AI 服務暫時不可用，以下為來源正文，待恢復後補全翻譯。
Every subfield of machine learning has a moment where it stops being a collection of papers and starts being a stack. NLP had it when Hugging Face Transformers turned “reimplement BERT from the appendix” into a single from_pretrained call. Image generation had it with Diffusers. Robotics is having that moment right now, and the stack is called LeRobot. Here is the strange thing about robot learning a few years ago: the models were mostly fine. ACT worked. Diffusion Policy worked. The problem was everything around the models. Every lab had its own dataset format, its own teleoperation rig, its own training loop, its own robot driver written at 2am before a deadline. Nothing composed. If NLP had operated this way, every research group would still be maintaining its own tokenizer and its own incompatible copy of Wikipedia. LeRobot’s bet, and I think it is the right one, is that robotics today is not architecture-limited, it is coordination-limited. So instead of shipping a model, Hugging Face shipped a protocol. The library, now backed by an ICLR 2026 paper and contributions from NVIDIA (GR00T, Isaac Teleop), has quietly become the default substrate for open robot learning. LeRobot is to robot learning what USB was to peripherals: boring on purpose, and transformative because of it. Everything is a policy Read more