Symbols and Dynamics in Embodied Cognition: Revisiting a Robot Experiment
This paper introduces novel analyses that clarify why the dynamical systems approach is essential for studies of embodied cognition by revisiting author’s prior robot experiment studies. Firstly, we argue that the symbol grounding problems as well as the “situatedness” problems should be the consequences of lacking a shared metric space for the interactions between the higher cognitive levels based on symbol systems and the lower sensory-motor levels based on analog dynamical systems. In our prior studies it was proposed to employ recurrent neural networks (RNNs) as adaptive dynamical systems for implementing the top-down cognitive processes by which it is expected that dense interactions can be made between the cognitive and the sensory-motor levels. Our mobile robot experiments in prior works showed that the acquired internal models embedded in the RNN is naturally situated to the physical environment by means of entrainment between the RNN and the environmental dynamics. In the current study, further analysis was conducted on the dynamical structures obtained in the experiments, which turned out to clarify the essential differences between the conventional symbol systems and its equivalence realized in the adaptive dynamical systems.
KeywordsSensory Input Recurrent Neural Network Global Attractor Motor Program Symbol System
Unable to display preview. Download preview PDF.
- 7.Kolen, J.: Exploring the computational capabilities of recurrent neural networks. PhD thesis, The Ohio State University (1994)Google Scholar
- 8.Kuipers, B.: A qualitative approach to robot exploration and map learning. In: AAAI Workshop Spatial Reasoning and Multi-Sensor Fusion, Chicago, pp. 774–779 (1987)Google Scholar
- 10.Matsuno, K.: Physical Basis of Biology. CRC Press, Boca Raton (1989)Google Scholar
- 11.Maturana, H., Varela, F.: Autopoiesis and cognition: the realization of the living. D. Riedel Publishing, Boston (1980)Google Scholar
- 12.Pollack, J.: The induction of dynamical recognizers. Machine Learning 7, 227–252 (1991)Google Scholar
- 13.Rumelhart, D., Hinton, G., Williams, R.: Learning internal representations by error propagation. In: Rumelhart, D., Mclelland, J. (eds.) Parallel Distributed Processing. MIT Press, Cambridge (1986)Google Scholar
- 14.Tani, J.: Model-Based Learning for Mobile Robot Navigation from the Dynamical Systems Perspective. IEEE Trans. on SMC (B) 26, 421–436 (1996)Google Scholar
- 15.Tani, J.: An interpretation of the ”self” from the dynamical systems perspective: a constructivist approach. Journal of Consciousness Studies 5, 516–542 (1998)Google Scholar
- 18.Tani, J., Ito, M.: Self-organization of behavior primitives as multiple attractor dynamics by the forwarding forward model network. Trans On IEEE SMC-B (2003) (in print)Google Scholar