Applying artificial neural networks to object and orientation recognition for robotic handling

Describes the application of artificial networks to recognition of objects and their orientation for the purpose of robotic handling of the objects. Three scenarios are considered: (1) two similar objects in five orientations; (2) two dissimilar objects in five orientations; and (3) three objects in three orientations. The orientations are identified by a rotation of the viewing position around one of the principal axes of the object. The author compares two configurations: one involves a multilayer perceptron (MLP) with three hidden layers and the other consists of an MLP in series with a bidirectional associative memory (BAM). In both cases the backpropagation paradigm was used to train the multilayer perceptron. It is shown that the BAM can be used in conjunction with an MLP to recognize objects and their orientations with a level of accuracy and reliability that allows such a configuration to be useful for the rapid positioning of grippers during robotic handling