Learning object-object affordances for placement-position prediction in an item packing task