The future of advanced robotics likely requires LLM-scale models. With more bias towards vision and locomotion than the usual LLM, of course.