how does the new end to end FSD work, need a block diagram from of data flow from the fleet to DoJO to an individual's car

[JB47394]

If inexperienced drivers are not allowed to drive, how will they ever become experienced drivers?

We want safe drivers, certainly, but how much effort should go into training drivers up when we're trying to eliminate the task of driving? I wonder how many lives would be saved simply by implementing the driver attention monitoring system in all new cars.

I also wonder how quickly effective cheats would be marketed in order to defeat the monitor.

 

[ewoodrick]

Wut? I think we were saying pretty much the same thing?

Absolutely not!

"Not just as good as average/mediocre/impaired/careless drivers"

If FSD is better than "average/mediocre/impaired/careless drivers" then it will save lives.

If does not have to be better than "good skilled conscientious human drivers"

 

[Ben W]

Absolutely not!

"Not just as good as average/mediocre/impaired/careless drivers"

If FSD is better than "average/mediocre/impaired/careless drivers" then it will save lives.

If does not have to be better than "good skilled conscientious human drivers"

Wait, are you talking about L4 or L2? For L4, your contention would only be true if the system were only used by average/mediocre/impaired/careless drivers. Since it would also be used by good drivers (who are the majority of drivers), it would need to be much better than average/mediocre/impaired/careless in order to save lives on average. L2 of course is a different story.

 

[ewoodrick]

Ok, then you’re right, I disagree. That contention would only be true if the system were only used by average/mediocre/impaired/careless drivers. Since it will also be used by good drivers (who are the majority of drivers), it will need to be much better than average/mediocre/impaired/careless in order to save lives on average. (To be clear, this is talking about L4. Supervised L2 is a different story.)

You are coming up all sorts of bogus conclusions.

ALL drivers can be distracted.
Pretty much ALL drivers are worse than they think.
Careless is easy.

You are also assuming that if FSD is a worse driver than the elite group that you list, that it will kill people. Definitely not a given.

Just because FSD may balk at something, doesn't mean that someone is going to get hurt. Cruise has had situations in San Francisco where the cars accidentally "gathered." That was a failure, but no one was hurt.
If a FSD vehicle gets stuck at an intersection and it has to be remotely driven through it, that's expensive, but not an accident.

FSD failure is far from being a death.

 

[stopcrazypp]

You are coming up all sorts of bogus conclusions.

ALL drivers can be distracted.
Pretty much ALL drivers are worse than they think.
Careless is easy.

Yep, AAA did a study on self driving car acceptance and 73% of people thought they drive better than the average driver, with it increasing to 80% for men.

Americans are dangerously overconfident in their driving skills —but they're about to get a harsh reality check

Americans think they're great drivers — and that could be holding back self-driving cars.

www.businessinsider.com

 

[stopcrazypp]

Wait, are you talking about L4 or L2? For L4, your contention would only be true if the system were only used by average/mediocre/impaired/careless drivers. Since it would also be used by good drivers (who are the majority of drivers), it would need to be much better than average/mediocre/impaired/careless in order to save lives on average. L2 of course is a different story.

Not following your logic. If FSD drives better than average and are used by both populations (both the average and below and the "good" drivers), it will still increase the average, thus likely saving lives once enough miles have been used on it.

 

[Ben W]

Not following your logic. If FSD drives better than average and are used by both populations (both the average and below and the "good" drivers), it will still increase the average, thus likely saving lives once enough miles have been used on it.

You're right, I should have excluded "average" from my list and just used "mediocre/impaired/careless"; that's what I was thinking when I wrote the post. (Although the average of { average, mediocre, impaired, careless } is still decidedly less then average.)

 

[Ben W]

You are coming up all sorts of bogus conclusions.

ALL drivers can be distracted.
Pretty much ALL drivers are worse than they think.
Careless is easy.

You are also assuming that if FSD is a worse driver than the elite group that you list, that it will kill people. Definitely not a given.

Just because FSD may balk at something, doesn't mean that someone is going to get hurt. Cruise has had situations in San Francisco where the cars accidentally "gathered." That was a failure, but no one was hurt.
If a FSD vehicle gets stuck at an intersection and it has to be remotely driven through it, that's expensive, but not an accident.

FSD failure is far from being a death.

You're right in that I should have excluded "average" from my list and just used "mediocre/impaired/careless".

