Learning an approximate map of the environment by unsupervised bimodal landmark exploration

Technological methods for navigation have achieved a high level of perfection, currently. However, this perfection is mostly achieved at the cost of additional artificial equipment which is extraneous to a freely-navigating agent such as an autonomous robot. Cognitive, biological navigation is based on functionality which is intrinsic to the cognitive system and thus more flexible and autonomous in the true sense. In this paper, exploratory experiments in navigation are performed, based on proximity events, visual landmarks, and distances traveled. It will be shown that a robot is able to learn a Kohonen self-organized landmark map of image and sonar data. An approximate 2-D map of the environment can be computed on the basis of the two major principal components within a sparse distance matrix between landmarks.