Virtual-reality-based point-and-direct robotic inspection in manufacturing

Abstract

This paper explores a flexible manufacturing paradigm in which robot grasping is interactively specified and skeletal images are efficiently used in combination to allow rapidly setting up surface flaw identification tasks in small-quantity/large-variety manufacturing. Two complementary technologies are combined to make implementation of inspection as rapid as possible. First, a novel material handling approach is described for robotic picking and placing of parts onto an inspection table using virtual tools. This allows an operator to point and give directives to set up robotic inspection tasks. Second, since specification may be approximate using this method, a fast and flexible means of identifying images of perfect and flawed parts is explored that avoids rotational or translational restrictions on workpiece placement. This is accomplished by using skeleton pixel counts as neural network inputs. The total system, including material handling and skeleton-based inspection, features flexibility during manufacturing set-up, and reduces the process time and memory requirements for workpiece inspection.