Robots Are Learning Faster—Should We Be Worried?

Robots Are Learning Faster—Should We Be Worried?

If robots can now open doors in any environment, what’s stopping them from entering your home uninvited?

Researchers have developed a set of AI models, called robot utility models (RUMs), that allow robots to perform tasks in new environments with a 90% success rate — without the need for further training.

The team, made up of members from New York University, Meta, and Hello Robot, created five models capable of completing basic tasks like opening doors and picking up objects. Traditionally, robots required extensive data and retraining in every new setting, but these models change that.

Using a simple tool—a reacher-grabber stick attached to an iPhone — they gathered data from various homes in New York and Jersey City to train the models. Testing the RUMs on a robot called Stretch, the researchers improved the success rate by integrating GPT-4o for task verification.

With this leap forward, the dream of downloading robotic skills and deploying them in homes could soon become a reality. Are we prepared for a world where robots adapt this quickly?

Read the full article on MIT Technology Review.

----