Viewpoint detection models for sequential embodied object category recognition