Particularly since it is rooted in the fallacy that chores are purely vision based.
Driving a car has been tricky, but at least vaguely plausible because alit is a vision/audio based task with course inputs. General manual labor gets into more complex dextrous manipulation and feeling, and machine learning isn’t great at extrapolating vision to things that aren’t vision based. They don’t watch a steering wheel with a camera, they track exactly what forces are provided. Watching a hand do something isn’t that helpful to make a machine move a hand in a similar way.
It is why despite the stocastic parrots making for some incredible looking interactions, human shaped robots still look useless in the most advanced demos they have mustered.
Particularly since it is rooted in the fallacy that chores are purely vision based.
Driving a car has been tricky, but at least vaguely plausible because alit is a vision/audio based task with course inputs. General manual labor gets into more complex dextrous manipulation and feeling, and machine learning isn’t great at extrapolating vision to things that aren’t vision based. They don’t watch a steering wheel with a camera, they track exactly what forces are provided. Watching a hand do something isn’t that helpful to make a machine move a hand in a similar way.
It is why despite the stocastic parrots making for some incredible looking interactions, human shaped robots still look useless in the most advanced demos they have mustered.