Subsymbolic representations, self-organizing maps, and object motion learning