Coast to coast drive happening this year for all FSD Teslas!

[Electroman]

99% to me means - it can drive autonomously 99% of the roads and conditions, and remaining requires human attention or manual driving.

Now you could end up driving on a route that is 100% within that 99% and that drive is fully autonomous, or vice-versus.

With that metric Waymo and Uber are 1% autonomous

 

[mongo]

Someone doesn't understand annualized failure rates.

Someone doesn't understand numbers with no units. Nor calling out a specific driver 'you' versus a set amount of time/ distance/ events.

99% of the time means you crash 3 times a year.

Regardless of distance or time?

 

[banned-66611]

Remember that it is Musk who is claiming full level 5 "functionally complete" this year. You can argue over percentages but the bar he has set is incredibly high, and he claimed a lower one back in 2017 and failed to reach that too.

 

[mongo]

Remember that it is Musk who is claiming full level 5 "functionally complete" this year. You can argue over percentages but the bar he has set is incredibly high, and he claimed a lower one back in 2017 and failed to reach that too.

He did not say functionally complete(pedantic level high since you used double quotes) even if that is what certain Twitter accounts claim...
.

“I think we will be feature complete — full self-driving — this year,”

also:

“My guess as to when we would think it is safe for somebody to essentially fall asleep and wake up at their destination? Probably towards the end of next year,” he said.

Elon Musk "certain" full self-driving tech will be ready next year

 

[banned-66611]

Oh right, I forgot that when he says "full self driving" he means level 2 driver aids where you must be fully alert at all times. And then from there to level 5 in a year.

We know what he has been smoking.

 

[mongo]

Oh right, I forgot that when he says "full self driving" he means level 2 driver aids where you must be fully alert at all times. And then from there to level 5 in a year.

We know what he has been smoking.

No, you 'forgot'; that it is a good idea to validate software in the real world, across many use cases, for million of miles in a variety of weather, before saying it is hand offs, no attention needed FSD.

 

[banned-66611]

No, you 'forgot'; that it is a good idea to validate software in the real world, across many use cases, for million of miles in a variety of weather, before saying it is hand offs, no attention needed FSD.

When you say "validate", you mean use Tesla customers as crash test dummies, right?

Other car manufacturers seem able to develop their own full self driving features without forcing the public to test them. For example Nissan is releasing navigate on autopilot later this year and didn't get one single customer to beta test it.

 

[willow_hiller]

Rather than all of us going back and forth conjecturing whether or not this drive will happen, why don't we each define what we think is the definition of a "Coast to coast autonomous drive" and assign some sort of rating system, so we can compare across automakers when these features start rolling out?

To me, I would define a coast-to-coast drive as being completed with the following criteria fulfilled:

1. In my driveway in Maryland, I enter "Fremont, California" as the destination. We're starting with 100 possible points.
   
2. Every time I need to touch the steering wheel, brakes, or accelerator (excluding any nags as required by law), I deduct a point.
3. The vehicle makes it to California with at least 70 points remaining.
   
4. If I'm required to manually drive for more than 100 yards at a time, the test is automatically failed.

Allowing 30 minor required interactions seems like a lot, but in my SR+ there are 26 supercharging stops along the way. I'll be impressed if FSD can pull up to a supercharging spot and put itself in park, but I don't think it's a requirement of the coast-to-coast.

So a 100 point test would be simply "Enter destination, never touch steering wheel, brakes, or accelerator, make it to California."

A barely passing test would be "Enter destination, need to manually pull up to supercharging spots, have 4-5 temporary disengagements throughout the trip, make it to California."

A failing test would be "Enter destination, car gives up on a short stretch of local road and requires manual driving for more than 100 yards."

What would your ideal coast-to-coast tests look like?

 

[mongo]

When you say "validate", you mean use Tesla customers as crash test dummies, right?

Other car manufacturers seem able to develop their own full self driving features without forcing the public to test them. For example Nissan is releasing navigate on autopilot later this year and didn't get one single customer to beta test it.

No I do not mean that.
That would imply telling driver the system was hands off attention off BEFORE validating it.
No one is being asked to be a crash test dummy, they are being TOLD that they are still the driver and still responsible for their safety.

I'll be interested to see how well the Nissan system works.

 

[Snuffysasa]

No, not really. If the safety driver has to take over suddenly for edge cases, then it is not self-driving. If the system can reliably anticipate the need for the driver to do so and alert the driver (or, if the driver is incapacitated, pull over to the side of the road until the driver is sober), then it is self-driving.

Case in point, there's some debate about whether Teslas can safely handle 65 MPH T-junction intersections. But the car knows it is coming up on such an intersection. If it can pull over and force the human driver to drive manually through that intersection, then the car can reasonably be seen as self-driving, because 99.9% of the time, the car won't be dealing with that sort of degenerate case, and the car's navigation system can try as hard as possible to avoid those sorts of intersections, making them even more rare. It is still self-driving so long as the car can always reliably "fail safe" by pulling over prior to situations that it can't handle.

