Robots That Understand Ambiguity

MIT researchers have cracked the code on teaching robots to handle messy, real-world chores—like bringing coffee without crashing into your laptop—by using two clever language models. One decodes vague requests into precise actions, the other filters out distractions to focus on what really matters. This system, called Masked IRL, learns faster, needs less human demo time, and even masters new tasks after training. In tests, robots successfully navigated obstacles and executed nuanced commands. Next up: equipping them with cameras to see and prioritize objects visually—making them smarter, safer, and ready for homes and factories. Support the show:Get a discount at https://solipillow.com/discount/dnn. Advertise on DNN:[email protected] This is an automated, high-level news summary based on public reporting.Report issues to [email protected]. View sources & latest updates:https://sources.thednn.ai/cdeb37f2167c8b43