Vision based control can make the robot to be more robust and flexible to unknown changes in the environment. Generally the most common task for robots is reach-to-grasp task. Calibration of the vision system prevents the extension of the vision system on robots. Self-calibration of the vision system and uncalibrated visual control is more suitable for vision based robots. In this dissertation, some issues concerned with visual guided reach-to-grasp for robots are discussed, including camera calibration, visual measurement and visual control. Firstly, a method for feature extraction of a rectangle is presented. Hough transform and Lease Square Error (LSE) method are employed to fit the edge of the rectangle. A ROI window based tracker was designed to improve the robustness of feature extraction. Secondly, the range and precision of the vision system was analyzed. The influence of the lens distortion was also discussed. The steps for selecting the configure parameters of the vision system for robot control was presented. The calibration error was presented by an error proportional coefficient and an equivalent rotation matrix. With this error model, the stability of the 3DOF visual servo system was analyzed. Thirdly, an active vision based hand-eye calibration method was presented. The rotation part of the robot hand-eye geometry was determined with three translation movement of the manipulator. Sensitivity analysis indicates the calibration precision satisfying the requirement of visual control. Since the calibration can be achieved without knowing the intrinsic parameters of the camera, it is easy to carried out and require less computational efforts. Fourthly, a measurement model about relative position for stereo rig is provided. The stereo parameter was calibrated online with two motions of the robot. The robustness of self-calibration can be improved with the confidence function. Fifth, with the online calibrated vision system, a visual based control law for the approach movement for grasping is provided. Compare to the traditional PBVS and IBVS, the new method improve the flexibility of visual control to the parameters of the vision system. Sixth, a hierarchical framework for visual guided grasping system was designed, and an industrial manipulator grasping system was developed with this framework. With the combination of eye-to-hand/eye-in-hand camera-robot configuration, the reliability and robustness of visual guided grasping can be improved. Seventh, a position based/image based hybrid visual servo method was presented for a humanoid robot. The movement of the arm was divided into two parts: PBVS on Z-axis and IBVS on XY plane. The desired image feature was determined with an adaptive estimation method. The visual guided grasping was achieved with the presented methods. Finally, the obtained research result was summarized and future work is addressed.
修改评论