Computational Models of Space in Navigation and Other Domains

Bowdoin College, Brunswick ME

Investigators

Abstract

This research will explore computer algorithms capable of the same sorts of spatial reasoning demonstrated by humans. The models constructed will in turn further a deeper understanding of how humans process space and how that processing is sensitive to different kinds of domains. The PI has proposed a model of cognitive mapping called PLAN (Prototypes, Locations, and Associative Networks) which has been partially implemented on an advanced mobile robot. This project will develop versions of the PLAN architecture that can be implemented on low cost robots using off-the-shelf technology. The goal is to develop a general-purpose spatial architecture independent of implementation details such as pattern recognition. Such an architecture should be applicable to non-visual environments as well (e.g., music), and even to completely abstract domains. The algorithms developed in the proposal will be aimed at constructing a general methodology for parsing complex domains into generic spatial structures in a fashion similar to how humans break up large-scale spaces into smaller regions. Humans appear to rely on special places called "gateways" to provide clues as to how to parse environments and the gateway structure has the capability of generalizing even to abstract domains. In music, for example, long pauses serve as a kind of gateway denoting a change in movements. The PI, who first introduced the gateway construct, has proposed that the kinds of gateways found in an environment can significantly impact human performance within the environment. A deeper understanding of this relationship will impact the evaluation and design of environments ranging from music to web sites to courses. The research will take place at an undergraduate liberal arts college, and will provide students with the opportunity to participate in advanced research and to gain hands-on experience with robots.