Cruise has a tiny number of cars deployed, and only at low speeds, in a geofence, pre-mapped to the centimeter, with LiDAR and sensors up the wazoo. This is the polar opposite of what Tesla is planning with Robotaxi/L4. I don't think you can generalize from Cruise's situation to Tesla's in this way.

Granted that most accidents are not fatal accidents or even injury-causing accidents; it's a power law distribution. Also true that some of the most egregious accident types (e.g. 120mph speeding, outright drunkenness, high-speed pursuits, road rage ramming, voluntarily driving off cliffs, etc.) won't have self-driving parallels. But there is no magical way for AV's to be programmed to avoid major accidents while still getting into lots of minor ones. My Tesla on FSD clipped a curb pretty hard at 15mph the other day; if that had happened at 65mph the car might have completely lost control and possibly flipped or crashed. FSD will not magically be able to see curbs (or road hazards) at 65mph that it can't see at 15mph. (Or process them better, if it was a planning failure and not a perception failure.) And FSD has its own failure modes that humans generally don't have; in particular, sudden system crashes and panics. Red Hands of Death will not be an OK thing in a L4 Robotaxi at highway speed.

So, technically, I agree that _most_ FSD failure is far from being a death. But so is most human failure. It's a very high and necessary bar to make FSD safer than humans, one that is still quite far away IMO, particularly with pure vision in a wide ODD.

 

[OxBrew]

If FSD is average, and you make everyone use it, the accident rates will be the same as they are now, but all above average drivers will be at higher risk of crashing than before.

If FSD is 75th percentile, then yeah, the over all crash rate might go down, but the top 25% are at higher risk than before. That still sucks if you are actually a good driver.

I've driven about 725,000 miles (conservative estimate) and never caused a crash over walking speed (parking lot taps, 2 total, would not count the way FSD is being judged, has to be over 12mph?).

Average drivers crash every 200,000 miles. I'm not using it until it's better than me, plus a factor of safety. Not holding my breath.

 

[spacecoin]

It all boils down to that road safety will not improve until the system - by itself - is safer than the average human.

Attention is a human failure mode, that computers don’t have. OTOH computers and ML have plenty of failure modes that humans don’t have.

Driver monitoring can’t detect all sorts of inattentiveness.

Being a test operator in a vehicle is a paid job that requires your full concentration and focus at all times.

 

[ewoodrick]

It all boils down to that road safety will not improve until the system - by itself - is safer than the average human.

Attention is a human failure mode, that computers don’t have. OTOH computers and ML have plenty of failure modes that humans don’t have.

Driver monitoring can’t detect all sorts of inattentiveness.

Being a test operator in a vehicle is a paid job that requires your full concentration and focus at all times.

It boils down differently for me.

If the worst driver selects to use FSD and FSD is better than the worst driver, then road safety improves. At least in the 100 yards around the worst driver.

 

[enemji]

It boils down differently for me.

If the worst driver selects to use FSD and FSD is better than the worst driver, then road safety improves. At least in the 100 yards around the worst driver.

In other words, it might become mandatory for people who have a certain history to be allowed to drive only using FSD

 

[ewoodrick]

In other words, it might become mandatory for people who have a certain history to be allowed to drive only using FSD

I'd word it a little different. If you've lost your license, you can still travel in the car that you aren't driving.

 

[spacecoin]

It boils down differently for me.

If the worst driver selects to use FSD and FSD is better than the worst driver, then road safety improves. At least in the 100 yards around the worst driver.

That’s not how statistics works…

There will as many good drivers over-trusting these types of systems.

 

[ewoodrick]

That’s not how statistics works…

There will as many good drivers over-trusting these types of systems.

It absolutely is how statistics works. If the probability of a driver having an incident is X and the probability of FSD having an incident is lower, then when the driver is removed from the larger pool and FSD is added, the overall probability of the pool will decrease. In might be small, but it is there.

It's basically like saying that if all the DUIs weren't under the influence, the roads would be safer.

 

[enemji]

I'd word it a little different. If you've lost your license, you can still travel in the car that you aren't driving.

That is always the case.

However, just like the safety score on our Tesla, Insurance companies can force people to engage FSD if your safety score is below a certain level.

 

[ewoodrick]

That is always the case.

However, just like the safety score on our Tesla, Insurance companies can force people to engage FSD if your safety score is below a certain level.

I'm pretty sure that they can't force you. But they definitely can give you a discount if you do.

 

[enemji]

I'm pretty sure that they can't force you. But they definitely can give you a discount if you do.

The DMV might get some ideas