If you can't sleep in it, it is not self-driving

[diplomat33]

Alex Roy who works at Argo AI offers a simple litmus test for determining if a car is really self-driving or not:

"Pick a vehicle. Can you get in, pick a destination and safely go to sleep? If yes, it’s self-driving. If no, it’s not"

We need a self-driving litmus test. Not a technical test, or debate over how it works. We need a self-driving litmus test because we need to know whether someone is talking about a real self-driving car, or something simpler, like the cars with driver assistance features people use everyday.

Besides, normal people don’t really care what things are called. We say we do, but we don’t. We care about what things do, and how well they do them.

So here’s my self-driving litmus test. It’s a simple test, as easy as flicking a light switch to check if my power is on.

Can I sleep in it?

That’s it. That’s the test. Pick a vehicle. Can you get in, pick a destination and safely go to sleep? If yes, it’s self-driving. If no, it’s not.

That’s the thing. “Self-driving” has to mean you can sleep in it, because if it doesn’t, what will we call self-driving vehicles that will let you close your eyes?

So remember, my friends. Don’t believe the hype. If you can’t sleep in it, it’s not self-driving.

Source: It's Not a Self-Driving Car Unless You Can Sleep In It

Alex's "sleep" litmus test seems to be a special case of the "mind off" litmus test. Sleep works because you can't be controlling the car in your sleep so the car is definitely doing all the driving. So it does work to prove the car is truly self-driving.

But we could probably think of other "mind off" activities that could also serve as a litmus test:
- Can I ride in the back seat with no driver in the front seat?
- Can I read a book while the car drives?
- Can I watch a movie while the car drives?

I think we might be able to generalize Alex's litmus test: "if you can't safely take your mind off of driving, it is not self-driving".

 

[linux-works]

my guess, fwiw, is that sleeping in the car can only happen in a 100% closed course, with all other players on that course also being self-driving and being aware of each other.

but while we have this dangerous mix of human and machine, and roads completely WRONG for machine sensors and use, its not going to be sleepable. not with this mix. there aren't enough nines. its that simple, sadly.

I want next generation roads where its a closed off system and we can truly trust it. I think we'll have that, too, if we dont wipe ourselves out (long term; 50 years or less time, but definitely not in the next 10 or 20).

 

[Terminator857]

I can sleep in an uber.

 

[TresLA]

That general litmus test is a good final goal, but it also highlights a main reason why this is so foreign a concept for us now: what’s in between cruise control and fully self driving? General public doesn’t have any steps in between, just features (e.g. - self park, lane keep assist), but this is how I find it so difficult to describe Tesla’s advancements. I see improvements over the years, but for people I’m trying to describe the improvements to, it’s neither plain cruise control nor is it self driving, so it’s just confusing.

 

[diplomat33]

I can sleep in an uber.

Yes, but there is a human in the driver seat, driving the car. So it is clearly not self-driving. The "sleep" litmus test only applies to automated driving systems to determine if the systems are actually self-driving or just an advanced driver assist. It is basically a test to distinguish between "true FSD" and "driver assist pretending to be FSD".

 

[Daniel in SD]

I still think highway L3 systems are viable and are self-driving so I don't agree that you have to be able to sleep.

 

[diplomat33]

I still think highway L3 systems are viable and are self-driving so I don't agree that you have to be able to sleep.

Fair enough. But L3 would only be self-driving in certain circumstances since they would not be self-driving when the driver has to take over. So L3 would probably be partial self-driving, right?

 

[diplomat33]

That general litmus test is a good final goal, but it also highlights a main reason why this is so foreign a concept for us now: what’s in between cruise control and fully self driving? General public doesn’t have any steps in between, just features (e.g. - self park, lane keep assist), but this is how I find it so difficult to describe Tesla’s advancements. I see improvements over the years, but for people I’m trying to describe the improvements to, it’s neither plain cruise control nor is it self driving, so it’s just confusing.

That's where the SAE levels can help.

Cruise control is just L1.

Tesla's Autopilot and so-called FSD are L2.

