Tutor Intelligence CTO Alon Kosowsky-Sachs and CEO Josh Gruenstein with Alla Simoneau from the AWS Generative AI Innovation Middle, at Tutor’s Digital Manufacturing facility. Credit score: Eugene Demaitre
For robots to grow to be extra adaptable, they should be taught as people do however at scale, based on Tutor Intelligence Inc. The corporate has constructed DF1, its “kindergarten” or Information Manufacturing facility of 100 bimanual manipulators. Together with distant teleoperators or “tutors,” they’re coaching its Ti0 vision-language-action, or VLA, mannequin.
“I’ve been constructing robots since I used to be 9 years outdated,” stated Josh Gruenstein, co-founder and CEO of Tutor Intelligence. “Not like LLMs and the Web, there’s no equal of Wikipedia for robots, and we want a mass switch of intelligence from 8 billion people. DF1 is just not about constructing fashions; it’s find out how to get the appropriate information from individuals educating robots.”
In 2021, Gruenstein based the company with Alon Kosowsky-Sachs, now chief expertise officer, out of the Massachusetts Institute of Expertise’s Laptop Science and Synthetic Intelligence Laboratory (MIT-CSAIL). In December 2025, the corporate raised $34 million in Collection A funding, and it held an open home final month.
Tutor Intelligence was a part of the primary cohort of the Bodily AI Fellowship, receiving mentorship and help from Amazon Internet Companies, NVIDIA, and MassRobotics. It has continued to work with AWS for large-scale cloud compute and with NVIDIA CUDA for modeling.
Editor’s word: On the 2026 Robotics Summit & Expo on Could 27 and 28 in Boston, MassRobotics will showcase startups within the second cohort of the Bodily AI Fellowship. Registration is now open.
Information Manufacturing facility trains bodily AI with human assist
“We intention to create typically succesful robotic AI in partnership with the economic world,” stated Gruenstein. Tutor’s purpose is to determine a “virtuous cycle” wherein robots are commercially deployed to get information at scale and proceed coaching.
Moderately than prepare its robots largely by means of simulation, Tutor Intelligence has arrange DF1 to achieve real-world information at its new headquarters in a historic mill in Watertown, Mass. The startup claimed that, to its data, DF1 is the most important “robotic information manufacturing facility” within the U.S.
The 100 Sonny semi-humanoid robots began with piece-picking duties frequent to e-commerce, kitting, and different business functions. Like all group of kindergartners, they had been clumsy at first with the buyer packaged items like sponges or luggage of snacks.
In just some weeks, nevertheless, the robots have already demonstrated the worth of Tutor’s strategy to gathering real-world information, with supervision by 45 to 50 tutors in Mexico and the Philippines utilizing proprioceptive teleoperation, in addition to staffers on website. They use the same scoring methodology to the one which Gruenstein developed as a highschool scholar over a decade in the past.
“By evaluating the identical coverage throughout all 100 robots, we’re in a position to detect and proper robotic behaviors 100x quicker,” stated Tutor Intelligence. “An edge case that will usually require eight hours of robotic operation to note will likely be seen in solely 5 minutes of DF1 operation.”
Gruenstein likened DF1 to the Giant Hadron Collider in that it’s “an instrument of discovery and scientific exploration for scaling humanoids.”
AWS has supplied a staff of technical strategists and scientists to assist startups like Tutor Intelligence, famous Alla Simoneau, bodily AI expertise lead on the AWS Generative AI Innovation Middle.
“We give assist from ideation to manufacturing,” she informed The Robotic Report. “The Bodily AI Fellowship is exclusive in that our specialists help the startup neighborhood to securely and securely go to market.”
To cut back inconsistency in fleet-scale information for “industrial intelligence,” Ti0 is educated with “velocity normalization,” which Tutor described as a “novel preprocessing methodology that aligns the velocity profiles of demonstrations throughout completely different teleoperators.”
