在2026年11月03日发布的一期播客中，Tutor Intelligence联合创始人兼首席执行官乔希·格伦斯坦（Josh Gruenstein）做客节目，与播客主持人共同探讨了机器人技术、制造业和供应链的现状。Tutor Intelligence总部主要位于波士顿，拥有约160名员工，致力于训练前沿机器人人工智能模型、大规模制造机器人，并将其直接部署到美国最大的供应链企业中。 乔希在大约五年半前创立了Tutor，其核心论点是：将机器人部署到实体经济中将构建必要的数据集、模型和供应链。Tutor的机器人已经帮助制造并向美国消费者运送了超过2.5亿件产品。大多数客户都是从未购买过机器人的全新买家，美国有十分之九的工厂和仓库缺乏机器人。 乔希解释说，机器人技术需要一个涵盖模型、硬件、软件、供应链、客户和产品的全栈平台，而不仅仅是解决单一的狭窄层面。公司的名称源于将人类远程操作员（导师/tutor）与机器人连接起来，赋予它们智能，从而创建一个类似于互联网为大语言模型（LLM）提供数据的机器人训练数据互联网。随着时间的推移，具身智能模型、3D空间数据和仿真工具有助于加速机器人的能力。 Tutor在马萨诸塞州沃特敦制造自己的硬件，以便利用前沿模型共同设计机器人。他们生产两种主要的机器人类型： 1. **卡西（Cassie）**：一款由基础模型驱动的、依靠蛮力的单臂机器人（参数量在2亿到4亿之间），使用吸盘处理诸如搬运托盘和纸箱等重型任务，以解决繁重的体力劳动。 2. **桑尼（Sunny）**：一款带轮子的工业半人形机器人，具有更高的负载能力、更快的速度以及工业协作安全标准，在波士顿的数据工厂和现实世界场所中运行，用于进行灵活的双臂操作。有一百台桑尼机器人在波士顿运行，并且一个不断增长的车队正在现实世界中部署。 客户的采用是由以每小时16美元的价格提供机器人且对客户而言零经济风险所驱动的。从第一次通话到签署合同的平均时间为40到60天，从签署合同到部署大约为40天，生产启动通常在数小时内完成。中小企业以及财富50强和财富100强公司都在使用它们。运营团队将投入运行后的机器人视为可靠的设备，人机交互（HRI）技术——例如在重复订购时以客户的儿童、足球运动员和流行歌星的名字给机器人命名——培养了与工人的强烈关系。 展望未来，乔希探讨了全球供应链的演变，指出在2026年，全球供应链依然相互连接、相互依存。虽然中国在前一阶段的工业自动化中进行了大量投资，但美国公司正在利用通用能力机器人建设全新的产能，而无需拆除现有基础设施。Tutor继续推进垂直整合，将更多的制造和零部件业务引入马萨诸塞州沃特敦的内部，并大力投资以构建成熟的本土机器人供应链。 Tutor的研究项目利用了波士顿一个名为“数据工厂一号”（Data Factory One）的100台机器人组成的机群，来大规模执行现实世界的强化学习。公司正在积极招聘研究人员、全栈各层的工程师以及用于制造机器人的技术人员。最终，Tutor旨在提供一个通用的劳动力基底，以重塑实体经济、降低商品价格并帮助企业扩大实体运营规模。

In a podcast published on 2026-11-03, Josh Gruenstein, co-founder and CEO of Tutor Intelligence, joins to discuss the state of robotics, manufacturing, and supply chains with the podcast hosts. Tutor Intelligence, based primarily in Boston with about 160 people, trains frontier robot AI models, manufactures robots at scale, and directly deploys them into the largest supply chain businesses across the United States. Josh started Tutor about five and a half years ago with the core thesis that deploying robots into the physical economy would build the necessary data sets, models, and supply chain. Tutor robots have helped manufacture and ship over a quarter of a billion products to American consumers. Most customers are greenfield buyers who have never purchased a robot before, with 9 out of 10 US factories and warehouses lacking robots. Josh explains that robotics requires a full-stack platform incorporating model, hardware, software, supply chain, customers, and product, rather than just solving one narrow layer. The company's name stems from connecting human teleoperators (tutors) to robots to embody them with intelligence, creating an internet of robot training data that mirrors how the internet provided data for LLMs. Over time, physical intelligence models, 3D spatial data, and simulation tools help accelerate robot capabilities. Tutor manufactures its own hardware in Watertown, Massachusetts, to co-design robots with frontier models. They produce two main robot types: 1. **Cassie**: A brute force, single-arm robot powered by foundation models (200 to 400 million parameter range) using suction cups to handle heavy-duty tasks like transporting pallets and cardboard boxes, addressing backbreaking physical work. 2. **Sunny**: An industrial semi-humanoid with wheels, a higher payload capacity, faster speed, and industrial collaborative safety standards, operating in Boston data factories and real-world sites for dexterous bimanual manipulation. A hundred Sunny robots operate in Boston, and a growing fleet is deployed in the real world. Customer adoption is driven by offering robots for 16 bucks an hour with zero economic risk on the customer. The average time from the first call to a signed contract is 40 to 60 days, and from signed contract to deployment is about 40 days, with production standup often within hours. Small and mid-market businesses as well as Fortune 50 and Fortune 100 companies utilize them. Operations teams view the robots as reliable equipment once operational, and human-robot interaction (HRI) techniques—such as naming robots after customer children, football players, and pop stars upon repeat orders—cultivate strong relationships with workers. Looking forward, Josh discusses the evolution of the global supply chain, noting that in 2026, the global supply chain remains interconnected and interdependent. While China invested heavily in the previous regime of industrial automation, U.S. companies are building greenfield capacity using generally capable robots without needing to tear down existing infrastructure. Tutor continues to verticalize, bringing more manufacturing and componentry in-house to Watertown, Massachusetts, and investing heavily to build a mature domestic robotic supply chain. Tutor's research program leverages a 100-robot fleet called Data Factory One in Boston to perform real-world reinforcement learning at scale. The company is actively hiring researchers, engineers across every layer of the stack, and technicians to build robots. Ultimately, Tutor aims to provide a general-purpose labor substrate that transforms the physical economy, makes goods cheaper, and helps businesses scale physical operations.