• MIT graduates launch Tutor Intelligence to create essential training data for robots performing everyday physical tasks.
• America’s largest robot data factory operates in Watertown, Massachusetts, described as “kindergarten for robots.”
• Robots practice simple actions like picking items, packing boxes, and folding laundry but frequently make mistakes.
• Founders aim to build adaptable robots capable of learning new skills in any environment.
• Technology expected to advance rapidly, with robots becoming common in daily life within five years.
Watertown, Massachusetts — Artificial intelligence is advancing into the physical world as robots at a specialized factory learn to handle simple human tasks that lack existing training data.
Co-founders Josh Gruenstein and Alon Kosowsky-Sachs, MIT graduates from the Computer Science and Artificial Intelligence Laboratory (CSAIL), established Tutor Intelligence to address a key gap in robotics development. While language models like ChatGPT draw from vast written sources, no comparable dataset exists for teaching robots physical actions.
Gruenstein, the company’s CEO, explained that most people have yet to encounter robots in daily life, but that is set to change quickly. At their Watertown facility — the largest robot data factory in the United States — the machines practice fundamental movements in what he calls “kindergarten for robots.”
The robots attempt basic chores such as picking up individual items and placing them into boxes or trying to fold laundry. In these early stages, they often mess up because the necessary training data for varied physical human tasks simply does not exist yet.
Gruenstein noted the contrast with specialized factory robots that repeat identical motions reliably, like those in car manufacturing. Most real-world physical work, however, requires flexibility and adaptation to changing conditions.
The goal is to develop robots that can learn any behavior in any situation, much like humans do. “For robots to do all of the jobs and kind of engage with our world in the same way that humans do, they have to be able to learn new skills on the fly,” Gruenstein said.
The team is actively generating the data needed to train these systems alongside dozens of robots working together. While the machines remain clumsy for now, rapid progress is anticipated.
This Watertown initiative represents a foundational step toward more capable and versatile robotics. Gruenstein predicts that within the next five years, robots will become a visible and exciting part of everyday technological and social shifts.