If Tesla's FSD gets to the point where Tesla removes driver supervision, it would become L3 or higher. Most people define "self-driving" as L3 or higher.

 

[PDX-Y]

This is just a semantic argument, and I don't see the point. I mean, put a normal person in the passenger seat, engage AP (not even FSD, no need to even bother to get technical here), enter a turn or three, and ask them what it is doing. They will say "The car is driving itself." Straightforward language expressed in a standard way.

Yet you want a straightforward reorganization of the same word roots to be false: "The car is NOT self-driving."

That seems to me to obscure more than it helps. It sounds good if you want the answer to be "FSD is bad", I guess? It doesn't help explain anything to anyone who doesn't already understand how vehicle autonomy works. Nor does it address the fact that there is value in all those intermediate ("bad") states that require some level of supervision.

 

[Daniel in SD]

This is just a semantic argument, and I don't see the point. I mean, put a normal person in the passenger seat, engage AP (not even FSD, no need to even bother to get technical here), enter a turn or three, and ask them what it is doing. They will say "The car is driving itself." Straightforward language expressed in a standard way.

Yet you want a straightforward reorganization of the same word roots to be false: "The car is NOT self-driving."

That seems to me to obscure more than it helps. It sounds good if you want the answer to be "FSD is bad", I guess? It doesn't help explain anything to anyone who doesn't already understand how vehicle autonomy works. Nor does it address the fact that there is value in all those intermediate ("bad") states that require some level of supervision.

Then what do you call a vehicle that can be legally operated without supervision?
I would actually categorize FSD beta as a prototype L5 self-driving car. Tesla can't really do that because there are laws about testing prototype self-driving vehicles on public roads.

Fair enough. But L3 would only be self-driving in certain circumstances since they would not be self-driving when the driver has to take over. So L3 would probably be partial self-driving, right?

It is self-driving when in its ODD, that's the way a look at it.

 

[FloridaJohn]

I would actually categorize FSD beta as a prototype L5 self-driving car. Tesla can't really do that because there are laws about testing prototype self-driving vehicles on public roads.

Of course they can do that. Other companies are categorizing their test cars as L5 already. It’s just that once you do that in California, you then have to report accidents and disengagements. And Tesla does not want to do that.

 

[PDX-Y]

Then what do you call a vehicle that can be legally operated without supervision?

You can call it anything you want! SAE even went and gave you numbers for this stuff. In that jargon, the word you want is "four". How about we use that if we want to get technical?

But arguing about what to call it says nothing about what it does, and IMHO arguments like this are really just fronts for side-channel fights about whether or not "FSD is bad". Changing the language to make you right is a logical fallacy.

I'm just saying that making pure semantic arguments about stuff like this, especially when it flies in the face of millenia-old English grammar (c.f. this makes "driving itself" and "it is self-driving" mean different things!).

 

[diplomat33]

This is just a semantic argument, and I don't see the point. I mean, put a normal person in the passenger seat, engage AP (not even FSD, no need to even bother to get technical here), enter a turn or three, and ask them what it is doing. They will say "The car is driving itself." Straightforward language expressed in a standard way.

The problem is that there is a big difference between "the car is steering and braking but it can't react safely to certain situations so the driver needs to pay attention" and "the car is steering and braking and it can react safely to all situations so the driver does not need to pay attention". In both instances, to the untrained eye, the car appears to be driving itself since it is doing the steering and braking. But in the first case, you better be paying attention or risk an accident whereas in the second case, you don't need to pay attention. That's a pretty big difference. And if you confuse the two, you risk a serious accident if you assume the car can handle something when it can't. That is why clearly understanding the difference between a "driver assist that looks like self-driving" and "true self-driving" is so important.

 

[Daniel in SD]

You can call it anything you want! SAE even went and gave you numbers for this stuff. In that jargon, the word you want is "four". How about we use that if we want to get technical?

But arguing about what to call it says nothing about what it does, and IMHO arguments like this are really just fronts for side-channel fights about whether or not "FSD is bad". Changing the language to make you right is a logical fallacy.

