-
Members: FreeCIS
IEEE Members: Free
Non-members: FreeLength: 00:48:39
16 Apr 2013
Abstract: Recently the technologies of �intelligent space� rapidly and widely are applied various systems with advanced network system. Various sensors, actuators, and monitors are distributed into our residence space and wearable machines. KUKANCHI is a term that represents such intelligent space designed for humans and robots to interact. We have approached to the required intelligent system for KUKANCHI based on the concept of affordance. Affordance could be regarded as a feature of an object or environment that implies how to interact with the object or environment. We have taken constructivist approach to model this cognitive ability and have proposed some models applied to real humanoids with recurrent neural networks. In this talk, I will present following two topics relating the concept of affordance from the viewpoint of dynamical systems perspective.
1) Identification of self-other bodies in the visual space The �body-scheme� enables us to convert between motor commands and movement of the body (e.g. hand) in view. We introduce predictability of forward model to segment such objects from the robot�s body. Concretely, recurrent neural model is actually implemented on the humanoid robot HIRO for modeling the forward-inverse model. The experimental result shows that the model can clearly separate the self and other bodies and acquire the internal representation of �body-scheme� reflecting the body structure.
2) Tool-body assimilation model (body enhancement to others) Through trial and experience, humans are capable of using tools as if they are part of their bodies. We modeled this phenomenon called the tool-body assimilation. The model uses recurrent neural net for modeling the �body� dynamics. Furthermore the network dynamics is modulated by the bias input for �body with tool�. Experiments with humanoid robot Actroid show that the generalization capability of neural networks provides the model the ability to deal with unknown tools.
The above studies based on the dynamical cognitive model could provide fundamental findings for designing advanced intelligent spaces.
1) Identification of self-other bodies in the visual space The �body-scheme� enables us to convert between motor commands and movement of the body (e.g. hand) in view. We introduce predictability of forward model to segment such objects from the robot�s body. Concretely, recurrent neural model is actually implemented on the humanoid robot HIRO for modeling the forward-inverse model. The experimental result shows that the model can clearly separate the self and other bodies and acquire the internal representation of �body-scheme� reflecting the body structure.
2) Tool-body assimilation model (body enhancement to others) Through trial and experience, humans are capable of using tools as if they are part of their bodies. We modeled this phenomenon called the tool-body assimilation. The model uses recurrent neural net for modeling the �body� dynamics. Furthermore the network dynamics is modulated by the bias input for �body with tool�. Experiments with humanoid robot Actroid show that the generalization capability of neural networks provides the model the ability to deal with unknown tools.
The above studies based on the dynamical cognitive model could provide fundamental findings for designing advanced intelligent spaces.
