Just posted to my Forbes column about a fascinating shift in how researchers are thinking about training robots.
Instead of relying on ever-larger AI models with more parameters and more data, some scientists are exploring ways to help robots learn more like humans.
I spoke with Imperial College London robotics lab director Edward Johns about a new method that lets robots learn up to 1,000 real-world manipulation tasks in less than a day, sometimes from just a single human demonstration.
(Check out our conversation here.)
The approach breaks complex actions into reusable pieces and transfers what the robot learns from one task to another, dramatically improving learning efficiency without massive datasets or billion-dollar budgets.
“If robots are to become genuine alternatives to humans, they probably have to learn quickly like humans,” Johns told me.