MIT FLEX Lets Robots Reason Beyond Literal Commands
AI-generated summary Researchers from MIT and New York University introduced MIT FLEX on November 9, 2026, an artificial intelligence learning pipeline that enables physical robots to interpret conversational human instructions and infer unstated common-sense rules in roughly five seconds. The research team evaluated FLEX across three simulated manipulation scenarios designed to mirror household environments: serving coffee, moving a lit candle, and transporting a heavy bowling ball.