Exploring Uncertainty and Movement in Categorical Perception Using Robots
Cognitive agents are able to perform categorical perception through physical interaction (active categorical perception; ACP), or passively at a distance (distal categorical perception; DCP). It is possible that the former scaffolds the learning of the latter. However, it is unclear whether ACP indeed scaffolds DCP in humans and animals, nor how a robot could be trained to likewise learn DCP from ACP. Here we demonstrate a method for doing so which involves uncertainty: robots are trained to perform ACP when uncertain and DCP when certain. We found evidence in these trials that suggests such scaffolding may be occurring: Early during training, robots moved objects to reduce uncertainty as to their class (ACP), but later in training, robots exhibited less action and less class uncertainty (DCP). Furthermore, we demonstrate that robots trained in such a manner are more competent at categorizing novel objects than robots trained to categorize in other ways.
KeywordsUncertainty Active categorical perception Robotics
This work was supported by the National Science Foundation under projects PECASE-0953837 and INSPIRE-1344227.
- 5.Lungarella, M., Sporns, O.: Information self-structuring: key principle for learning and development. In: Proceedings of the International Conference on Development and Learning, pp. 25–30. IEEE (2005)Google Scholar
- 7.Pfeifer, R., Bongard, J.: How the Body Shapes the Way We Think: A New View of Intelligence. MIT Press, Cambridge (2006)Google Scholar
- 10.Seung, H.S., Opper, M., Sompolinsky, H.: Query by committee. In: Proceedings of the Fifth Annual Workshop on Computational Learning Theory, pp. 287–294. ACM (1992)Google Scholar
- 12.Zieba, K., Bongard, J.: An embodied approach for evolving robust visual classifiers. In: Proceedings of the Genetic and Evolutionary Computation Conerence, pp. 201–208. ACM (2015)Google Scholar