Robot hand - eye coordination using neural

This paper focuses on static hand-eye coordination. The key issue that will be addressed is the construction of a controller that eliminates the need for calibration. Instead, the system should be self-learning and must be able to adapt itself to changes in the environment. In this application, only positional information in the system will be used; hence the above referencèstatic.' Three coordinate domains are used to describe the system: the Cartesian world-domain with elements ~ x = (x; y; z), to describe the position of the target object and the robot end-eeector in world coordinates; the vision domain with elements ~ , to describe the observation of the target object relative to the camera position; and the robot domain with elements ~ known as the joint values of the robot, to describe the position of the robot end-eeector relative to the robot base. The task that is set out to be solved is the following. A robot manipulator has to be positioned directly above a pre-speciied target, such that it can be grasped. The target is speciied in terms of visual parameters. Only the (x; y; z) position of the end-eeector relative to the target is taken into account; this suuces for many pick-and-place problems encountered in industry. (In a number of cases, also the rotation of the hand is of importance, but this rotation can be executed separate from the 3D positioning problem.) Thus the remaining problem is 3 degrees-of-freedom (DoF). Of the robot no explicit model is needed, but only some basic assumptions (explained below) are incorporated. It is known that the robot is controlled by specifying changes of its joint values ~ #, and that thècurrent' joint values ~ can be measured. The ~ # are delta values of ~ ; i.e., # i i. Secondly, from the visual system it is only known that it gives positional information ~ of the observed object. Given the visual data ~ and the robot positional data ~ , a model-free adaptive controller learns to generate robot commands ~ # which position the end-eeector directly above the target object. We are interested in the relation between the visual features (i.e., positions of point features) ~ on the one hand, and joint positions ~ on the other. Traditional model-based hand-eye coordination tends to consider this as two separate problems: rst, a translation from visual to Cartesian domain, followed by a translation …