavior
ACM Multimedia 2012, .(2012)


2-2. Interactive Automatic Song Writer using Nonparametric Bayesian Approach

Music composition is one of the human creative activities. However, song writing is still a difficult problem for most people without professional skills and knowledge about music. We developed an automatic song generation algorithm that enables users to interactively write a song on the basis of a machine learning technique, such as the Bayesian nonparametrics.

1. Akira Shirai and Tadahiro Taniguchi, A Proposal of an Interactive Music Composition System Using Gibbs Sampler, 14th International Conference on Human-Computer Interaction ( Human-Computer Interaction, Part I, HCII 2011, LNCS 6761), 490-497 .(2011)  [PDF]

2. Akira Shirai and Tadahiro Taniguchi, A Proposal of the Melody Generation Method using Variable-order Pitman-Yor Language Model,  Journal of Japan Society for Fuzzy Theory and Intelligent Informatics, Vol.25 (6), pp. 901-913 .(2013) [PDF] (in Japanese)

Example:IWATE fall in love[youtube]

2-3. Recommendation Methods on Social Network Service

Social networks have a lot of information on their networks and in the documents that the users output. We can develop many kinds of services by mining the latent information from these social networks. We have developed a recommendation algorithm on a social network service.


1. Hiroyuki Koga and Tadahiro Taniguchi, Developing a User Recommendation Engine on Twitter Using Estimated Latent Topics, 14th International Conference on Human-Computer Interaction ( Human-Computer Interaction, Part I, HCII 2011, LNCS 6761), 461-470. (2011)

2. Kenya Sudo, Shogo Nagasaka, Kuniaki Kobayashi, Tadahiro Taniguchi, and Toshiaki Takano, Encouraging User Interaction of Social Network Through Tweet Recommendation Using Community Structure, 2013 Conference on Technologies and Applications of Artificial Intelligence, 300-305 . (2013)


3. Decentralized Autonomous Systems

Our society is considered a decentralized autonomous system. In a centralized system, a central controller controls the entire system. In contrast, the individual components in a decentralized autonomous system behave independently, to some extent. For example, people within a given company individually make decisions, but behave in a coordinated way as a company. A city is not a living thing but changes like a living thing depending on the decision making of its components (people). We take a model-based approach and use multi-agent simulation techniques for better understanding such decentralized autonomous systems and for developing useful methods for our future society.


3-1. Decentralized Autonomous Smart Grid: i-Rene

Since the industrial revolution, fossil fuels have been of the main source of energy in our society. However, fossil fuels will disappear from our world in the future. It is important to make use of the renewable energies including solar, wind, and water power to overcome this crisis. We have to change our power grid into a decentralized power network to make effective use of such renewable energies. We proposed a decentralized autonomous smart grid, named i-Rene. We are studying this system from the viewpoint of intelligent system, economics, game theory, and control theory in order to develop and evaluate such a new power grid.

1. Tadahiro Taniguchi, Algorithm for Adaptive Intelligent Agent Trading Electric Power in Decentralized Autonomous Smart Grid Analysis of Dynamics of Bottom-up Price Formation by Intelligent Power Router and Demand Response, Transactions of the Japanese Society for Artificial Intelligence, Vol.28(1), 77-87. (2013) (in Japanese)  [LINK]
2. Tadahiro Taniguchi and Shiro Yano, Decentralized Trading and Demand Side Response in Inter-Intelligent Renewable Energy Network, The 6th International Conference on Soft Computing and Intelligent Systems / The 13th International Symposium on Advanced Intelligent Systems. (2012)
3. Shiro Yano and Tadahiro Taniguchi, Dynamic game approach for Decentralized optimal control of Re-
newable energy network, Fourth International Workshop on Agent Technologies for Energy Systems
(ATES2013). (2013)
URL: i-Rene project

3-2. Emergence of Communication and Role-differentiation

When people gather with others, they form an organization and/or a community. They can differentiate their roles and effectively collaborate as a complete system. We develop computational models that help with the emergence of communication and role-differentiation in order to better understand adaptive behaviors and collective intelligence of humans.

1. Tadahiro Taniguchi, Kazuma Tabuchi, and Tetsuo Sawaragi, Adaptive Design of Role Differentiation by Division of Reward Function in Multi-agent Reinforcement Learning, SICE Journal of Control, Measurement, and System Integration (JCMSI), Vol.3 (1), pp. 26-34 .(2010)  [PDF]
2. Tadahiro Taniguchi, Kenji Ogawa, and Tetsuo Sawaragi, "Implicit estimation of other's intention without direct observation of actions in a collaborative task: Situation-Sensitive Reinforcement Learning'', SICE Annual Conference 2007, in CD-ROM .(2007)SICE Annual Conference 2007 International Award Finalist [PDF]

4. Human Communication and Community Empowerment

4-1. BIBLIOBATTLE: a Social book-review game

On-line communication is insufficient for solely gaining knowledge about our society. We have to design off-line communications as well. We develop a social book-review game called BIBLIOBATTLE, which encourages people to share their knowledge and their personalities.

1. Tadahiro Taniguchi, Hiroshi Kawakami, and Osamu Katai, Bibliobattle : an informal social interaction design mediated by book reviews, The Transactions of Human Interface Society , Vol.12 (4), pp.93-104. (2010)  [PDF]  (in Japanese)
2. Kenta Oku, Yuma Akaike and Tadahiro Taniguchi, Evaluation of Bibliobattle as Recommender Systems, The Transactions of Human Interface Society , Vol.15 (1), pp. 95-106. (2013)  [PDF]
3. Tadahiro Taniguchi, Hiroshi Kawakami and Osamu Katai, Bibliobattle: Informal community scheme based on book review sessions, The 8th International Workshop on Social Intelligence Design (SID2009), pp. 92-98. (2009) [PDF]


4-2. Dealing Rights to Speak: Mechanism Design for Communication in a Meeting

imageA mechanism named the “dealing rights to speak” was designed to facilitate the participants in a meeting to discuss more frankly and logically and to equalize the time for speech of each participant. The mechanism was inspired by using a micro economics model. We are trying to develop mechanisms that make our daily communications more productive.

1. Hiroyuki Koga and Tadahiro Taniguchi, Dealing Rights to Speak: Mechanism Design for Communication in a Meeting, Journal of Japan Industrial Management Association, (accepted) (in Japanese)

  • Contact