The robot's task is to reach for and grasp an object that is seen through a camera. The literature on related tasks is briefly reviewed. In the present study, the object is a rectangular shaped brick positioned on a table. Redundant robot arm postures exist to grasp the brick at a given location and orientation. The abstract RNN can cope with this redundancy. However, it cannot associate the arm posture simply from the original image. The high-dimensional image data require preprocessing.