Robot self-awareness: usage of co-training for distance functions for sequences of images