Another case in point, there's no guarantee that the camera coverage can perfectly handle 90 degree nose-in parking. The vast majority of 90-degree parking spaces do not require vehicles to park nose-in, though, and a car can do its best to avoid parking spots that do require it. So a car can reasonably be called self-driving even if it requires a human to drive the first few feet when pulling out of such a spot, so long as it warns the human that he/she will have to unpark manually if it parks in the spot in question and gives the driver the option to search for a different spot without that restriction.

You are saying the same thing as me. I agree with you.

 

[timewasted]

Rather than all of us going back and forth conjecturing whether or not this drive will happen, why don't we each define what we think is the definition of a "Coast to coast autonomous drive" and assign some sort of rating system, so we can compare across automakers when these features start rolling out?

To me, I would define a coast-to-coast drive as being completed with the following criteria fulfilled:

1. In my driveway in Maryland, I enter "Fremont, California" as the destination. We're starting with 100 possible points.
   
2. Every time I need to touch the steering wheel, brakes, or accelerator (excluding any nags as required by law), I deduct a point.
3. The vehicle makes it to California with at least 70 points remaining.
   
4. If I'm required to manually drive for more than 100 yards at a time, the test is automatically failed.

Allowing 30 minor required interactions seems like a lot, but in my SR+ there are 26 supercharging stops along the way. I'll be impressed if FSD can pull up to a supercharging spot and put itself in park, but I don't think it's a requirement of the coast-to-coast.

So a 100 point test would be simply "Enter destination, never touch steering wheel, brakes, or accelerator, make it to California."

A barely passing test would be "Enter destination, need to manually pull up to supercharging spots, have 4-5 temporary disengagements throughout the trip, make it to California."

A failing test would be "Enter destination, car gives up on a short stretch of local road and requires manual driving for more than 100 yards."

What would your ideal coast-to-coast tests look like?

You are in Fremont California and your car is in your driveway in Maryland. You open your app and click the “come pick me up” button. After however many hours you get a notification that your car is 10 minutes away.

I mean, this is what they sold back in 2016, so why not keep things consistent?

 

[willow_hiller]

You are in Fremont California and your car is in your driveway in Maryland. You open your app and click the “come pick me up” button. After however many hours you get a notification that your car is 10 minutes away.

I mean, this is what they sold back in 2016, so why not keep things consistent?

That's a good, easy to measure definition!

If a lot of people have the same expectations, I can see why so many people say "No way is this happening this year." I wasn't even considering buying a Tesla until about March of this year, so the history of promises doesn't affect my expectations as much.

Your definition would be dependent on things other than FSD being completed, right? The Federal government would need to pass legislation allowing it in all 50 states (we're more likely to see things go state-by-state to begin with), and Tesla would need to man Superchargers with attendants until they finish that snake-charger or develop an alternative solution.

It's really hard to be able to say "Yes, this could drive itself cross-country IF the laws allowed and IF it could park and charge itself." because it's never possible to fully test hypothetical scenarios.

 

[Bladerskb]

He did not say functionally complete(pedantic level high since you used double quotes) even if that is what certain Twitter accounts claim...
.

also:


Elon Musk "certain" full self-driving tech will be ready next year

In 2016 he said L5 driverless in 2018. In 2017 he said L5 sleep in your car in 2019. In 2019 Autonomy day he said L5 dont need to pay attention and can sleep by Q2 2020.

 

[timewasted]

That's a good, easy to measure definition!

If a lot of people have the same expectations, I can see why so many people say "No way is this happening this year." I wasn't even considering buying a Tesla until about March of this year, so the history of promises doesn't affect my expectations as much.

Your definition would be dependent on things other than FSD being completed, right? The Federal government would need to pass legislation allowing it in all 50 states (we're more likely to see things go state-by-state to begin with), and Tesla would need to man Superchargers with attendants until they finish that snake-charger or develop an alternative solution.

It's really hard to be able to say "Yes, this could drive itself cross-country IF the laws allowed and IF it could park and charge itself." because it's never possible to fully test hypothetical scenarios.

So as a less flippant/more thought out answer, about the only thing I'd accept for an actual full self driving demo would be something along the lines of a typical tourist doing a cross country drive. What I mean by that is that the car will take a couple detours to see local attractions along the drive. For example when the car gets to Vegas, it gets off the highway and drives down the strip, maybe pulling up to a casino. Each area along the trip has some traffic scenario that the locals think of that make them think "a self driving car will NEVER be able to handle this!" So take a short 15-20 minute detour on a couple of stops and prove that it CAN handle it.

I'd also say that a human interacting with the movement of the car at all would be a failure, outside of truly exceptional circumstances. No, I'm not going to try to define "truly exceptional circumstances" beyond saying something like "the lane lines weren't obvious", "the car next to me swerved erratically", or "an accident shut down three lanes of traffic" don't count. Drivers deal with these every day. I'd likely accept "the car hit a nail in the road and got a flat tire" as being an acceptable interaction... maybe. The car should still handle that safely by itself though.

 

[diplomat33]

You are in Fremont California and your car is in your driveway in Maryland. You open your app and click the “come pick me up” button. After however many hours you get a notification that your car is 10 minutes away.

