Visually estimating workpiece pose in a robot hand using the feature points method

Abstract

An automatic method of visually estimating the pose of a workpiece in a robot hand has been developed. This method is not subject to the constraint that the workpiece have one of a finite number of states in the robot hand. Workpiece pose is estimated using the three dimensional locations of workpiece feature points. Workpiece features can be located in space by image feature tracking during a known arm motion and by using trigonometric relations. The correspondence to features on a model of the workpiece can be established by using transformations which verify interfeature distances and angles. The ability to estimate workpiece pose is necessary to enable robots to accomplish many industrial tasks.