Researchers from MIT and New York University introduced FLEX, an AI-assisted learning platform intended to help robots interpret broader intent behind natural-language commands. According to the report published Nov. 9, 2026, the work was to be presented at the annual Conference on Robot Learning that month. The study was led by Andreea Bobu, an assistant professor of aeronautics and astronautics at MIT, with co-authors Audrey Lee of MIT and Ilia Sucholutsky of NYU.

FLEX uses a three-step process: adding features named in a command, generalizing with a large language model (GPT-4o), and masking objects no longer present. The team said the process takes about five seconds. In simulations, the robot carried out tasks including serving coffee, moving a candle and carrying a bowling ball; volunteers typed instructions such as avoiding spills near a laptop. With FLEX, the robot followed the commands literally and also reasoned more broadly, such as avoiding electronics and fragile objects.

The research was demonstrated only in simulated scenarios. It was partly supported by the MIT Generative AI Impact Consortium and MIT's Social and Ethical Responsibilities of Computing program.