Note that NoA/city and recognition of stoplights and stop signs are only the first steps toward driving in the city as well as EAP presently does on the highway. There are a million edge cases where the driver will have to take over. And even just Level 2 NoA/city is immensely more difficult than NoA/freeway because on the freeway it's a reasonable assumption that other cars will continue to move with the traffic, changing speed gradually and changing lanes occasionally. In the city you are constantly anticipating other cars' actions. How will it respond to a driver at a 4-way stop who does not move when it is their turn? Will it be so over-cautious about turning left across traffic that it takes four times as long to make the turn as you would? We've all been in the situation where someone moves out of turn, or doesn't move when it is their turn, and then both cars start to move at the same time, and then both stop, and finally one makes a hand gesture to let the other one go. Will the car recognize such hand gestures? Or will you have to take over at 2 out of 3 intersections?
Tesla will gather data from billions of miles driven, but how long will it take to develop the software that can adequately respond? Neural Networks seem to show remarkable learning ability, but they are in their infancy, and so far have mostly been used for things like learning to play games with very specific rules. We really have no understanding of their potential for tasks that have not yet been automated.
Landing an airplane is very difficult for a human because the plane must be lined up precisely with the runway, and speed and glide angle are critical. Such things are easy for a computer. And when an airplane is landing, the control tower has given it clearance and there will be no other plane in its way. But the driver of a car has to interact constantly with other drivers, anticipating their actions and responding according to their expectations. Computers are untested in this kind of situation.
Ten billion miles of data are useless without a program that can make good decisions. Don't get me wrong: I think we'll get there. But the task is far more complicated than some seem to think, and Elon is the worst person to say how long it will take because he is such an optimist. This is good because maybe it's the only reason we have Tesla. But it is risky to take his timeline predictions at face value.
I think that @AlanSubie4Life is spot-on when he says that the first people with self-driving cars will not be the ones who buy the FSD package today; they'll be the ones who buy the new cars that have the necessary hardware, because our cars don't, and it was simply wrong to say that our cars had all the needed hardware before the software existed. Before you have the software you cannot know what hardware will be needed.
Tesla will gather data from billions of miles driven, but how long will it take to develop the software that can adequately respond? Neural Networks seem to show remarkable learning ability, but they are in their infancy, and so far have mostly been used for things like learning to play games with very specific rules. We really have no understanding of their potential for tasks that have not yet been automated.
Landing an airplane is very difficult for a human because the plane must be lined up precisely with the runway, and speed and glide angle are critical. Such things are easy for a computer. And when an airplane is landing, the control tower has given it clearance and there will be no other plane in its way. But the driver of a car has to interact constantly with other drivers, anticipating their actions and responding according to their expectations. Computers are untested in this kind of situation.
Ten billion miles of data are useless without a program that can make good decisions. Don't get me wrong: I think we'll get there. But the task is far more complicated than some seem to think, and Elon is the worst person to say how long it will take because he is such an optimist. This is good because maybe it's the only reason we have Tesla. But it is risky to take his timeline predictions at face value.
I think that @AlanSubie4Life is spot-on when he says that the first people with self-driving cars will not be the ones who buy the FSD package today; they'll be the ones who buy the new cars that have the necessary hardware, because our cars don't, and it was simply wrong to say that our cars had all the needed hardware before the software existed. Before you have the software you cannot know what hardware will be needed.
