Learning objects by tactile perception

An artificial tactile procedure for learning about objects in a three-dimensional unstructured space by means of artificial intelligence methods is proposed. The main hypothesis adopted is to select polyhedrons as objects to be recognized and learned by the system, because polyhedrons can describe the greater part of industrial pieces. In the first step, the object to be recognized is explored and sensory information is extracted (exploration primitives). In the second step, the information is processed to determine new integrated object characteristics (relations among surfaces). In the third step, the object is recognized and, if unknown, it is learned about in order to increase the system knowledge. The recognition/learning step is based on the exploration primitives extracted in the previous steps.<<ETX>>