[redpoint5]

I was just thinking how most robots used in industry have motions programmed by humans. In fact, painting robots often record the motions that a skilled painter makes in painting the body of a car, and then minor tweaks to the motion are programmed in to correct any little errors. Then the motion is repeated when the production line is turned on.

Instead of relying entirely on autonomous cars having code robust enough to navigate unfamiliar roads, I was thinking a similar approach could be used. Have a learning mode where the human driver makes a typical commute to work a few times while the car learns where the road markings are, obstacles such as potholes, lane positions, etc before taking over. This would give the AI a jumpstart on learning the route and being able to anticipate some things in advance, instead of having to calculate on the fly every single time.