While I'm all for a positive attitude and hope they have significant improvements up their sleeve, please all, stop with the delusional references to the 6B miles.
It isn't the 6B fleet miles coming up,
nor is it the total autopilot miles,
nor is it the AP2 miles.
It's the 6B level 5 capable while still being supervised by humans miles to prove their implementation, starting when this capability is deployed sometime in the future!
And it might not be 6B miles. That particular number seems to have started with Elon in the
Secret Master Plan Part Deux.
We expect that worldwide regulatory approval will require something on the order of 6 billion miles (10 billion km). Current fleet learning is happening at just over 3 million miles (5 million km) per day.
I'm taking the fleet learning to be a reference to the Tesla vehicles with AP hardware installed (EAP enabled or not), that are collecting driving data and watching over our shoulders while we're driving.
The thing is, he provides no further evidence or support beyond what he says, that 6B is the right number to achieve worldwide regulatory approval. Nor do I take what he said as being the standard that will achieve the outcome - rather I take it to be directionally accurate and what he believed to be the order of magnitude of the effort, as of the time of writing (July 2016).
The reality is that nobody, including Elon, knows today what level of activity and fleet learning will be needed to achieve worldwide regulatory approval and initial implementation. We can guess, as Elon has done.
My guess is he's short by 1 or 2 orders of magnitude (60 or 600B miles).
One reason I'm guessing that, is that if you study vehicle accident statistics and other driving metrics, when units are needed, the standard unit of measure today (US) is units of 100 million miles driven. So deaths for 100 million miles driven, as an example. If 100 million miles is a standard unit of measure, then 6B miles represents 60 units of work; thus my guess that Elon is 1 or 2 orders of magnitude short of the fleet learning activity needed for FSD.
(Sidebar - some of the numbers per 100 million miles driven are getting small enough, that I'm expecting the unit of work to change to billion miles driven. If that happens, then 6B miles represents 6 units of work. Oh - and if we're thinking the autonomous drivers need to be an order of magnitude safer, then that further sounds to me like 6 units of work just isn't enough to prove it out. It's enough to say the results are encouraging, but that doesn't sound like enough to me).
What I love about Tesla's approach is I don't need to wait 2-4 years, or 2-4 decades, to start gaining some of the benefits today. I can get driver assistance (adaptive cruise, lane keeping) in a usable and improving form today. The work that gets us closer to FSD will also approve my fancy cruise control along the way (and my use of the fancy cruise control along the way will get us all to FSD sooner).
In fact, it's the approach that looks to me like it makes FSD economically achievable. The rest of the competitors that are working this with car fleets measured in the 0's or 00's - those look like approaches that won't be able, economically, to get to the end result. They are all going to have to find a way to do what Tesla is doing today - putting cars on the road with some of the functionality, and trick humans into using it, so the data can be collected and fleet learning can occur. For free (or better - pay the company, instead of being paid by the company).
I don't know of anybody taking a rules based approach to FSD - I believe everybody is doing some form of supervised learning, and that means training data to train and validate the models.
