We Need FSD Full Self-Driving to Be Imperfect

[Pstryker]

Absolutely. If we want it to be better we must use it as much as possible. Give Tesla data and we all win.

Sponsored

 

[Darthamerica]

I think I'm talking right past you.

Why do you think fatalities are resulting from the FSD development program? The statistics clearly show supervised FSD results in less fatalities than a human driver alone!

I think you may be misunderstanding me. I am not saying FSD is more dangerous than humans or more likely to cause a fatality. My experience tells me that it is close to as safe as human drivers. However, it’s not perfect and an inevitability of that is going to be fatalities. That might be because the SW screws up. It might be because the user is an idiot. It might be no fault of FSD or the driver. But it has and will continue to happen because the road is a dangerous place!

Tesla will for obvious reasons put a lot of scrutiny on those incidents as will regulators. As a result, customers will get improvements to the overall system that might otherwise never happen.

 

[HaulingAss]

I think you may be misunderstanding me. I am not saying FSD is more dangerous than humans or more likely to cause a fatality. My experience tells me that it is close to as safe as human drivers. However, it’s not perfect and an inevitability of that is going to be fatalities. That might be because the SW screws up. It might be because the user is an idiot. It might be no fault of FSD or the driver. But it has and will continue to happen because the road is a dangerous place!

Tesla will for obvious reasons put a lot of scrutiny on those incidents as will regulators. As a result, customers will get improvements to the overall system that might otherwise never happen.

It looks like I'm still talking past you.

The statisitics show there are already fewer fatalities with Supervised FSD than with a human driver alone. It's not "close to as safe" as a human alone, it's far safer. And it won't go to unsupervised FSD until the data proves there will be fewer deaths than with humans alone.

Regulators look at the big picture, at whether something is causing more accidents. Tesla uses disengagements to find the needles in the haystack (the scenarios in which more training data is needed). Sure, a fatality would be included in that, but Tesla doesn't need fatalities to develop the system. Because a fatality is a driver who wasn't paying attention. If they were payng attention, it would only have been a disengagement.

Why is this so difficult for you to understand? Fatalities are not a necessary component of the development of FSD, except to the extent that they are going to happen with or without FSD Supervised, just at a lower rate with FSD Supervised.

It's another false narrative that fatalities are necessary to perfect FSD.

 

[Darthamerica]

It looks like I'm still talking past you.

The statisitics show there are already fewer fatalities with Supervised FSD than with a human driver alone. It's not "close to as safe" as a human alone, it's far safer. And it won't go to unsupervised FSD until the data proves there will be fewer deaths than with humans alone.

Regulators look at the big picture, at whether something is causing more accidents. Tesla uses disengagements to find the needles in the haystack (the scenarios in which more training data is needed). Sure, a fatality would be included in that, but Tesla doesn't need fatalities to develop the system. Because a fatality is a driver who wasn't paying attention. If they were payng attention, it would only have been a disengagement.

Why is this so difficult for you to understand? Fatalities are not a necessary component of the development of FSD, except to the extent that they are going to happen with or without FSD Supervised, just at a lower rate with FSD Supervised.

It's another false narrative that fatalities are necessary to perfect FSD.

Ok then I 100% respectfully disagree with you that it’s “far safer.” That’s a marketing claim with cherry picked statistics. I use it everyday and on every single drive I’ve experienced something from FSD(S) that I’d consider unsafe. Not some drives, every drive! But I agree that it’s safe enough for an attentive driver to use on public roads.

For reference, I’ve had two accidents in Tesla vehicles. One using AP, and once when I was driving manually. FSD(S) is way better than AP, but like AP it still has a long long way to go before really matches the marketing claims.

 

[HaulingAss]

Ok then I 100% respectfully disagree with you that it’s “far safer.” That’s a marketing claim with cherry picked statistics. I use it everyday and on every single drive I’ve experienced something from FSD(S) that I’d consider unsafe. Not some drives, every drive! But I agree that it’s safe enough for an attentive driver to use on public roads.

