Jeremy Stober, Risto Miikkulainen and Benjamin Kuipers. 2011.
Learning geometry from sensorimotor experience.
Proceedings of the First Joint Conference on Development and Learning and Epigenetic Robotics, to appear.
A baby experiencing the world for the first time faces a considerable challenging sorting through what William James called the "blooming, buzzing confusion" of the senses. With the increasing capacity of modern sensors and the complexity of modern robot bodies, a robot in an unknown or unfamiliar body faces a similar and equally daunting challenge.
In order to connect raw sensory experience to cognitive function, an agent needs to decrease the dimensionality of sensory signals. In this paper a new approach to dimensionality reduction is presented, allowing an agent to extract spatial and geometric information from raw sensorimotor experience. The method uses distinctive state abstractions to organize sensorimotor experience and manage the firehose of experience, and sensorimotor embedding to infer low-dimensional representations of space from sensorimotor experience.
This approach is evaluated by learning the geometry of Gridworld and RovingEye robot domains. The results show that sensorimotor embedding provides a better mechanism for extracting geometric information from sensorimotor experience than classic dimensionality reduction methods, and that it provides a scalable method for improving geometric knowledge incrementally and robustly based on agent actions.