| With the increasing aging population and the growing number of "empty-nesters", how to care and accompany "empty-nesters" has become a serious social problem. In daily life, the operation of household objects is one of the most frequent behavior, the elderly and the disabled are hard to accomplish these daily acts due to limited mobility or many other reasons, they need home service robots for help. So not only moving in unstructured home environment, it is more important for the service robots to have the ability of operation, to help people accomplish the handling and delivering of household objects.The thesis takes home service robots providing services of handling and delivering household objects as application background, applies NAO humanoid robot as experimental platform. To enable NAO to proceed smarter and more autonomous motion control, the kinematics model of the NAO’s five degrees of freedom arm is established. The noumenal monocular vision and binocular stereo vision are used to constitute visual servo control system, accurate pose estimation algorithm is proposed and position-based visual servoing (PBVS) control method is designed and iterative learning control is applied to improve the PBVS control law. Experiments verify that the proposed method enables the robot to accomplish the task of handling and delivering household objects, which is of great significance for the robots to provide home services for people.To grasp the objects around the robot and in the sight of the robot, the rapid identification and localization of object is achieved by using Naomark label based on noumenal monocular vision and the position and orientation information of object is estimated through the world homography matrix decomposition rapidly and accurately. At the same time, the PBVS control law is designed for either unimanual or bimanual grasping, by using inverse kinematics to acquire the joint control angle of realizing object grasping. And the pose constraint is proposed during the process of transporting objects bimanually. To avoid the collision of the end-effector with the obstacles and the target object during the grasping process, path planning of the end-effector is added to the original control law. Experimental results show the superiority of improved PBVS control law.For the objects without Naomark label and the moving objects during delivery, the noumenal monocular vision is hard to recognize and locate them accurately due to the low resolution, the unknown depth information, difficulties of image feature extraction and fast image processing, motion blur. To address these challenges of grasping the objects which is difficult to realize base on noumenal monocular vision, the thesis uses binocular stereo camera Bumblebee2 to constitute visual servo system. The rapid identification of object is achieved through natural feature extraction by image processing, the pose of object is estimated by epipolar geometry and fundamental matrix decomposition. The 3D coordinates of feature points are acquired by linear triangular method, then the relative position and orientation of the camera coordinate system and the object coordinate system is calculated. To solve the localization problem of moving objects during delivery process, an extended Kalman filter based object pose estimation and motion prediction method is proposed. The PBVS Lyapunov asymptotic stable control law is designed for either static object grasping or moving object grasping. However, the presence of mechanical error and object motion prediction error makes the end-effector hard to completely track the desired grasping trajectory. Iterative learning control is added to PBVS control law to improve the performance of the servo system. Iterative experiments verify that the improved PBVS control law can correct errors quickly and improve the response speed of system. The proposed method can help the robot achieve object grasping and delivery rapidly and stably, accomplish home services. |
PDF Full Text Request