| Popis: |
Maintenance and more generally intervention in hostile environment use remote handling tools; between them, robotics technology is of prime importance, because operations are then performed without the presence of man in the loop for most actions; besides flexibility and automation of routine jobs, the use of robotics tools can improve safety aspects, by allowing the operator to focus on supervision tasks. Furthermore, off-line programming of such robotics systems is highly desirable, because other functioning mode, such as “teach-repeat” mode, are no longer feasible in hostile environment. Nevertheless, off-line programming of the teleoperation tasks requires robot’s accuracy for safety and reliability reasons. This implies kinematic calibration of the robot, i.e. identification of a more accurate relationship between joint encoder readings and the actual position of the end-effector. This paper presents two techniques to achieve this goal. The fisrt one (Part I) deals with geometrical parameter identification, by using the traditional estimator theory in statistics (maximum likelihood method), while the second one (Part II) uses neural network techniques for the correction of non-geometrical errors. Experimental and simulation results show that the robot accuracy can be improved by a factor of more than 50. |
