The paper introduces a novel approach for the kinematic coordination of mechanical robot micro-grippers on the basis of neural networks. Conventional robot systems use specialized grippers for specific tasks. For objects with an amorphous structure, variable shape or small dimensions, conventional grippers become unreliable due to several reasons. The present paper presents an approach on the basis of a tentacle shaped micro-gripper with a high number of links and rotational articulated joints. The proposed method for gripping an object is based on wrapping the manipulator's links around the object in order to establish a firm grip. Sensors located near the joints of the micro-manipulator detect the corresponding distances to the object. Since the multi-link manipulator has a high degree of kinematic redundancy (consider up to several hundred links in a chain), the implementation of an effective trajectory control unit is a challenging task, especially if sensor-based real-time coordination is required. In this paper, we use an optimized geometrical path generator in order to teach dynamic neural nets a certain motion behavior, dependent on distance sensor signals. We show that the neural net is able to learn the procedural knowledge for the gripping process with ability of generalization and discuss the results