FSD Beta v11.x

EVNow · Feb 6, 2022

whitex said:
Yes, you can. There can be a rather large difference in the burden to prove vs. disprove something. More information, including mathematical notation you might remember from math, since stats 101 probably did not cover mathematical proofs.

Mathematical Proof/Methods of Proof/Counterexamples - Wikibooks, open books for an open world

en.wikibooks.org

So, let’s say in your LA to NY example, there are 3 faults. What does that prove ?

whitex · Feb 7, 2022

EVNow said:
So, let’s say in your LA to NY example, there are 3 faults. What does that prove ?

If the car required even a single human intervention to correct those faults, it didn't make it, so it's not as good as an average driver who would have made it. Of course you could account for some percentage of human drivers who also would have not made a coast to coast trip, so perhaps you should send 2 Tesla cars (not together, perhaps one from L.A one from Seattle) since chances of 2 average drivers starting from Seattle and L.A. and not making it to N.Y. are fairly slim. Note that if the cars do make it however, that does not prove it's as good as an average driver (for that you need a much larger sample and detailed data), however if they both fail, it's a pretty good indicator that FSD is not as good as an average driver. If your argument that 2 average drivers driving coast to coast are not that unlikely to both not make it, send a 100 cars and calculate how many should make the complete trip based on how many cars an insurance company would expect not make the trip out of a 100 average drivers (I suspect that number of expected cars would be 99 to 100).

As it stands now, FSD cannot even handle the Boring Co. Las Vegas tunnel, which is extremely simplified driving scenario as compared to a cost to coast drive, and not subject to road regulations so no excuse for Elon that he's waiting for regulators (there are driverless trains in Vegas already, a Tesla in a tunnel should not require additional regulations).

Elon should focus on more attainable goals, like recognizing when a car is going into a carwash and automatically enable "carwash mode", an average human should be able to do that. I'm not saying it's easy to do, but way simpler than a downtown L.A. to downtown N.Y. trip. The Las Vegas tunnel driving would also be a good stepping stone. Walk before you run approach.

Knightshade · Feb 7, 2022

whitex said:
As it stands now, FSD cannot even handle the Boring Co. Las Vegas tunnel, which is extremely simplified driving scenario as compared to a cost to coast drive, and not subject to road regulations so no excuse for Elon that he's waiting for regulators

This is outright false.

The tunnel cars are absolutely subject to regulation (which is why they're so speed limited right now- even the # allowed at a time is regulated) and many articles have cited regulators as the reason they're not autonomous right now as well.

The financial pickle facing Elon Musk's Las Vegas Loop system | TechCrunch

Please see update below... Restrictions put in place by Nevada regulators are making it difficult for The Boring Company (TBC) to meet contractual targets

techcrunch.com

Clark County regulators have approved just 11 human-driven vehicles so far, set strict speed limits and forbidden the use of on-board collision-avoidance technology that is part of Tesla’s “full self-driving” Autopilot advanced driver assistance system.

Boring have since proven that even manually driven, with the low speed limit imposed, they can meet the required # of people moved per hour in their contracts (though in part because they've at least been able to increase the # of cars)-- but there's been no change in the regulators stance on the other restrictions last I heard.

Mardak · Feb 7, 2022

EVNow said:
I'm sure they want to spend most time on city and minimize highway regressions for the next few quarters

I totally agree FSD Beta City Streets quality has a lot more opportunity to improve relative to production Autopilot Highway quality, so it would be natural to focus more effort on the former. However, depending on the metrics that Tesla is trying to achieve, e.g., for convincing regulators FSD Beta is significantly safer than humans, Tesla might want to focus on "single stack" to include highway driving to better match typical commutes, which cover all types of roads including highways.

People have criticized Tesla's Vehicle Safety Reports because Autopilot typically is used on highways whereas the NHTSA average crash data is from all roads. So similarly, if FSD Beta's safety is evaluated only from City Streets driving instead of "inflating" safety with highway driving, Tesla is making the problem harder for themselves. On the flip side, achieving significantly better safety on the harder problem should mean it's actually safer too, and various companies have spoken up about The Disengagement Myth: "driving on a well-marked highway or wide, suburban roads is not the same as driving in a chaotic urban environment."

