Fully autonomous robots are much closer than you think – Sergey Levine

Sergey Levine describes Physical Intelligence as building foundation models intended to control different robots across many tasks. The company's early demonstrations—folding laundry, handling a box, and attempting kitchen cleanup—are presented as basic building blocks, not as proof that general-purpose home robotics is solved. Levine's longer-term target is a robot that can manage ongoing household routines, use common sense, seek information, and recover from mistakes. He gives 2030 as a median estimate for fully autonomous household operation, but distinguishes that endpoint from the earlier milestone that matters to him: deploying a useful, narrowly scoped robot so it can collect real-world experience and start a self-improvement flywheel. He hopes something useful may be deployed in one or two years, while acknowledging uncertainty.

The proposed flywheel is not simply a model improving itself without oversight. Levine describes a human-in-the-loop stage in which people observe robot behavior, provide guidance, and help resolve failures. Physical tasks make some errors visible: a person can see that a shirt was folded poorly, intervene, and help the system avoid the same mistake. Language can also act as supervision, as in telling a robot to pick up a cup and put it in the sink. The hard research problem is to represent useful feedback and turn it into stable learning signals. Levine believes this is tractable, but says the details remain challenging; the timeline depends on finding a useful scope where robots can operate safely and gather experience.

Levine expects robot capability to expand gradually, much like coding assistants moved from completing small pieces of code toward taking on larger, more defined tasks. The analogy is about scope, not a sudden switch in which all workers are replaced. A robot might first make coffee or fold boxes, then take on a wider set of tasks as its common sense and repertoire improve. When pressed for a timeline, Levine gives five years as a reasonable median for household autonomy but emphasizes that the answer depends on unresolved research questions and on the right synthesis of existing ideas.