By my reading of the article, stale2002 is correct. Pratt is clearly talking about the difference between Level 4 and Level 5 in his first two answers. Level 4 doesn't require driver attention, it just requires the driver to be (instantly?) available if the system has a problem, and this change-over is the situation that Pratt worries about in his fourth answer.
(NB: When he says "The article was talking about level 5", it's clear from context that stale2002 means that the article was talking about level 5 in terms of technological difficulty and timelines.)
I don't think so, mostly because I don't think a qualifying system would actually require an instant switch over. I can't find anywhere that it the timescale is defined; if I had to guess, I'd say it's defined in terms of whatever the minimum safe timescale is (empirically), i.e., Level 4 is whatever it needs to be to allow a human to read a book safely, but not necessarily sleep.
For instance, a Level 4 car probably can't deal with unexpected road construction, and might simply bring the car to a halt, smoothly but quickly. It would be dangerous if the driver was asleep and the car stopped dead on the highway for the 30 seconds it takes someone to wakeup and get oriented, but it could be pretty safe if the driver just needs to look up from their movie and grab the wheel.
> Pratt is clearly talking about the difference between Level 4 and Level 5 in his first two answers
No, Pratt is saying that manufacturers are hyping level 4 as being around the corner, but it is not. Pratt does not think the claims for level 4 are sufficiently backed up:
"...That’s Level 4. And I wouldn’t even stop there: I would ask, “Is that at all times of the day, is it in all weather, is it in all traffic?”
> Level 4 doesn't require driver attention, it just requires the driver to be (instantly?) available if the system has a problem
How exactly would that work? You don't have to be attentive but you must be available to take over?
At any rate, this is not the definition provided by the Society of Automotive Engineers [1], which is the definition NHTSA has adopted for levels of autonomous vehicles.
SAE says for level 4:
"the driving mode-specific performance by an automated driving system of all aspects of the dynamic driving task, even if a human driver does not respond appropriately to a request to intervene"
In other words, the system could request human intervention, but it will still do something even if the human does not respond.
Wikipedia appears to have either summarized this in different words or has a different version.
> this change-over is the situation that Pratt worries about in his fourth answer.
I wholeheartedly agree with Pratt that "In some ways, the worst case is a car that will need driver intervention once every 200,000 miles..."
> "...That’s Level 4. And I wouldn’t even stop there: I would ask, “Is that at all times of the day, is it in all weather, is it in all traffic?”
But that doesn't actually contradict stale2002 or diminish the enormous potential economic importance by more than (say) a factor of 2. Level 4 systems that are available to use 60% of the time are still revolutionary, and compatible with both stale2002's and Pratt's comments.
> How exactly would that work?
See my comment to function_seven for my speculation. Although I used the Wikipedia definition, I believe everything I said is compatible with the definition from SAE you have quoted. In particular "the system could request human intervention, but it will still do something even if the human does not respond" does not conflict with "it just requires the driver to be available if the system has a problem" because the "something" the system may do is bring the car to a rapid (or even emergency) stop. This can become dangerous if the human can't resume control within a few seconds, but is an easily acceptable risk to occur once every 100k miles as a price of Level 4 autonomy.
stale2002 said the article is talking about level 5, and that level 4 is around the corner.
Yet the article is talking about level 5, 4, and below. It is saying some car companies are overhyping their capability to reach levels 4 and 5.
Pratt feels even level 4 is not right around the corner.
> Level 4 systems that are available to use 60% of the time are still revolutionary
To reach a certain level, you must operate under the definition of that level 100% of the time. You can't be level 4 60% of the time. That's level 3.
I agree it would be revolutionary to be at level 3. Right now no car is there. They're all at level 2, which require human monitoring of the driving situation.
> To reach a certain level, you must operate under the definition of that level 100% of the time. You can't be level 4 60% of the time. That's level 3
No. I'm looking at the SAE definitions, where it says
> A particular vehicle may have multiple driving
automation features such that it could operate at different levels depending upon the feature(s) that are engaged.
clearly indicating that the levels describe modes of operation, not immutable car classifications.
Regardless, this has just become a semantical dispute. stale2002's comment makes the most sense under the interpretation "cars that that can drive at Level 4 60% of the time are just around the corner", and that comment is a valuable counterpoint to the reasonable interpretation of Pratt's interview.
> the levels describe modes of operation, not immutable car classifications
Look at the column where human attention is required ("Monitoring of Driving Environment"). Every car in existence requires human monitoring in every driving mode.
The moment attention is not required, the car company becomes liable. Volvo will release a level 4 car this year where they assume liability, however, that is only for a hundred people in Sweden.
"Around the corner" means widely available, and that's just not the case.
Right now Tesla does not reliably save the data from its accidents. How could a car company assume liability when they can't even save the data?
> Level 4 doesn't require driver attention, it just requires the driver to be (instantly?) available if the system has a problem
Based on my best understanding of the article and the good description on these levels in the CES keynote linked in the article...
This is Level 2, when the driver must be available instantly. Level 3 is when there driver must be available but there's plenty of time (the article mentioned 15 seconds) to swap.
Level 4 should not need any human intervention when engaged. It can only be engaged when it's safe to do (e.g. in a city in good weather conditions) but once it's enabled, it fulfills the task given to it without ever needing to fall back to human hands. You should be able to sit on the back seat drinking beer. Level 4 should be able to safely stop the car and wait for assistance if things go bad without driver intervention.
Level 2 is where the cars out there are now. Level 5 is a long way away.
So basically by your definition, level 5 will never ever happen? Seems a bit useless then TBH.
Also, as pointed out in another comment, Level 4 does require driver changeover, it's just that the car is supposed to safely handle the event of driver not responding to that request (e.g. by stopping in the middle of the road).
> So basically by your definition, level 5 will never ever happen? Seems a bit useless then TBH.
Well, even trains on tracks have humans ready to pull the break if something is wrong up ahead.
Would you ride a train that had an accident caused by something for which its driverless system was not prepared, but that a human conductor could have avoided?
I think it's possible level 5 will happen. There will still be accidents on roads where only driverless vehicles exist. Some people will choose to ride in such vehicles, and some won't.
No, what I'm trying to get at is GP post changed the definition of level 5 from "some cars don't have steering wheels" to "all cars don't have steering wheels". A non-self-driving car will (by definition) always be cheaper than a self-driving car, and some people just like driving or distrust machines. Together with the fact that (especially in the United States of Freedom) governments will never ban cars with steering wheels, that definition of level 5 will never happen.
> A non-self-driving car will (by definition) always be cheaper than a self-driving car
I don't think that's true at all. A self-driving car can save money on not having to have a steering wheel, pedals, hand-brake etc. Possibly even won't require airbags or seat belts if safety is improved massively
non-self driving cars could become expensive luxury cars
Surely you jest; that the cost of the wide array of self driving sensors, the data input subscription and the hardware and software to run the self-driving system is cheaper than a few mass-produced bits of plastic and metal?
Maybe in the future. Especially 'hardware and software to run the self-driving system' will cost a few dollars (small computer, zero amortized cost of software)
(NB: When he says "The article was talking about level 5", it's clear from context that stale2002 means that the article was talking about level 5 in terms of technological difficulty and timelines.)