on Object-based Visual Attention for Robots