Is it possible that AVs will never be as safe as human drivers with ADAS?

[Daniel in SD]

I was just thinking that all the advancements in the AV space are being applied to the new generation of ADAS (advanced driver assistance systems) and will make cars much safer. This will raise the bar on how good AVs must be to be safer than humans. It seems possible that collision avoidance systems, speed limiters, and driver vigilance monitoring could make human driven cars impossible to beat in the foreseeable future.
For example, if you develop a pedestrian detection and collision avoidance system for an AV that is 99.9% effective you've built a killing machine but if you put that same system in a regular car you've reduced pedestrian collisions by a factor of a thousand.
At some point it seems like you might have to decide that AVs are safe enough even if they are less safe than machine assisted humans.

 

[diplomat33]

Considering how many car fatalities we still have, even with better ADAS, I think there is definitely room for AVs to be safer.

 

[Frizull]

Any sufficiently advanced ADAS system is indistinguishable from an AV.

 

[Lasairfion]

Any sufficiently advanced ADAS system is indistinguishable from an AV.

By simple definition that cannot be true. An autonomous vehicle does not assist a driver, it replaces them.

Also, the real question is 'how much safer does the autonomous vehicle need to be compared to the average human driver for regulatory approval'.

Because at the end of the day, your average driver is pretty awful. As someone who teaches advanced driving I see some shocking cases come to us, and that's people who actually want to improve. The average driver on the street... it's a wonder any of us survive at all. So really the autonomy doesn't have to be all that good to be better than the average driver.

It also appears that American driving test standards are particularly low compared to the UK. So I can only assume that your driving would be worse, although your roads seem much easier than ours to navigate.

 

[Daniel in SD]

Any sufficiently advanced ADAS system is indistinguishable from an AV.

I’m talking about systems that run in the background to prevent accidents. Like automatic emergency braking but more advanced.
Sort of like the way a human assisted by a computer is better at chess than either humans or computers. There are things the computers are good at and things that humans are good at and they don’t completely overlap.

 

[Daniel in SD]

By simple definition that cannot be true. An autonomous vehicle does not assist a driver, it replaces them.

Also, the real question is 'how much safer does the autonomous vehicle need to be compared to the average human driver for regulatory approval'.

Because at the end of the day, your average driver is pretty awful. As someone who teaches advanced driving I see some shocking cases come to us, and that's people who actually want to improve. The average driver on the street... it's a wonder any of us survive at all. So really the autonomy doesn't have to be all that good to be better than the average driver.

It also appears that American driving test standards are particularly low compared to the UK. So I can only assume that your driving would be worse, although your roads seem much easier than ours to navigate.

And yet even though humans are horrible drivers they only get into accidents every 150,000 miles on average. Imagine how good they could be with a computer in the background to keep them from making stupid mistakes.

 

[Frizull]

I’m talking about systems that run in the background to prevent accidents. Like automatic emergency braking but more advanced.

Embedded within are two assumptions that might contribute to the proposition being false. First is the assumption that drivers with ADAS equipped vehicles will keep their vigilance at a constant level with higher levels of assistance. Second is the assumption that the human driver would not have a negative effect on safety once ADAS systems reach AV-level competence.

By simple definition that cannot be true. An autonomous vehicle does not assist a driver, it replaces them.

The quick take is meant to be read as a reference to Arthur C. Clarke's famous "Any sufficiently advanced technology is indistinguishable from magic."
That being said, the Autopilot team is on a path to make very competent ADAS that will organically morph into AV. The distinction might be useful for other players that have already removed the steering wheel or have no safety driver, but it's not applicable here. Separating ADAS from AV is a false dichotomy in Tesla's case.

 

[Daniel in SD]

Embedded within are two assumptions that might contribute to the proposition being false. First is the assumption that drivers with ADAS equipped vehicles will keep their vigilance at a constant level with higher levels of assistance. Second is the assumption that the human driver would not have a negative effect on safety once ADAS systems reach AV-level competence.

I agree that this could be an issue but I think most people would react to the car automatically doing a violent braking or avoidance maneuver as evidence that they should pay more attention not less. I think that if AVs reach current humans levels of safety then the accidents they have will still be easily avoidable by a human. This seems obvious since almost all accidents are caused by human error and not technical skill.

 

[AlanSubie4Life]

First is the assumption that drivers with ADAS equipped vehicles will keep their vigilance at a constant level with higher levels of assistance

I think the hypothetical here is saying that this would very much be a background system which doesn't actively assist, like AP does. It just jumps in last minute and fairly reliably saves lives (while possibly not reliably preventing vehicle damage)

Autopilot team is on a path to make very competent ADAS that will organically morph into AV.

I agree that it could morph into AV, but for the AV side of things, I think the point is that the bar for the safety requirement to allow widespread AV use could be a moving target, due to mandated ADAS systems which dramatically improve all vehicles' safety overall. Tesla doesn't control the bar that regulators think is good enough - and what if by the time they are ready to convert their ADAS to AV, all new vehicles on the road driven by human drivers are dramatically safer than they are now, so the AV is actually less safe than those vehicles (though much safer compared to today's standards)?

 

[ewoodrick]

And yet even though humans are horrible drivers they only get into accidents every 150,000 miles on average. Imagine how good they could be with a computer in the background to keep them from making stupid mistakes.

Absolutely correct. But take it a step further, take the human out of the scenario. Accidents go down, traffic congestion decreases, traffic laws get followed and people arrive even earlier.

A simulation last year from MIT showed computer AV handling of a 4-way intersection. Cars are able to pass through at full speed with very little gap between cars. Some humans can not make a turn within 10 seconds of the adjacent vehicle.

Once humans stop driving, AVs can be significantly dumber.

 

[AlanSubie4Life]

Once humans stop driving, AVs can be significantly dumber.

I understand that networked cars, etc. may make some problems easier to solve. But there are a lot of situations.
So, I’m not sure your statement is true. It does depend on the ultimate level of safety desired of course, but I want an AV that is extremely good at dealing with some pretty odd situations which can occur in every day driving which have little to do with other vehicles on the road. Humans (even bad ones) are remarkable in terms of what they can figure out about how to deal with unusual situations. They just aren’t very good at the more routine attentiveness issues.

 

[theboom1]

I think the eventual end game for ADAS is simply the full self driving software still running in the background even when your driving and when it detects significant mistake, it will subtly but instantly turn itself on, fix the problem, and then fade back off. This way it allows people to keep driving if they want/need to nut it will make it just as safe as full AV so best of both worlds. They could also have cameras inside the car that track your eyes that automatically turn on AV when you are not paying attention.