VISION BASED CONTROL OF A 3-DOF MANIPULATOR

Abstract

In this research, vision based control of a 3-DOF manipulator system is studied. Research and experiments in "Look and Move" control, and dynamic visual servoing for robot control are conducted separately. A dual loop control structure is adopted for the robot control in this research. The inner loop is a low level joint feedback controller. The outer loop is either a trajectory planner or a dynamic compensator, which corresponds to "Look and Move" control method or dynamic visual servoing control strategy respectively. It is obvious that the main difference between the two vision based control methods is in the structure of their outer loop. In "Look and Move" robot control, the inner loop joint feedback controller is implemented based on either resolved motion rate joint control(RRC) or resolved motion acceleration joint control(RAC) strategies. The outer loop is a trajectory planner which is designed to make the manipulator move smoothly to reduce the effect of image blurring. In general, our results suggest that implementing a joint controller based on RRC has more robustness than that based on RAC. In dynamic visual servoing, the outer loop is designed to be a dedicated visual Proportional-Integral(PI) controller so as to compensate for the dynamics of the vision loop. Proportional control is used in independent inner loop joint controllers and are designed to make the manipulator move smoothly. Our experiments suggest that the tracking errors are larger in this case than in the case of "Look and Move" control. This is because of the longer delay in visual processing needed for this strategy. It is observed in the experiment that the manipulator can track the moving object more gently in the case of dynamic visual servoing. The influences of delay which occurs in the processing of visual information by the vision system are also analyzed. A long delay incurred in the visual information processing can cause the whole control system to become unstable or undermine the system's robustness in the dynamic visual servoing system. On the other hand, in "Look and Move" control, longer delay in vision processing provides more time for the robot controller to make the manipulator move more smoothly and precisely. As a result, fewer image blurs occur when there is a longer trajectory planning time. Based on our experiment, the performance of the dynamic visual servoing outplays the performance of "Look and Move" control method. This is because of the additional compensation of dynamics of the whole system in dynamic visual servoing method. In order to improve the performance of the vision based control of the 3-DOF manipulator, a more sophisticated vision system capable of high speed processing of visual information is required.