Guiding Attention for Grasping Tasks by Gestural Instruction: The GRAVIS-Robot Architecture