So yes I think FSD Beta 11 could be more useful and ready sooner if it skipped "single stack" to include highway driving, but there's probably practical business goals Tesla wants to achieve that might not align with consumer expectations.

aronth5 · Feb 7, 2022

Mardak said:
So yes I think FSD Beta 11 could be more useful and ready sooner if it skipped "single stack" to include highway driving, but there's probably practical business goals Tesla wants to achieve that might not align with consumer expectations.

Don't assume single stack is linked to business goals. Maybe yes or maybe not. But consider this approach would eliminate any improvements that can only be implemented in the "single stack" or can only partially be implemented. I have no idea if his is a valid concern specific to FSD but something to consider. I suspect it's likely though.

EVNow · Feb 7, 2022

Mardak said:
However, depending on the metrics that Tesla is trying to achieve, e.g., for convincing regulators FSD Beta is significantly safer than humans, Tesla might want to focus on "single stack" to include highway driving to better match typical commutes, which cover all types of roads including highways.

I don't think Tesla cares about convincing regulators FSD Beta is significantly better than humans.

In business terms - Tesla doesn't want to take liability for FSD or AP/NOA until they are internally convinced that it is a risk that is worth taking. SO, they would take on that liability only if the money they make on FSD/NOA is a lot more than expected payouts because of accidents. This would only happen if FSD/NOA is 10x or so better than humans.

Considering a serious accident could result in $60k in damages and FSD costs $12k - it would mean they can only allow max of 1 serious crash per life of 5 cars or about 500k miles. That is atleast 2x human driving failure rate. This is assuming breakeven and it costs Tesla $0 in marginal costs. Taking into account actual h/w and software costs, the margin is probably $8k. Then, if you want something significantly better than break-even, we are looking at FSD/NOA being 5x to 10x better than humans.

Irrespective of what Tesla claims in terms of regulators - these are the real business parameters at work here - not what fickle regulators might think. Until Tesla can convince their finance department & investors that risk of taking on liability for NOA/FSD is truly low - they won't be interested in submitting anything to regulators for unsupervised AV approval.

EVNow · Feb 7, 2022

whitex said:
If the car required even a single human intervention to correct those faults, it didn't make it, so it's not as good as an average driver who would have made it.

This is where learning basic stats would help you. Don't want to waste time teaching you basic Stats. Have a good day.

aronth5 · Feb 7, 2022

EVNow said:
I don't think Tesla cares about convincing regulators FSD Beta is significantly better than humans....

Well they do care but as you pointed out not until after they have convinced the people that matter internally that there is a good business case before proceeding to the regulators. That is pretty much how it works in any regulated industry when you're rolling out a new product. What helps Tesla is they are already collecting valuable metrics on FSD.

EVNow · Feb 7, 2022

aronth5 said:
Well they do care but as you pointed out not until after they have convinced the people that matter internally that there is a good business case before proceeding to the regulators. That is pretty much how it works in any regulated industry when you're rolling out a new product. What helps Tesla is they are already collecting valuable metrics on FSD.

Right - 11 is nowhere near that tipping point. Infact 10.10 needs to become probably 1000x better to just match human accident rate (1 in 10 vs 1 in 10,000 miles). So, I suspect they have a couple of years to go before convincing regulators becomes even a consideration.

ps : That doesn't mean Tesla doesn't want to convince the customers, investors etc. But they don't need single stack for that ....

Mardak · Feb 7, 2022

EVNow said:
I suspect they have a couple of years to go before convincing regulators becomes even a consideration

Tesla works with regulators even for driver assistance features, so this is well before Tesla takes liability for unsupervised/robotaxi systems. For example, EU regulations (following UN/ECE regulations) require driver assistance Autopilot to suddenly abort if it would take a curve too quickly, and separately CA DMV has questioned Tesla about driver assistance features like Traffic Light and Stop Sign Control.

I don't know what types of communications Tesla has with say Canadian regulators in preparations for deploying FSD Beta driver assistance, but having better metrics to share (or just ready in one's back pocket) probably provides a better argument for why it should be allowed. Even within the US, having good metrics could help in potential push back from NHTSA for when Tesla wants to release FSD Beta (still driver assist) more widely (without Safety Score requirements?) -- this would probably allow for recognition of some deferred FSD revenue, so Tesla has some financial incentive too.

