Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register
This site may earn commission on affiliate links.
Are you figuring Tesla will start out with small geofenced operations like Waymo? Remote assistance is feasible when you’re running something like Waymo with 250 vehicles total and only about 100 on the road at any one time, but probably not realistic with a fleet of any significant size.
Yes - when Tesla starts their robotaxi network it will be geofenced small scale.
 
  • Like
Reactions: spacecoin
1 in 10,000 is human level minor accidents (like curbing). You can search where we have discussed this before ( the stats are from Virginia tech study).
Sure but that would not be the level where you’d say you could go robotaxi (with remote assist)! It’s just not the metric you’d be looking at.

It’s not clear at all that minor-accident (curbing) rates would even correlate with important-accident rates.
 
Sure but that would not be the level where you’d say you could go robotaxi (with remote assist)! It’s just not the metric you’d be looking at.

It’s not clear at all that minor-accident (curbing) rates would even correlate with important-accident rates.
You need to read the Virginia Tech study.

More importantly, once the disengagement rates get better, Tesla will be able to analyze the disengagements and figure out a much better "likely accident rate by type". This more than just one general disengagement rate would be the basis of any robotaxi deployment decision. I'm sure a group of data scientists and actuaries would deeply analyze the data before making decisions.

They have to figure out the likely accident & payout rates before starting any robotaxi service.

This is also true for unsupervised "level 3" FSD. This is why Elon keeps saying it has to be >10x better .... if you are getting paid only $15k for what is probably 150,000 miles of driving per lifetime of car - even if they pay out for one big accident per 10 cars / 1.5 Million miles of travel - that will wipe out any FSD revenue they got from those 10 cars - just $150k !

BTW, those who crib about the FSD cost - and think for that money they should be getting real level 3 unsupervised are not doing these calculations.
 
Last edited:
That sounds like a good question to pose to all the companies who aren’t doing it and appear to be not close to attempting it, or directly to Elon because he’s the one citing these numbers and I’m merely regurgitating them — with my own color in thinking that even 100x better or 10,000% safer may not be sufficient.
Again, why is 100x or more necessary? Let's assume for a moment (perhaps incorrectly) that "safer" directly translates to "fewer deaths". So if FSD is at 2x (NOT saying it is atm), then I'm not sure there is any real argument against using it. There is of course the risk associated with bad publicity when "EV catches fire!!" is a huge blaring headline even when 100's of ICE cars to that every day.

Of course, there are many caveats. For example, if FSD *refused* to allow the car to move AT ALL it would already be 100x better, since the chances of dying in a stationary parked car are quite low. :)
 
  • Funny
Reactions: EVNow
Again, why is 100x or more necessary? Let's assume for a moment (perhaps incorrectly) that "safer" directly translates to "fewer deaths". So if FSD is at 2x (NOT saying it is atm), then I'm not sure there is any real argument against using it. There is of course the risk associated with bad publicity when "EV catches fire!!" is a huge blaring headline even when 100's of ICE cars to that every day.
From the company liability perspective it makes a big difference.

At what x will the jury decide it's not the manufacturer fault?
 
From the company liability perspective it makes a big difference.

At what x will the jury decide it's not the manufacturer fault?
It will always be the manufacturer's fault if found liable/at fault regardless of ideas of theoretical reliability. It will just happen less often if better and not make manufacturer go bankwupt.
 
Because those are supervised.

That is why Tesla very closely tracks disengagement rates (according to the biography). When that gets to be better than human accident rates, they will declare supervision is not needed. That is still very far away as I keep saying - currently the disengagement rate is around 1 every 10 miles from crowd sourced data. When it gets to 1 in 10,000 miles, we are getting to unsupervised levels (1x human level). At that time Tesla can start robotaxi like service with remote assistance like Waymo.

When disengagement rate is 1 in 100,000 Miles I.e. 10x human level, Tesla can let you sleep in the car.
how is the 1 every 10 mile disengagement rate calculated? I seem to recall a lower disengagement rate being touted shortly after FSD was expanded to highway driving. If you're including highway driving then it will automatically be lower because the number if decisions per mile is much lower. one needs to look at city vs highway miles. You can further break it down to rural vs urban highway miles since the former is about the easiest task FSD has to handle.
 
  • Informative
Reactions: pilotSteve