The Robots Already Working in American Factories | Tutor Intelligence

Tutor Intelligence is putting robots to work in American factories and warehouses by combining robot development, manufacturing, and direct deployment in one business. Founded in 2021, the company has about 160 employees, primarily in Boston, and serves customers ranging from small businesses to Fortune 50 and Fortune 100 companies. Its robots perform manual supply-chain tasks at customer sites, while human operators called tutors can step in when needed. Josh Gruenstein’s central thesis is that paid deployments generate both operating experience and training data, creating a flywheel that neither models nor hardware alone can provide.

Tutor makes two platforms in Watertown, Massachusetts, using components including actuators and cameras from outside suppliers. Cassie is a heavy-duty, one-armed robot with suction cups for moving pallets and cases; Sunny is a wheeled industrial semi-humanoid built for indoor factory floors, higher payloads, faster movement, and work around people. Gruenstein estimates the pair could cover 80% to 90% of manufacturing-logistics work, but that is a company estimate, not a demonstrated coverage rate across all operations. Sunny is too industrial for the home, and Tutor’s current robots cannot handle outdoor surfaces such as gravel.

The company says nine out of ten US factories and warehouses have no robots, even though tasks such as moving cases, packing boxes, and assembly remain manual. Gruenstein describes some of these jobs as physically demanding and hard to staff, citing three-month turnover and an average worker age of 55. Tutor presents its robots at a price of $16 an hour, emphasizing work cost and quality over whether a human tutor or learned model is controlling the machine. The intended design life is three to ten years at human rates and reliability; this is a requirement for the product, not a claim that every deployed unit has achieved it.

Tutor reports an average of 40 to 60 days from first call to signed contract, about 40 days from contract to arrival at a customer site, and production work within a day—often within hours—of arrival. The company says it deploys general-purpose robots without customizing them or putting economic risk on customers. A deployment can expand from one robot at one site to a dozen or more across multiple sites if operations teams want the same work elsewhere. Tutor argues that this approach suits the mostly small and mid-market US manufacturing and logistics sector better than projects tailored over years to one large customer; Gruenstein says he has seen startups struggle after pursuing highly customized work about five to ten times.

Tutor’s claimed advantage is its deployment and data flow, supplemented by simulation, internet-scale data, and 3D sources such as CAD. The company has 100 Sunnies in Data Factory 1, its Boston facility for internal data collection and research, and says the fleet enables real-world reinforcement-learning work at scale. Gruenstein says reinforcement learning performs poorly with small batches and describes the facility as, to the company’s knowledge, the largest semi-humanoid data-collection farm in the United States. Tutor has moved from convolutional neural networks to frontier models for general robot intelligence, but the speakers remain uncertain about how well models will generalize beyond their training distributions. Whether to retain human tutors depends on model performance, operating complexity, and the cost of further engineering.

Tutor accepts tradeoffs across the robotics stack to get robots working economically, then aims to reduce those tradeoffs as the platform develops. It says manufacturing processes can be updated every two weeks, robot code ships daily, and major releases arrive every two weeks. Ongoing costs—including inference, teleoperation, maintenance, and break-fix—currently outweigh upfront costs, so the value proposition depends on how much work a robot can perform relative to its total cost. Gruenstein sees general-purpose robots as a possible way to expand US industrial capacity and eventually lower the cost of goods, while stressing that home use presents harder technical, safety, and economic problems than factory work. He also expects supplier relationships to evolve as volumes rise, with more components potentially built in the United States.