Intelligent robot gripper for general purposes

Abstract

In the past many different robot grippers have been developed to grasp one or a few specific objects. Those grippers are well suited for continuous work in structured environments and are thus employed for most of todays industrial applications. Some researchers, on the other hand, have focused their attention on sophisticated general purpose grippers with the dexterousness and kinematics similar to the human hand. This approach leads to mechanically refined but usually heavy and expensive grippers, which may be difficult to mount on a robot wrist. This paper presents another approach to the design and development of general purpose grippers. The basic idea is to integrate various taskdependent sensors with a mechanically still fairly simple gripper and combine it with an appropriate control software involving artificial-intelligence methods. This approach has led to the construction of the “COR-Gripper”* described in this paper. The gripper consists of three independent, parallel fingers and includes a force & torque sensor, tactile sensors as well as optical and ultrasonic proximity-sensors. It is currently mounted on a Puma 560, which is coupled to a vision system. The control-software is a combination of classical algorithms and AI-methods, such as neural networks. The latter is used to calculate the optimal finger-positions for grasping objects of arbitrary shape.