MIT's FLEX lets robots infer the broader intent behind spoken instructions

Andreea Bobu's CLEAR Lab at MIT built FLEX, a pipeline that edits a robot's motion-planning features from conversational commands and uses GPT-4o to generalize them into wider rules, such as avoiding all fragile objects after being told to avoid a mug. In simulated arm tasks it took about five seconds and could make home and assistive robots easier to direct.

Summary written by DroidBazaar. The original article belongs to its publisher.