“Our goal is, and I feel pretty good about this goal, that we’ll be able to do a demonstration drive of full autonomy all the way from LA to New York, from home in LA to let’s say dropping you off in Time Square in New York, and then having the car go park itself, by the end of [2017]”
At this pace the coast-to-coast test will take centuries :-)
I would love to know what Tesla's internal roadmap for autonomous driving is. Because there's no way that internally they're still going "just one more tweak"
The way the metric is designed is dumb so they do their testing elsewhere. It is a California rule. Basically if you wanted to game the metric you could just drive an autonomous car around a circular track and have no disengagements. But anyone trying for true level 5 is going to have to have lots of disengagements along the way towards training the system. So if Tesla put their actual numbers it would look really bad because city driving is very different than highway driving. And of course all the Tesla haters would go around talking about how their FSD system sucks which would hurt their stock price which would hurt their ability to raise money which would reduce their R&D spending.
It makes a clear to be argument, which is not about Tesla but about the parent comment: that what the parent wrote as an apology for Tesla and against the "haters" doesn't seem to put Tesla in better light. If anything it validates the anti-Tesla complaints.
My comment wasn't meant to be some kind of flamebait (and I find the accusation rude and ad-hominen: you could have avoided it, and just say that my comment appears to not make any argument to you - which is not the same as a flamebait). I merely pointed out my observation about the parent's comment.
The amount of progress done in machine learning in the last 2 years is crazy. People who think of Waymo vs Tesla vs Cruise vs AutoX underestimate how much of the progress was made by researchers developing new algorithms in the open.
Tesla is working hard on moving from camera view based algorithms to estimatin birds eyed view using self-supervised learning (just like all competitors I guess). We just started to be able to separate moving objects from static objects on videos efficiently.
It would surprise me if by 2030 there wouldn't be an open source self driving system with open model and dataset.
13 comments
[ 3.1 ms ] story [ 39.0 ms ] threadAt this pace the coast-to-coast test will take centuries :-)
So, in practice, they are probably constantly "lets see what happens after x more tweak".
Then Elon takes the time estimate of x more tweaks and announces it as "FSD by this date".
Well, nothing of what you wrote looks like an argument for why they would be wrong in saying that.
My comment wasn't meant to be some kind of flamebait (and I find the accusation rude and ad-hominen: you could have avoided it, and just say that my comment appears to not make any argument to you - which is not the same as a flamebait). I merely pointed out my observation about the parent's comment.
If Tesla wants to achieve 3, 4, 5, it'll need a place to learn without being crucified for it.
That's what I got from the comment.
Tesla is working hard on moving from camera view based algorithms to estimatin birds eyed view using self-supervised learning (just like all competitors I guess). We just started to be able to separate moving objects from static objects on videos efficiently.
It would surprise me if by 2030 there wouldn't be an open source self driving system with open model and dataset.