For reference, I’ve had two accidents in Tesla vehicles. One using AP, and once when I was driving manually. FSD(S) is way better than AP, but like AP it still has a long long way to go before really matches the marketing claims.

OK, I'm done with talking past you.

You believe what you want to believe.

 

[Darthamerica]

OK, I'm done with talking past you.

You believe what you want to believe.

It’s not what I believe, I know that FSD(S) requires human supervision for a reason. And for that reason, you can’t make a direct comparison and assert that it’s safer. Well you can, but I assure you regulators won’t. FSD(S) is indeed “safer” under ideal circumstances such as on the HWY/FWY in flowing traffic. But on city streets or edge cases, It’s not safer and often requires disengagements. Safer and safe enough are two different concepts that people using this system need to understand Better.

 

[HaulingAss]

It’s not what I believe, I know that FSD(S) requires human supervision for a reason. And for that reason, you can’t make a direct comparison and assert that it’s safer. Well you can, but I assure you regulators won’t. FSD(S) is indeed “safer” under ideal circumstances such as on the HWY/FWY in flowing traffic. But on city streets or edge cases, It’s not safer and often requires disengagements. Safer and safe enough are two different concepts that people using this system need to understand Better.

Sigh, it's supervised because disengagements are currently a part of FSD Supervised. That's why FSD Supervised is safer than a human alone. And it won't go to unsupervised until that is safer as well.

Your argument is circular.

 

[Darthamerica]

Sigh, it's supervised because disengagements are currently a part of FSD Supervised. That's why FSD Supervised is safer than a human alone. And it won't go to unsupervised until that is safer as well.

Your argument is circular.

Disengagements have nothing to do with it being supervised or not. There will be disengagements even when unsupervised. So my argument is not circular. My argument is that it’s not accurate to make a broad statement asserting that FSD Supervised is safer than a human alone. And if you doubt that, I have a few streets here in SoCal you can find out on. FSD Supervised is likely safer than a human alone on average in a few scenarios. Scenarios like a long cross country HWY drive. But in urban environments and vs edge cases that is not the case.

 

[HaulingAss]

Disengagements have nothing to do with it being supervised or not. There will be disengagements even when unsupervised. So my argument is not circular. My argument is that it’s not accurate to make a broad statement asserting that FSD Supervised is safer than a human alone. And if you doubt that, I have a few streets here in SoCal you can find out on. FSD Supervised is likely safer than a human alone on average in a few scenarios. Scenarios like a long cross country HWY drive. But in urban environments and vs edge cases that is not the case.

FSD Supervised is safer than a human in only a few scenarios?

No, it's safer over 1.3 billion miles of actual use, all-inclusive.

 

[Darthamerica]

FSD Supervised is safer than a human in only a few scenarios?

No, it's safer over 1.3 billion miles of actual use, all-inclusive.

This is why the term “safer” used for FSD(S) is misleading. You can’t use all-inclusive miles because the data used to calculate FSD(S) usage is heavily skewed towards highway mileage. It’s also biased towards certain types of drivers. In those cases, yes, it’s safer. But on city streets, for the general public(ie people who can’t afford $40,000 cars), it’s only just beginning to be used.

 

[Pops]

FSD Supervised is safer than a human in only a few scenarios?

No, it's safer over 1.3 billion miles of actual use, all-inclusive.

https://www.tesla.com/VehicleSafetyReport

I personal do believe FSD is safer, however there is a problem with Tesla self reported numbers, which makes it hard to be certain.

1. They are naturally going to be biased.
2. They only report on Autopilot(not FSD), which is mostly highway driving. City streets are more dangerous than highways, that gives Tesla's stats an immediate edge. Because they compare to total crashes per mile for all US roads.
3. Tesla numbers are more exact, because they are fleet reported. NHTSA are estimates and might be incorrect.
4. Tesla's data directly contradicts insurance industry reports that show "Teslas crash more than gas-powered cars". Even with the huge reported advantage of Autopilot, they are still in more crashes over all per mile? Something doesnt seem to match up with that.