Then again, I suppose Tesla probably didn't need to provide metrics for Autopilot/Autosteer safety before giving the features to customers. Although, it does seem like FSD Beta 10.9 rollout was halted due to NHTSA/regulators, so the timing of FSD Beta 11 could similarly be affected even though it's not robotaxi.

aronth5 · Feb 7, 2022

EVNow said:
ps : That doesn't mean Tesla doesn't want to convince the customers, investors etc. But they don't need single stack for that ....

They may need single stack if technical improvements to improve upon the human accident rate are dependent on the single stack architecture. Smart customers and investors will expect to see accident rate information plus Tesla will want to point to ever improving FSD accident rates as justification for FSD/Robotaxi/Insurance.

EVNow · Feb 7, 2022

aronth5 said:
They may need single stack if technical improvements to improve upon the human accident rate are dependent on the single stack architecture. Smart customers and investors will expect to see accident rate information plus Tesla will want to point to ever improving FSD accident rates as justification for FSD/Robotaxi/Insurance.

BTW, aren't you assuming a single stack would allow Tesla to call both NOA/City FSD by the same name and there by combing their accident / disengagement rates ? I'm not sure this would fly. When Tesla asks Canada for approval to introduce FSD Beta, they would have to provide (if at all) FSD Beta related stats. Not NOA+FSD Beta stats.

BTW, I don't see any reason why single stack is needed to improve quality. Only place where single stack might help is NOA to FSD City handoff (and the other way). Rest of it is just an internal architecture thing. This is the reason I keep saying its just a distraction (and apparently sounds nerdy so all the twitter folks use it).

RowdyMY · Feb 7, 2022

Been a few days since it was mentioned but still seems relevant to the convo. I’ve also brought it up before.

Tesla collects insane amount of data from its (our?) fleet of FSD beta vehicles.

If my car were representative of the entire fleet, based on the participation numbers from the earnings call, something like 2.5+ petabytes of data is sent back to Tesla every month.

Not sure what specific type of data is tagged and uploaded but they are collecting, and hopefully processing it.

whitex · Feb 7, 2022

Knightshade said:
This is outright false.

Really? Can you list which road regulations the cars in the tunnels are subject to? Or did you assume that the same regulations apply to all transport vehicles, be they cars on the roads, a subway underground, or monorail above ground, and can be referred to as road regulations?

whitex · Feb 7, 2022

Knightshade said:
Boring have since proven that even manually driven, with the low speed limit imposed, they can meet the required # of people moved per hour in their contracts (though in part because they've at least been able to increase the # of cars)-- but there's been no change in the regulators stance on the other restrictions last I heard.

Where did you get your data? Their goal was 57,000 passengers per hour.

Elon Musk's Boring Company 'Tesla in Tunnels' in Las Vegas to expand throughout the city

The Boring Company’s first commercial loop, deployed in Las Vegas, has officially been approved for expansion throughout the city. Tesla...

electrek.co

As for regulations, if the likes of Waymo, or even Cruise, can get regulators let them ferry people around without a driver onboard on city streets of San Francisco and other cities (yes robo-taxis are here already, just not Tesla ones), you'd think Boring should be able to convince the regulators to let them use in a tunnel without any pedestrians, intersections, weather, etc. Perhaps Boring/Tesla driving tech is not as convincing as the other technologies? Oh, and Nevada is actually leading in autonomous vehicle legislation, IIRC they were the first state to legalize completely autonomous cars.

Nevada continues to embrace an autonomous vehicle future – Sierra Nevada Ally

Legislation paves way for autonomous delivery vehicles

www.sierranevadaally.org

Knightshade · Feb 8, 2022

whitex said:
Really? Can you list which road regulations the cars in the tunnels are subject to?

I literally just cited a news story debunking your claim the lack of automation was NOT due to regulators.

It literally said it was due to regulators

Are you gonna avoid admitting your error by playing a semantic game about if a tunnel is a "road"?

Sad.

Knightshade · Feb 8, 2022

whitex said:
Where did you get your data? Their goal was 57,000 passengers per hour.

Uh... what?

Your link is about FUTURE plans for the REST of the system...not the existing convention center system.

You can tell if you actual read the story

Or the date for that matter.

Shame you didn't bother with either.

The convention center systems goal was capacity for 4400 per hour.

Which they did in a test of the system. Even driving manually, with the 30-something mph regulated speed limit.

