A numeric-symbolic expert system for 2-D and 3-D object recognition in robotic applications

In this paper an expert system for 2-D and 3-D object recognition combining numeric and symbolic representations is presented. Starting with some basic concepts, a new method for the boundary representation on 2-D objects is given, and then a method for 3-D object recognition is provided which is based on images of the object. Included in the system is a module that implements an algorithm for learning physical descriptions from functional definitions, examples and precedents, and an inference module for reasoning under uncertainty using Naylor's rule value technique. The expert system involves both numeric modules (written in Pascal) and symbolic modules (written in Prolog). The Pascal and Prolog modules communicate through a special program that plays the role of an adaptive interface between them. Some representative examples are provided.