MIT and NYU researchers introduce FLEX for robot motion planning

MIT and New York University researchers introduce FLEX, a feature-editing pipeline that lets people give robots natural-language commands for motion planning, according to an MIT News report.

FLEX uses GPT-4o to generalize from a user's instruction, such as avoiding fragile objects, without the user specifying each case. The three-step process, Add, Generalize and Mask, takes about five seconds, and the team tested it in simulations of a robot arm. The researchers plan to present the study at the annual Conference on Robot Learning.