Vision-language models for robot grounding in clutter.