neroden
Model S Owner and Frustrated Tesla Fan
Good argument, I guess. The counter-argument is that until you have someone who knows what good driving looks like (hint: it doesn't look like the average car on the Interstate), you're not going to know what problems you need to solve. When you finally get the people in to work out how to make the car drive *well*, you may discover you have to totally redo your vision stack, because a really good driver is looking for something you weren't looking for.Until we're a lot closer to full self driving being reality it's just an engineering/computing exercise, so why take on the expense?
There are a lot of situations where the correct path to follow is *not obvious* to an average driver. Particularly when it's a question of whether to exit the pavement and go onto the shoulder, whether to go into the opposite lane, whether to exit the shoulder onto dirt, etc. This is HARD and most people get it wrong most of the time. Even me.
In some cases, it's literally "make a K turn and go to a totally different route".
If they have a few driving experts -- say three, with different backgrounds -- and they show each tough situation to them and go "OK, what should the car do here?", and go with the consensus (and also ask them what cues they're looking at, particularly if they disagree) -- then they'll be doing a good job at training.
That I agree on. Also, LIDAR is a total dead end.Fortunately for us as investors, Tesla seems to have the hardware on the right track.
From an investing point of view this is probably another example of "you don't have to run faster than the bear". If everyone fails, but Tesla is the best in the bunch, they probably still make a mint.
If Tesla has to restart their NN path choice training mostly from scratch, but everyone else doesn't even have the right hardware, Tesla definitely makes a mint.
Last edited: