Vision and navigation in man-made environments: looking for syrup in all the right places

People are often able to act eeciently in places like grocery stores, libraries, and other man-made domains even if they haven't been to those particular places before: They are exercising useful knowledge about how these environments are organized in order to facilitate their tasks. In this paper we show that everyday environments exhibit useful regularities an autonomous agent can use in order to accomplish tasks eeciently. In particular, we identify useful regularities of grocery stores, and show how they're used in the design of an agent. We discuss how our planning system, Shopper , uses these regularities to nd items in Grocery-World, a simulated grocery store. Suppose I stop to buy Kellogg's Raisin Bran at an unfamiliar grocery store on the way home. How should my task proceed? A slow, but sure way is to systematically walk through the store slowly moving down aisles while looking at anything, careful not to miss the Raisin Bran. However, I can use information of how grocery stores are organized to speed up my search, for example I know: Most stores provide signs outlining an aisle's contents. Signs are placed at the ends of aisles. Cereals are clustered together. Raisin Bran is a cereal. In coming up with a method for nding Raisin Bran, we can use this knowledge as a basis for a strategy: Move across aisles and stop at a \cereals" sign. Enter the aisle under the sign. Find any kind of cereal. Look for Raisin Bran among the cereals. This strategy is eeective for many items in a grocery store. It works because it relies on speciic features of the environment. These features are ensured to exist so as to make shopping easier for people. Thus this strategy should be easily extensible to all medium-sized grocery stores (in the United States). Shopper For a robot operating in an existing man-made domain , knowledge of organization and strategies can prove useful for accomplishing tasks like this. With the Shopper project, we are examining the types of functional knowledge needed for an agent to work in a man-made domain as well as the sensing and control mechanisms needed to use this knowledge. In this paper , we describe the Shopper system: an integrated system incorporating planning and vision techniques for the task of grocery store shopping. Grocery store shopping is a common task everyone does at least occasionally. Since everybody is able to …