lukestern:
Whoopsy:
 

Autopilot was and still will be a gimmick feature. As long as human drivers are still mixed in with automated cars, there won't be a perfect system. Autopilot features are passive, unlike a human which is active. You have been driving for a long time, you have the experience to look around for tell tale signs, like seeing traffic started to back up a few miles up the road on your lane and you change lanes, or you see the guy to the right moving his hand from the wheel to th turn signal, that tells you he intended to change lane in front of you and you back off the throttle to leave room. These are 2 prime examples why autopilot will never work, cameras, sensors cannot pick these up. You will also see a car coming up behind you fast and you move over, autopilot won't. The 'brain' or lack of in the Autopilot system cant process these things. 

The whole point with Neural Networks is just that they in due course will master every tiny little case that can happen out there. And with 500k cars (and growing) driving around right now training the NN they will eventually get there. Even the edge cases that you mention will be managed. If you eyes can see those things and you take a certain decision, the computer will be trained to act the same.

It will take many years to get there and Elons timeline is probably far off like always. But eventually it will happen and I would certainly not underestimate the power of Neural Nets.

Also, the Autopilot is far from being a gimmic. It's a really nice assistant system that get's better every month. It's actually quite impressive to see how far they have come and how much progress they have been made since i started out driving with Autopilot on my work commute. Now it manage most of my driving without any issues, that was not the case just a few months ago. When you understand how Autopilot works and why it behaves in certain ways, it is really impressive how it drives. Actually quite much as a human. A human that is learning and getting better and better over time.

It's important to understand that what Tesla customers have in their cars is not representative for what the system in development mode can achieve.

 

Tesla's cameras don't have the resolution to see movements inside other cars. Their system is actually better than LIDAR as those movements will be invisible to LIDAR which can only see outlines. Cameras could theoretically compare back to back images to check for movement.

Human brains process informations very differently the silicon brains, silicon can process simply tasks very quickly but it's useless again human brains in predicting the 'unpredictable' so to speak. Human learn from experiences, neural network wanted to work the same way but they are far far far from even at 1/100 of what a human brain can do.

Yes AI can beat chess masters, but there is only a finite umber of moves a piece can do, ad there are rules for chess, so it can be brute forced. But driving on the road, while there are suppose to be rules, human don't always obey and other humans also expect the others to to obey the rules, and that fact confuses any silicon brains, they really have no way to process and use the information they got from sensors. 

Autopilot is a fancy name for assisted active cruise control, nothing more. Like I said, most people consider it gimmicky and not a must have feature for automobiles.

Tesla is decades away from real autopilot. 

I think currently, the 'autopilot-ed' Tesla should be confined to perhaps bus lanes or something, a giant target for it to follow and it stays within one single lane and no need to worry about other cars merging in and out. It should also require the driver to hold the steering wheel at all times, and eyes looking straight ahead to look for danger.

 


--