HuggingFace 联合创始人 Thomas Wolf 讨论公司新的机器人项目 LeRobot，以及他为何认为机器人技术正处于与几年前 Transformer 和语言模型相似的转折点。
Thomas Wolf, co-founder of HuggingFace, discusses the company's new robotics initiative, LeRobot, and why he believes robotics is at a similar inflection point to transformers and language models a few years ago.
要点 · TL;DR
机器人技术正处于转折点，硬件已就绪，软件正在追赶。 Robotics is at an inflection point with hardware ready and software catching up.
机器人领域的开源通过防止连接丢失导致的灾难性故障来确保安全。 Open source in robotics ensures safety by preventing catastrophic failures from lost connectivity.
数据多样性是主要瓶颈；跨社区共享数据集可以解决泛化问题。 Data diversity is the main bottleneck; sharing datasets across the community can solve generalization issues.
核心观点 · Key points
机器人技术正处于与多年前 Transformer 相同的转折点，硬件已就绪，软件正在追赶。 Robotics is at the same inflection point as transformers were years ago, with hardware ready and software catching up.
机器人领域的开源对安全至关重要，本地模型可防止因连接丢失导致的灾难性故障。 Open source in robotics is crucial for safety, as local models prevent catastrophic failures from lost connectivity.
数据多样性是机器人技术的主要瓶颈；跨社区共享数据集可以解决泛化问题。 Data diversity is the main bottleneck in robotics; sharing datasets across the community can solve generalization issues.
世界模型和视频生成模型正在快速发展，可以为机器人训练生成合成数据。 World models and video generation models are advancing rapidly and can generate synthetic data for robotics training.
Hugging Face 的角色正在从推广自有库演变为赋能更广泛的社区，成为元社区构建者。 Hugging Face's role is evolving from pushing its own libraries to enabling the broader community as a meta-community builder.
反共识 · Contrarian takes
人形机器人可能不是终极形态；更便宜、多样化的机器人可能推动更广泛的采用。 Humanoids may not be the ultimate form factor; cheaper, diverse robots could drive broader adoption.
中国成为开源 AI 模型的冠军令人惊讶，这是由内部竞争驱动的。 China becoming a champion of open source AI models is surprising and driven by internal competition.
AI 模型还不擅长提出正确的科学问题；它们更适合作为助手。 AI models are not yet good at asking the right scientific questions; they are better as assistants.
对人形机器人的恐怖谷担忧可能被夸大；人们会很快接受机器人。 The uncanny valley concern for humanoids may be overblown; people will quickly accept robots.
长期来看，开源模型将因成本和控制优势而胜出，尽管当前存在动荡。 Open source models will win in the long term due to cost and control, despite current turbulence.
本期章节 · Chapters（共 24）
Hugging Face 机器人入门Introduction to Robotics at Hugging Face
社区成长与硬件演进Community growth and hardware evolution
机器人社区三类开发者Three types of developers in the robot community
氛围编码与可及性Vibe coding and accessibility
机器人市场成熟与 ChatGPT 时刻Maturity phase of robotics market and ChatGPT moment
Rich Mini：机器狗重生与创业平台Rich Mini as a reincarnation of robot dogs and a platform for startups
Hugging Face 在机器人数据中的角色Hugging Face's role in robotics data
世界模型及其对机器人的影响World models and their impact on robotics
人形机器人及替代形态Humanoid robots and alternative form factors
人形机器人成本与采用Humanoid robot cost and adoption
基础模型 vs 专用模型Foundation models vs specialized models
开放与封闭模型及 OpenAI 在 HFOpen vs closed models and OpenAI on Hugging Face
开源采纳的动机Motivations for Open Source Adoption
Hugging Face 角色演变Hugging Face's Evolving Role
中国开源领导力China's Open Source Leadership
西方开源复兴Resurgence of Open Source in the West
开放模型与商业信任Open models and business trust
AI for Science 与超人类能力AI for science and superhuman capabilities
开放科学热情Passion for open science
AI 对科学发现的影响与开源角色AI's impact on scientific discovery and open source's role
科学中提出正确问题Asking the right question in science
AI 中有趣问题与分歧Interesting questions in AI and disagreement