Computer-use agents and self-clicking AI browsers make for great demos but struggle to offer obvious utility to most people.

Stanford researchers have become the first to demonstrate that machine-learning control can safely guide a robot aboard the ...