The Boring Company LVCC Loop capacity exceeds 4400 passengers per hour [Video]

The Boring Company’s (TBC) test drives for its Las Vegas Convention Center Loop (LVCC) exceeded 4,400 passengers per hour, according to results revealed on May 29. Steve Hill, the Chief Executive Officer and President of the Las Vegas Convention and Visitors Authority (LVCVA), announced the...

www.teslarati.com

whitex said:
As for regulations, if the likes of Waymo, or even Cruise, can get regulators let them ferry people around without a driver onboard on city streets of San Francisco and other cities (yes robo-taxis are here already, just not Tesla ones), you'd think Boring should be able to convince the regulators to let them use in a tunnel without any pedestrians, intersections, weather, etc. Perhaps Boring/Tesla driving tech is not as convincing as the other technologies? Oh, and Nevada is actually leading in autonomous vehicle legislation, IIRC they were the first state to legalize completely autonomous cars.

Nevada continues to embrace an autonomous vehicle future – Sierra Nevada Ally

Legislation paves way for autonomous delivery vehicles

www.sierranevadaally.org

You don't understand why different regulators have different regulations?

(or apparently the difference between state and local regulations)

Yikes.

Bladerskb · Feb 8, 2022

Knightshade said:
I literally just cited a news story debunking your claim the lack of automation was NOT due to regulators.

It literally said it was due to regulators

Are you gonna avoid admitting your error by playing a semantic game about if a tunnel is a "road"?

Sad.

@whitex is right no where in the article does it say that Boring requested and went through the process to deploy autonomous vehicles and the regulators said no. It does however state that the regulators forbad them from using their collision avoidance tech (which is not autonomous) most likely due to the fact that 1) its in a tight closed space 2) its not autonomous and would be reliant on the driver to make instant corrective actions when it fails. 3) Due to #1, driver won't have the necessary time and space required to take over instantly and correct any mistake the ADAS makes.

However, TechCrunch has been told that Clark County regulators have approved just 11 human-driven vehicles so far, set strict speed limits and forbidden the use of on-board collision-avoidance technology that is part of Tesla’s “full self-driving” Autopilot advanced driver assistance system. Tesla’s Autopilot system technically does not rise to the level of fully autonomous, even though it is branded as such. It is considered — even internally, according to exchanges between Tesla and California regulators — an advanced driver assistance system that can automate certain functions.

Knightshade · Feb 8, 2022

Bladerskb said:
@whitex is right no where in the article does it say that Boring requested and went through the process to deploy autonomous vehicles and the regulators said no. It does however state that the regulators forbad them from using their collision avoidance tech (which is not autonomous) most likely due to the fact that 1) its in a tight closed space 2) its not autonomous and would be reliant on the driver to make instant corrective actions when it fails. 3) Due to #1, driver won't have the necessary time and space required to take over instantly and correct any mistake the ADAS makes.

However, TechCrunch has been told that Clark County regulators have approved just 11 human-driven vehicles so far, set strict speed limits and forbidden the use of on-board collision-avoidance technology that is part of Tesla’s “full self-driving” Autopilot advanced driver assistance system.

Remarkable the nonsense semantic gymnastics you trolls go through.

Story: Literally says they can't use anything but human driven vehicles due to regulators.

Trolls: IT DOES NOT TECHNICALLY SAY THEY CAN'T USE FSD SINCE THEY DID NOT USE THE WORD FSD!

momo3605 · Feb 8, 2022

Knightshade said:
Remarkable the nonsense semantic gymnastics you trolls go through.

Story: Literally says they can't use anything but human driven vehicles due to regulators.

Trolls: IT DOES NOT TECHNICALLY SAY THEY CAN'T USE FSD SINCE THEY DID NOT USE THE WORD FSD!

Are you arguing that if Tesla wanted they could deploy fully autonomous cars in the tunnels, and the only thing holding them back was regulators? I think it's pretty obvious Tesla's technology is nowhere near good enough for that. So what exactly are you arguing?

FSD Beta v11.x

Well-Known Member

Well-Known Member

Well-Known Member

Active Member

Long Time Follower

Well-Known Member

Well-Known Member

Long Time Follower

Well-Known Member

Active Member

Long Time Follower

Well-Known Member

Member

Well-Known Member

Well-Known Member

Well-Known Member

Well-Known Member

Senior Software Engineer

Well-Known Member

Active Member

Similar threads