Would be interesting to combine a Boston Dynamics-type robot with (1) a GPT-like language model, (2) deep learning algorithms that analyze human movement (a la Dall-E but for human motion rather than images), and (3) text-to-speech, so that you can say "dance like a child" and it can generate a child-like dance.