 

[HaulingAss]

https://www.tesla.com/VehicleSafetyReport

I personal do believe FSD is safer, however there is a problem with Tesla self reported numbers, which makes it hard to be certain.

1. They are naturally going to be biased.
2. They only report on Autopilot(not FSD), which is mostly highway driving. City streets are more dangerous than highways, that gives Tesla's stats an immediate edge. Because they compare to total crashes per mile for all US roads.
3. Tesla numbers are more exact, because they are fleet reported. NHTSA are estimates and might be incorrect.
4. Tesla's data directly contradicts insurance industry reports that show "Teslas crash more than gas-powered cars". Even with the huge reported advantage of Autopilot, they are still in more crashes over all per mile? Something doesnt seem to match up with that.

The actual crash statistics Tesla publishes show a safety lead so great that I don't think even the factors you mention are significant enough to over come. Because it's not even close.

Do you have a link to those statistics from the "insurance industry"? Because I believe they were only comparing Tesla's to luxury cars. That's how you write a hit piece, not compare actual accident rates. I think the Model S and X could loosely be considered "luxury" cars but that only makes up less than 20% of Tesla vehicles on the road and they don't even have the same demographics as luxury cars. In other words, it's false that Tesla's have a higher accident rate than all cars on the road if they had to resort to comparing them only to luxury cars. It's a laughable "study", more like a hit piece.

 

[eswimm]

It’s not what I believe, I know that FSD(S) requires human supervision for a reason. And for that reason, you can’t make a direct comparison and assert that it’s safer. Well you can, but I assure you regulators won’t. FSD(S) is indeed “safer” under ideal circumstances such as on the HWY/FWY in flowing traffic. But on city streets or edge cases, It’s not safer and often requires disengagements. Safer and safe enough are two different concepts that people using this system need to understand Better.

FSD requires human supervision, the combination of FSD and the human supervisor is what makes it safer. Alone, it may be more or less safe given conditions, but with an attentive human, one who is evaluating potentially unsafe situation before the vehicle arrives and either preemptively disengaging or just being ready to correct as needed, it is safer. The vehicle can detect what the driver may not and the driver detects and corrects actions by the vehicle. An incident occurs only at the intersection of driver and vehicle failure and by that very nature occurs less than either alone.

 

[Outdoors]

Be bold, be cautious, enjoy the ride.

Thanks Dad. Seems like someone wanted to talk about something, then they continued to talk. I just can't understand why.

 

[Darthamerica]

FSD requires human supervision, the combination of FSD and the human supervisor is what makes it safer. Alone, it may be more or less safe given conditions, but with an attentive human, one who is evaluating potentially unsafe situation before the vehicle arrives and either preemptively disengaging or just being ready to correct as needed, it is safer. The vehicle can detect what the driver may not and the driver detects and corrects actions by the vehicle. An incident occurs only at the intersection of driver and vehicle failure and by that very nature occurs less than either alone.

You see what you said? “With an attentive driver.” And even when attentive, there is no way any human driver can evaluate some of the situations and edge cases FSD(S) can put you in fast enough to avoid an accident. That’s why we are seeing instances of it running traffic signals, hitting curbs or even having some accidents. So we shouldn’t make blanket statements or claims that it is “safer.” It is not, yet. Instead it’s reasonably safe enough to release.

You also need to remember that Teslas are still a tiny fraction of the cars on the road. Many Tesla drivers are also more educated, financially better and don’t live in some areas that would be a lot more difficult to handle. That’s why you have to be careful making comparisons because it’s not an apples to apples comparison.

For Tesla fanatics on this forum, none of this means I don’t think Tesla will eventually get there. Instead what this thread is about is that Tesla needs the imperfections and issues discussed here in order to get to the level of capability and safety some people think it has but doesn’t.

Sponsored