Abstract
Off-line programming of robot workcells offers the potential for reduced downtime when new tasks are to be taught so lowering costs and increasing production flexibility. A review of literature shows that a major limitation to the wide spread use of off-line programming has been the lack of cheap, simple tools to measure the pose of the robot for calibration of the arm signature and tool centre point offset. This thesis describes the design, development, and testing of a new, low-cost, vision-based sensor concept to meet this measurement need. Two versions of the sensor concept are implemented, one built around a single camera and the other using two cameras. These prototypes are capable of measuring both the position and orientation of a robot's end-effector. Tests demonstrate their measurement accuracy: about ±0.1mm and ±0.5 in position and orientation. Automatic image analysis routines are developed for identifying the tool centre point of a robot-mounted weld torch. A case study describes testing of the single-camera sensor in a factory workcell. This demonstrates the robustness of the measurements collected and the speed of operation of the sensor in an industrial environment. The tool centre point offset could be calibrated automatically in about one minute compared to tens of minutes using existing manual techniques. Measurements are used to calibrate the kinematic model of the robot arm and the tool centre point offset. Average accuracy of off-line generated programmes is improved from worse than 30mm errors to better than 0.5mm across the approximately 1m diameter working volume of a Puma 560 robot. This level of accuracy is suitable for the welding application which required an accuracy of 0.5-1.5mm. Further consideration is given to industrial application of the sensor in particular for on-line monitoring of the robot tool during extended production runs and future use could include calibration of other, more complex end-effectors.