Modeling Attention with Embodied Visual Behaviors