Disambiguating sensory interpretations using minimal sets of sensory data

The problem of recognizing and locating objects in the workspace of a robot entails finding consistent interpretations of sensory data. Demands of speed require that the recognition problem be solved using as little sensory data as possible. In this paper, we consider the problem of optimally acquiring position and surface orientation data about points on the surfaces of objects, for use in recognizing objects. Here, optimal is taken to mean that, subject to restrictions on the sensing geometry, we predict positions for acquiring sensory data that are most stable under errors in the interpretation of the data and that will most rapidly reduce the current set of consistent interpretations of the data. The technique has been implemented and tested.