Position based visual controlling method for industrial robot applicationsA position based visual controlling method for industrial robot applications has been developed, tested and preliminary applied to an automated welding seam grinding application. A hand-eye calibration method, which is a basis for position based visual control of robots, has also been developed and verified to be suitable for precise visual robot control. Introduction The Optical measurement laboratory of Kemi-Tornio Polytechnic has developed a position based visual robot controlling method and preliminary tested it to an industrial application, which can not be automated by using blind robot. Position based visual robot controlling means robot control by using measured real-world coordinates of the object(s) to be manipulated. These coordinates are measured with machine vision. In order to determine the coordinates needed in controlling, the machine vision system must be calibrated both intrinsically and also with respect to the robot. The calibration of the machine vision system with respect to the robot is called hand-eye calibration. In this article we will first introduce an efficient method for hand-eye calibration, then we are going to present a controlling method suitable for all kind of industrial robots, and after that we will report briefly the results of our aptitude tests. Basis for the control: Hand-eye calibration A suitable hand-eye calibration method is a key to position based visual robot control, when the visual sensor is attached to end-effector of the robot. The hand-eye calibration determines the position and orientation of the machine vision system respect to the robot. When the hand-eye calibration is performed, the coordinates measured by the vision system can be transformed to the robot coordinates. There are many hand-eye calibration methods, which are performed by moving the camera above a calibration object by using the robot. The hand-eye parameters can be solved, when there is at least one robot movement (two camera positions), by using the robot kinematics and the determined poses (positions and orientations) between the camera and the calibration object. A problem of these methods is that they are highly depending on the robot kinematics. The possible inaccuracies of the kinematic model of the robot causes inaccuracies also to the hand-eye parameters. The hand-eye calibration can be done without moving the whole robot and without of assumptions of robot kinematics by utilizing so called screw theory. The screw theory is a way to describe the transformation between two coordinate frames. A transformation from the coordinate frame to another can be determined with unique screw axis and few parameters. The screw theory can be efficiently applied to hand-eye calibration when rotating the machine vision system around the tool frame axes of the robot. Determining the poses of the vision system after and before the rotations and setting these rotations axes of the robot as congruent with the screw axis, the hand-eye parameters can be solved.
Our screw theory based hand-eye calibration method is suitable for all kind of industrial robots. The repeatability accuracy of the hand-eye calibration method was tested with five and six degrees of freedom (DOF) industrial robots. The calibration was performed ten times with both robots and the repeatability of the hand-eye parameters was analysed. Deviation of the translational hand-eye parameters (translation along x, y and z –axis) indicate the repeatability of the hand-eye calibration method. These deviations are presented in Table 1.
 Six DOF controlling method The position based visual robot controlling method is based on the homogenous coordinate transformations and discrete trajectory calculations. The 3D coordinates of the target object(s) measured by the visual sensor can be transformed to robot coordinates by using the hand-eye parameters and then the controlling trajectory of the robot can be calculated. The robot can be programmed to move along the trajectory with available robot programming functions. For example, the point-to-point movement function is possible for all industrial robots. The controlling method can be run in PC and the connection with the robot can established via the serial communication. The six DOF controlling method means, that the movement commands for the robot can be determined in all six degrees of freedoms of the 3D space. Even though the robot could not move in all six DOF in 3D space (e.g. in case of five DOF robot), all available movements of the robot can be calculated. Thus, the controlling method developed is suitable for all kind of industrial robots. The accuracy of the controlling is mainly based on the absolute movement accuracy of the robot. It is well known that the standard industrial robot without a kinematic calibration has not very high absolute accuracy. Optical measurement laboratory will implement a kinematic calibration method for industrial robots in the future. This method will improve both the absolute movement accuracy of the robots in use and it will also further increase the accuracy of the position based visual controlling method developed. Automated grinding of welding seam The developed visual controlling method has been preliminary tested in order to automate a grinding task. In this application a spiral shaped welding seam has to be grinded so that the grinded seam is not more than 0,1 mm above the level of the base material surround the seam.
The test set-up consists of a six DOF industrial robot (Figure 1.), a 3D visual sensor and two dial gauges used as the imaginary grinding tool (Figure 2.). The robot was not calibrated (no absolute accuracy -option). The 3D visual sensor was based on a video camera and a laser stripe.
  The 3D coordinates of the welding seam were measured by scanning the seam with 1 mm steps by moving the visual sensor with the robot. After the measurement, the controlling trajectory for the tips of the dial gauges was calculated and performed. The values of the dial gauges were collected manually during the seam tracking movements. The tracking accuracy (mainly in Z-direction) was determined from these values.
The tracking accuracy of the welding seam was tested in different work areas and arm poses of the robot. It was noted that the tracking accuracy was dependant on the work area and the arm poses of the robot. The results presented in Figure 3 show that ± 0.5 mm tracking accuracy of the welding seam in z-direction (normal to surface of the seam) was achieved.
 Summary The position based visual robot controlling method presented can control the robot in all six degrees of freedom in 3D space limited only by degrees of freedoms of the robot itself. The method can be applied to all kind of industrial robots and wide range of flexibility and intelligence demanding applications. The method has been tested and found out to be functional by four, five and six DOF industrial robots.
The developed method was also preliminary applied to a grinding application, which can not be solved without external sensors of the robot. The tests showed that the controlling method itself is applicable for the task, but the tracking accuracy is not yet high enough in every part of the working area of the robot. A kinematic calibration method for the robot will be developed in order to obtain accuracy demanded.
|
