| Summary: | 碩士 === 國立臺灣大學 === 機械工程學研究所 === 105 === For a robot to grasp a randomly placed irregular object is a challenging task. The process includes several steps. First, the robot needs to be capable of identifying the shape, position, and posture of the object. Second, the robot needs to determine a grasping posture. Next, even when the robot can achieve the previous two tasks, the success of grasping relies on adequate contact between the gripper and the object. This issue is especially crucial when the dimensions or positioning of the object have uncertainties or when the object is fragile or soft, because the mechanical gripper has less compliance unlike the human hand. In short, the grasping task requires delicate coordination between the hand, eye, and arm.
This study reports on a novel low-computation object grasping method that can classify complex objects into primitive shapes and then select the object grasping posture based on predefined grasping postures associated with the approximated primitive shapes. In this approach, the object is not precisely modeled, and the grasping posture is selected from a small number of candidates without massive search; thus, the grasping posture for the object can be quickly derived. Because the object and primitive shape have geometrical discrepancy, the gripper is compliant and equipped with infrared proximity sensors on the fingers to compensate for the geometrical uncertainties and provide adequate contact between the object and the grippers. Furtheremore, the gripper also equipped with the pressure arry for detecting the slipping and adjustinh the grasping force. The methodology is experimentally evaluated with several types of objects in different postures.
|
|---|
