Tutor Intelligence augments the B-Series to accommodate scale model training and robot deployment

Real-world data, human “tutors” and simulation are all part of the recipe for training AI models to enable next-generation robots to tackle an ever-wider variety of tasks. Tutor Intelligence Inc. said today it has raised $75 million in Series B funding for work in general robotics.

“It’s not just that Tutor models allow robots to do jobs that robots have never been able to do before,” said Josh Gruenstein, co-founder and CEO of Tutor Intelligence. “This type of intelligence also makes robots accessible on a whole new level. We’re unlocking much greater access to the solutions that businesses need to survive and that make basic goods cheaper for everyone. And the demand is there. We can hardly build robots fast enough.”

Tutor Intelligence was born in 2021 from the Massachusetts Institute of Technology Computer Science and Artificial Intelligence Laboratory (MIT CSAIL). The Watertown, Massachusetts-based company said it is building a full technology stack to “bridge the gap between what robots can do and what people can do.”

The data flywheel accelerates robot implementation

“AI has transformed the digital world, learning to converse and code from the vast amount of text and images on the internet,” Tutor Intelligence said. “The physical world has no precedent to train on, and AI has yet to transform it in the same way.”

The company has deployed fleets of internet-connected robots in manufacturing environments, where they perform real-world work for customers with input from remote operators known as “tutors.” It uses real-world data to train better frontier models, ultimately making robots more capable.

Tutor said its “full-stack platform” and “data flywheel” accelerate deployment, making robots more scalable, safer and more cost-effective. Instead of a robot-as-a-service (RaaS) model, companies pay the Tutor by the hour or on a per-fee basis, the same way they pay for manual labor.

Customers don’t make upfront purchases and don’t sign long-term contracts, Tutor added. The company said it handles implementation and maintenance and regularly pushes out software and intelligence updates. Tutor said his approach is bringing automation to tasks that have long challenged conventional, deterministic robotics. Earlier this year it opened a “Data Factory” to evaluate policies for 100 physical robots.

Tutor Intelligence catches on with Sonny and Cassie

Tutor’s Data Factory includes the Sonny two-armed semi-humanoid robots for every collection. Cassie is a mobile manipulator designed for collecting and palletizing boxes.

The company reported that twice as many of its robots are working in nearly four times as many facilities as a year ago. It said it is on track to have more than 1,000 robots under contract by the end of 2027, all of which will be built and distributed from the United States (tutors may be overseas).

Robots are currently moving, lifting and manipulating objects across some of the largest manufacturing and logistics networks in the country. They handle breakfast cereals, snacks, candy, cosmetics, household items, toys and more.

Tutor noted that its growth reflects the maturation of enterprise adoption of physical AI, as companies move from pilots to expectations of a return on investment from automating their operations. Nearly a third of organizations that received Tutor robots in the last 12 months were existing customers deploying multiple robots.

Funding to support greater training and production

Andreessen Horowitz led Tutor Intelligence’s Series B round, bringing its total capital raised to $117 million. Union Square Ventures and Neo also participated, leading the company’s Series A and seed rounds, respectively.

“Tutor’s insight is that implementation is not only the destination of robotics research, but also the way to overcome its limitations,” said Raghu Raghuram, general partner at Andreessen Horowitz and former CEO of VMware.

“Instead of asking the real world to adapt to search, Tutor starts with what real-world implementations require and works backwards through deployment, hardware, and search,” he added. “Continuous learning across an ever-widening range of implementations is the path from specialized capabilities to general-purpose robotics.”

The tutor said that his latest investment is in the formation of the Frontier Robots foundation model, as well as the production and implementation of Sonny and Cassie. It announced the latest models of its robots and underlying platform last month.