Action-Driven Perception for a Humanoid
We present active object categorization experiments with a real humanoid robot. For this purpose, the training algorithm of a recurrent neural network with parametric bias has been extended with adaptive learning rates. This modification leads to an increase in training speed. Using this new training algorithm we conducted three experiments aiming at object categorization. While holding different objects in its hand, the robot executes a motor sequence that induces multi-modal sensory changes. During learning, these high-dimensional perceptions are ‘engraved’ in the network. Simultaneously, low-dimensional PB values emerge unsupervised. The geometrical relation of these PB vectors can then be exploited to infer relations between the original high dimensional time series characterizing different objects. Even sensations belonging to unknown objects can be discriminated from known (learned) ones and kept apart from each other reliably. Additionally, we show that the network tolerates noisy sensory signals very well.
KeywordsActive Perception RNNPB Humanoid Robot
Unable to display preview. Download preview PDF.
- 1.O’Regan, J.K., Noë, A.: A sensorimotor account of vision and visual consciousness. Behav. Brain Sci. 24(5) (October 2001); 939–73; discussion 973–1031Google Scholar
- 3.Franceschini, N.: Combined optical, neuroanatomical, electrophysiological and behavioral studies on signal processing in the fly compound eye. In: Taddei-Ferretti, C. (ed.) Biocybernetics of Vision: Integrative Mechanisms and Cognitive Processes: Proceedings of the International School of Biocybernetics, Casamicciola, Napoli, Italy, October 16-22, 1994, vol. 2, World Scientific, Singapore (1997)Google Scholar
- 4.Steinman, S.B., Steinman, B.A., Garzia, R.P.: Foundations of binocular vision: a clinical perspective. McGraw-Hill, New York (2000)Google Scholar
- 8.Cuijpers, R.H., Stuijt, F., Sprinkhuizen-Kuyper, I.G.: Generalisation of action sequences in RNNPB networks with mirror properties. In: Proceedings of the 17th European symposium on Artifical Neural Networks (ESANN), pp. 251–256 (2009)Google Scholar
- 9.Kolen, J.F., Kremer, S.C.: A field guide to dynamical recurrent networks. IEEE Press, New York (2001)Google Scholar
- 10.Riedmiller, M., Braun, H.: A direct adaptive method for faster backpropagation learning: the RPROP algorithm. In: IEEE International Conference on Neural Networks, vol. 1, pp. 586–591 (1993)Google Scholar
- 12.Bradski, G.: The OpenCV Library. Dr. Dobb’s Journal of Software Tools (2000)Google Scholar
- 16.Kleesiek, J., Badde, S., Wermter, S., Engel, A.K.: What do Objects Feel Like? Active Perception for a Humanoid Robot. In: Proceedings of the 4th International Conference on Agents and Artificial Intelligence (ICAART), vol. 1, pp. 64–73 (2012)Google Scholar
- 18.Gibson, J.J.: The theory of affordances. In: Shaw, R., Bransford, J. (eds.) Perceiving, Acting, and Knowing: Toward an Ecological Psychology, pp. 67–82. Erlbaum, Hillsdale (1977)Google Scholar
- 19.Ogata, T., Ohba, H., Tani, J., Komatani, K., Okuno, H.G.: Extracting multi-modal dynamics of objects using RNNPB. In: Proc. IEEE/RSJ Int. Conf. on Intelligent Robots and Systems, Edmonton, pp. 160–165 (2005)Google Scholar