“By way of DF1 and Ti0, our purpose is to bootstrap the world’s first business humanoid deployment flywheel on the Sonny platform, unlocking compounding coverage enchancment and real-world worth over time,” stated the corporate in a weblog post.
Gruenstein stated he’s optimistic that Tutor can transfer from coaching to pilots by the tip of this 12 months. To realize velocity to scale, the corporate has used the identical collaborative robotic {hardware}, sensors, and contract producers for Sonny as for its single-armed Cassie system.
Cassie is designed for quick deployment
As Tutor Intelligence demonstrated at MODEX 2026 final month, the newest model of Cassie will be deployed in two days and handles bins weighing as much as 50 lb. (22.6 kg). Due to years of knowledge assortment and coaching that proceed in the present day, the case-picking and palletizing robotic can deal with a steady movement of SKUs with minimal changeover time and carry out at as much as 14 cycles per minute.
Cassie makes use of imaginative and prescient and integrates with exterior sensors utilizing the EVE native I/O module. Its security options together with lidar-informed zones and compliance with ISO/TS 15066 requirements. The cellular manipulator can transfer from station to station.
The system’s pill interface consists of help for a number of languages, and it doesn’t require coding experience or the setting of waypoints, asserted Tutor. The robotic will also be washdown-ready for meals dealing with.
Since Sonny and Cassie share elements, will Tutor’s bimanual selecting converge with its cellular manipulator?
“Proper now, we’re engaged on coaching the fashions for jobs the robots can do,” replied Kosowsky-Sachs. “At MODEX, we had been stunned to listen to individuals say, ‘I need this yesterday.’ Cassie cannot solely be helpful for big 3PLs [third-party logistics providers], however it’s additionally for different clients.”
Utilization-based pricing aligns Tutor growth, buyer wants
Tutor Intelligence costs $14 to $18 per hour for Cassie’s companies, which it stated is competitive with more and more scarce guide labor.
“Robotics as a service [RaaS] usually has hidden catches, so we speak by way of usage-based pricing,” stated Gruenstein. “For us, innovation is the expertise to get to a lower cost level. We now have a ‘no contract’ or ‘no dedication’ mannequin, which aligns our incentives to construct an clever robotic workforce and associate with companies, quite than integrators, which make cash by supporting complexity.”
BetterBody Foods has added the palletizer to its amenities in Utah and Massachusetts.
“For the deployment course of, Tutor lived as much as its phrase,” stated Jeff Pulley, facility supervisor on the firm. “The toughest half concerning the set up was organising {the electrical} and airdrop. With their staff onsite, we received the robotic plugged in and up and operating in a couple of hours. Now, it will possibly construct on a skewed pallet or straighten bins.”
Pulley informed The Robotic Report that he appears ahead to utilizing Sonny to run throughout a number of shifts. “It was cool to see the progress,” he stated after touring DF1.
Productiv Inc. conducts extremely variable kitting for e-commerce, medical system, and cosmetics clients. It assembles 30 million kits per 12 months with 10 to fifteen gadgets every. The Richmond, Va.-based firm checked out force- and power-limited robots and humanoids earlier than getting began with automated palletizing utilizing Cassie.
“We didn’t need to spend $150,000, however $14 per hour was aggressive,” famous Paul Baker, chief monetary officer at Productiv. “The robotic was worthwhile from Day 1 and retains up the road velocity.”
Productiv has 4 metrics: SKU protection, placement accuracy, pack time, and the way lengthy it takes to program a brand new SKU. Tutor met all of them, from 99.9% high quality and serving to to ship 350,000 kits to lowering programming time from 5 days to 1, Baker stated.
Tutor Intelligence can be working with main manufacturers and a significant 3PL. The corporate is hiring for roles together with analysis and software program engineers, buyer success managers, and salespeople.
The submit Tutor Intelligence builds Information Manufacturing facility to coach robotic AI in the true world appeared first on The Robotic Report.
