Invited speakers

    • (Main organizer) Takayuki Nagai, The University of Electro-Communication

      • Title:

  • Jun Tani, KAIST

    • Title:How can we develop deep mind of robots?

    • Abstract:

    • My research motivation has been to investigate how cognitive agents can acquire structural representation via iterative interaction with the world, exercising agency and learning from resultant perceptual experience. Over the past 20 years, my group has tackled on this problem by investigating the idea of predictive coding applied to development of cognitive constructs of robots. Under the principle of predictive coding, dense interaction take place between the top-down intention proactively acting on the outer world and the resultant bottom-up perceptual reality accompanied with the prediction error. Our finding has been that compositionality or systematicity enabling some conceptualization can emerge via such iterative interaction as the result of downward causation in terms of constraints such as multiple spatio-temporal scale property applied to the neural network modeling and the way of tutoring the robots applied to the behavioral interaction level. The talk will highlight our recent results on interactive and integrative learning among multimodality of perceptual channels including pixel level dynamic vision, proprioception and linguistic inputs using a humanoid robot platform Finally, I will point to one aim of future research, how the deep mind of a robot may arise through long-term tutoring.

      • Slide <PDF> # please check the file list at the bottom of this page.

  • Justus Piater, University of Innsbruck

    • Title: Bootstrapping Complex Skills by Building on Experience

    • Abstract:

    • Scaling up the ability of robots to interact with uncontrolled environments in sophisticated ways requires incremental learning capabilities that allow the robot to exploit already-learned concepts to facilitate future learning. I present two examples of such systems from my lab. Both systems involve playing behavior of the robot in order to gather relevant experience in an exploratory fashion.

    • In the first system, the robot first learns simple sensorimotor action-effect predictions (poking individual objects), which are then used as building blocks in a harder action-effect prediction task (placing one object on top of another). The end result is a robot that learns to stack building blocks much like a human infant does.

    • The second system is also motivated by infant development. Here, the central idea is that infants (and even adults) turn hard (or unfamiliar) sensorimotor tasks into familiar tasks by applying specific preparatory actions. Our robot discovers the utility of various actions in diverse contexts, and learns complex manipulation skills by appropriately chaining preparatory actions. Learned skills are in turn added to the repertoire of potential preparatory actions for future tasks. We illustrate this on the task of picking up a book from a table and placing it upright on a shelf.

  • Xavier Hinaut, INRIA

    • Title: Reservoir Computing for Robot Language Acquisition

    • Abstract:

    • How do children learn language? Could we use robots to model children language acquisition? This question is linked to a more general issue: how does the brain associate sequences of symbols to internal symbolic or sub-symbolic representations? I will present a Recurrent Neural Network (RNN), namely an Echo State Network (ESN) or Reservoir, that performs sentence comprehension and can be used for Human-Robot Interaction. The RNN is trained to map sentence structures to meanings (i.e. predicates). This model has interesting capabilities, for instance it can learn to "understand" French and English at the same time. Moreover, it is flexible and can be trained on different kinds of output predicate representations.

    • The objective of this model is double: to improve HRI and provide neural models of language acquisition. From the HRI point of view, this model enables one (1) to gain adaptability because the system is trained on corpus examples (no need to predefine a parser for each language), (2) to be able to process natural language sentences instead of stereotypical sentences (i.e. "put cup left"), and (3) to be able to generalize to unknown sentence structures (not in the training data set). From the computational neuroscience and developmental robotics point of view, the aim of this architecture is to model and test hypotheses about child learning processes of language acquisition (Tomasello 2003).

  • Kuniaki Noda, Waseda University

    • Title: Multimodal Integration Learning of Robot Behaviors using Deep Learning

    • Abstract:

    • Deep learning has recently been successfully applied not only to conventional pattern recognition tasks such as image classification or speech recognition but also to complex sensorimotor learning tasks such as object grasping or manipulation for robots. This talk investigates the potential of deep learning methodologies towards robotics research by introducing our recent studies regarding multimodal integration learning of robot behaviors. Specifically, this talk presents how our approach enables a robot to model correspondences between synchronous temporal inputs from different modalities and to utilize the acquired integrated representation for adaptively switching behaviors by retrieving motor commands relevant to perceived environmental changes.

    • Slide <PDF> # please check the file list at the bottom of this page.

  • Tadahiro Taniguchi, Ritsumeikan University

    • Title: Nonparametric Bayesian Word Discovery for Symbol Emergence in Robotics

    • Abstract:

    • Word discovery from speech signals is a crucial task for a human infant to learn a language. Differently from conventional approach towards automatic speech recognition, infants cannot use labeled data, i.e., transcribed text. They have to discover words from speech signals and learn meanings of the words in an unsupervised manner. We have been developing machine learning methods that enable a robot to learn words automatically. In this talk, I am introducing an unsupervised machine learning methods. The method is called nonparametric Bayesian double articulation analyzer (NPB-DAA) for learning phonemes and words directly from speech signals using hierarchical Dirichlet process hidden language model (HDP-HLM). The method is based on Bayesian nonparametrics. I am also introducing our research field called symbol emergence in robotics.

    • Slide <Slideshare>

  • Komei Sugiura, NICT

    • Title: Cloud robotics for building conversation robots

    • Abstract:

    • To build conversational robots, roboticists are required to have deep knowledge of both robotics and spoken dialogue systems. In this talk, challenges towards spoken dialogues with robots are summarized, and some recent advances using cloud robotics platforms are explained.