MIT system lets a robot arm figure out that 'don't spill on the laptop' also means keep coffee away from the phone
The method borrows an AI chatbot's common sense so people don't have to spell out every rule. So far it has been tested only in computer simulations.
Engineers at MIT and New York University have built a system that helps a robot grasp the larger point behind a casual command, according to an MIT News release published Nov. 9. Told “Don't spill on the laptop!” while carrying coffee, a robot arm also steered around a nearby phone, reasoning that liquids and electronics don't mix.
How it works
The system, called FLEX, works in three steps, the release says. First, the robot adds the named object, such as the laptop, to its list of things to avoid. Next, it asks a large language model what the objects on that list have in common. A large language model is the kind of AI behind ChatGPT; it learns patterns from huge amounts of text. This study used OpenAI's GPT-4o. Last, the robot's camera checks which objects are gone from the table and drops them from the list. The process takes about five seconds.
What it changes
Robots plan their movements with math, and older methods simply add one more object to avoid each time a person corrects them, the researchers said. That list grows long and goes out of date. FLEX turns a few corrections into a broader rule, such as staying away from anything fragile or flammable.
Volunteers typed commands to a simulated robot arm that served coffee, moved a candle and carried a bowling ball. No physical robot was tested. The study was led by Andreea Bobu, an MIT assistant professor of aeronautics and astronautics, and will be presented this month at the Conference on Robot Learning, a yearly research meeting.
Source: MIT News release, Nov. 9, 2026.