A Computational Neuroscience Model of Working Memory with Application to Robot Perceptual Learning

M. Tugcu, X. Wang, J.E. Hunter, J. Phillips, D. Noelle, and D.M. Wilkes (USA)


Learning, working memory, perception, and computer vision


An issue of critical importance to both robots and biological creatures is the efficient use of the limited resources available for survival. One strategy employed by higher animals is the focusing of attention onto the few items, e.g., percepts, salient to reaching the current goals, and ignoring distracting input. In these animals, the pre-frontal cortex working memory plays a significant role in the focus of attention. A recently developed Working Memory Toolkit (WMtk) is based on a computational neuroscience model of working memory. We apply this model/toolkit to two perceptual learning problems from robot vision related to navigation and landmark detection. Our system is described along with two perceptual learning experiments. The results of these experiments are given and show impressive performance both in terms of accuracy and speed of learning. To our knowledge, this is the first such application of a computational neuroscience model of working memory to a robot.

Important Links:

Go Back