We present a system for learning the 3 DOF fine-positioning taskof a robot manipulator (Puma 260) using a gripper mounted camera. Smalllateral gripper-target misalignments are corrected in one step. Largerones employ a previous coarse adjustment move in order to bound theparallax effects of the close camera focus. We build object specialized,neural network-based pose estimators with a rather small set of Gaborfilters. Gabor filters perform a spatially localized frequency analysisand resemble the spatial response profile of receptive fields found invisual cortex neurons. The system demonstrates efficiency w.r.t. speedand accuracy, as well as robustness against changing illumination andobject conditions
展开▼