I'm just saying that making pure semantic arguments about stuff like this, especially when it flies in the face of millenia-old English grammar (c.f. this makes "driving itself" and "it is self-driving" mean different things!).

I told you what I would call it. What would you call it?

 

[NHK X]

Alex Roy who works at Argo AI offers a simple litmus test for determining if a car is really self-driving or not:

"Pick a vehicle. Can you get in, pick a destination and safely go to sleep? If yes, it’s self-driving. If no, it’s not"





Source: It's Not a Self-Driving Car Unless You Can Sleep In It

Alex's "sleep" litmus test seems to be a special case of the "mind off" litmus test. Sleep works because you can't be controlling the car in your sleep so the car is definitely doing all the driving. So it does work to prove the car is truly self-driving.

But we could probably think of other "mind off" activities that could also serve as a litmus test:
- Can I ride in the back seat with no driver in the front seat?
- Can I read a book while the car drives?
- Can I watch a movie while the car drives?

I think we might be able to generalize Alex's litmus test: "if you can't safely take your mind off of driving, it is not self-driving".

You can really sleep in any car you desire.. the real question is will you stay sleeping forever or not. : )

 

[diplomat33]

You can really sleep in any car you desire.. the real question is will you stay sleeping forever or not. : )

Yeah. That's where the "safe" part comes in.

"Pick a vehicle. Can you get in, pick a destination and safely go to sleep? If yes, it’s self-driving. If no, it’s not"

 

[NHK X]

Yeah. That's where the "safe" part comes in.

"Pick a vehicle. Can you get in, pick a destination and safely go to sleep? If yes, it’s self-driving. If no, it’s not"

was a joke... man.

 

[PDX-Y]

I told you what I would call it. What would you call it?

Honestly I thought I was clear: I would not pick a fight on the semantic meaning of "self-driving" as this thread does, as it's clearly designed as spin and not technical discussion. If you want to say that Tesla offers "supervised autonomy" as a way to make clear the need for a driver, then I'd get behind that. If you want to say that the car is "full self-driving", I'm fine with that too, because to a normal English speaker it connotes that the car is driving itself (which it is).

Again, the goal of the spin hereis transparently not to explain how Tesla's autonomy solution works. The goal is to try to delegitimize an actually very impressive and objectively valuable product (I mean: people pay five figure sums for this thing!) by picking not on its capabilities, but on its... name. Don't buy into spin like that.

 

[Daniel in SD]

Honestly I thought I was clear: I would not pick a fight on the semantic meaning of "self-driving" as this thread does, as it's clearly designed as spin and not technical discussion. If you want to say that Tesla offers "supervised autonomy" as a way to make clear the need for a driver, then I'd get behind that. If you want to say that the car is "full self-driving", I'm fine with that too, because to a normal English speaker it connotes that the car is driving itself (which it is).

Again, the goal of the spin hereis transparently not to explain how Tesla's autonomy solution works. The goal is to try to delegitimize an actually very impressive and objectively valuable product (I mean: people pay five figure sums for this thing!) by picking not on its capabilities, but on its... name. Don't buy into spin like that.

The question was what would you call a car that does require a driver?

I would be interested in survey data because I always figured most people would say a car that drives itself should not require a driver. I figured it was only some Tesla owners who had a different definition of self-driving. The term "self-driving" has existed for decades and it's always meant driverless as far as I know. I often use the term "robotaxi" around here to avoid confusion though.
Like I said I actually think FSD beta has accurate nomenclature as long as it's emphasized that it's prototype self-driving system.

 

[PDX-Y]

The term "self-driving" has existed for decades and it's always meant driverless as far as I know.

I don't think that's true at all, do you have something specific in mind? As far as I know, this nonsense semantic argument about the "real" meaning of "self-driving" never existed at all until people wanted to yell about Tesla.

Note that in aviation, cockpit automation systems have decades and decades of history and no one every wanted to make a stink about needing a pilot, nor come up with a new word (beyond "missile" I guess) to describe vehicles without one.