I mean, this is what they sold back in 2016, so why not keep things consistent?

Well, yes, that is what FSD will be when it is completely finished. We should expect to get that and I think Tesla does intend to deliver. But we also have to be realistic. The fact is that Tesla is not going to achieve what you describe this year. Right now, Tesla is working towards that goal but is not there yet. They are getting "feature complete" done this year. So any coast to coast demo this year will be to prove that they have achieved "feature complete" by showing that the car can indeed drive all the way across country with little to no disengagements. But that does not mean that coast to coast demo this year is a failure if it's not a "summon the car across country". If Tesla can achieve a coast to coast demo with very minimal driver interventions (say 0-5), that would still be a huge intermediary success!

So I think @willow_hiller 's metric makes sense. It gives us a specific way to measure how close or how far Tesla is to the "summon the car across country" goal you describe.

Rather than all of us going back and forth conjecturing whether or not this drive will happen, why don't we each define what we think is the definition of a "Coast to coast autonomous drive" and assign some sort of rating system, so we can compare across automakers when these features start rolling out?

To me, I would define a coast-to-coast drive as being completed with the following criteria fulfilled:

1. In my driveway in Maryland, I enter "Fremont, California" as the destination. We're starting with 100 possible points.
   
2. Every time I need to touch the steering wheel, brakes, or accelerator (excluding any nags as required by law), I deduct a point.
3. The vehicle makes it to California with at least 70 points remaining.
   
4. If I'm required to manually drive for more than 100 yards at a time, the test is automatically failed.

Allowing 30 minor required interactions seems like a lot, but in my SR+ there are 26 supercharging stops along the way. I'll be impressed if FSD can pull up to a supercharging spot and put itself in park, but I don't think it's a requirement of the coast-to-coast.

So a 100 point test would be simply "Enter destination, never touch steering wheel, brakes, or accelerator, make it to California."

A barely passing test would be "Enter destination, need to manually pull up to supercharging spots, have 4-5 temporary disengagements throughout the trip, make it to California."

A failing test would be "Enter destination, car gives up on a short stretch of local road and requires manual driving for more than 100 yards."

What would your ideal coast-to-coast tests look like?

I like your point system a lot. I would be ok with excluding supercharger stops from the point system at least initially. Obviously, if the car could auto park at a supercharger stall autonomously that would be amazing but I don't think it is a main priority right now. I think the main purpose of the coast to coast demo this year should be to test the driving parts on roads. Again, we should remember that any coast to coast demo this year will only be to prove "feature complete" (ie the car has what it needs to drive across country with very minimal disengagements. Features like being able to autonomously park in a supercharger stall will come later.

I guess for me, a successful coast to coast demo this year is the car driving coast to coast, including every type of road, with zero or very minimal interventions (0-5) that were not serious, not counting supercharger stops. For example, interventions because the car was going to blow through a red light would be a fail for me because that is a traffic violation and a serious safety issue. But 1-5 imperfections on a 2000 mile drive like the car was a bit too slow starting after a red turned green or hesitated a few seconds to turn at an intersection but still did it, would be acceptable for me.

Now, next year or the year after that, my standards for a coast to coast demo will be higher because by then, I will expect FSD to be better than "feature complete".

 

[diplomat33]

Folks. I do think we need to have realistic expectations in terms of this year's coast to coast demo if it happens: remember that this year, Tesla is aiming to finish "feature complete". So any coast to coast demo this year will simply be to prove that they have "feature complete" done. It is not going to be perfect L5 autonomy. It is not going to be "summon your car driverless cross country". It may even have a couple driver disengagements. Now, next year or the year after, we should expect FSD to improve. Any coast to coast demo that Tesla does next year or the year after that, better be a lot better than what they do this year.

 

[timewasted]

Folks. I do think we need to have realistic expectations in terms of this year's coast to coast demo if it happens: remember that this year, Tesla is aiming to finish "feature complete". So any coast to coast demo this year will simply be to prove that they have "feature complete" done. It is not going to be perfect L5 autonomy. It is not going to be "summon your car driverless cross country". It may even have a couple driver disengagements. Now, next year or the year after, we should expect FSD to improve. Any coast to coast demo that Tesla does next year or the year after that, better be a lot better than what they do this year.

I'm being totally realistic, because Elon has been 100% consistent about what we're going to get for years now. This is not some arbitrary bar that I came up with for a forum post, but a consistent message from the CEO (and now head of Autopilot team) of Tesla. It has been a core part of Tesla's marketing and sales flow for years now.

 

[Bladerskb]

Well, yes, that is what FSD will be when it is completely finished. We should expect to get that and I think Tesla does intend to deliver. But we also have to be realistic. The fact is that Tesla is not going to achieve what you describe this year. Right now, Tesla is working towards that goal but is not there yet. They are getting "feature complete" done this year. So any coast to coast demo this year will be to prove that they have achieved "feature complete" by showing that the car can indeed drive all the way across country with little to no disengagements. But that does not mean that coast to coast demo this year is a failure if it's not a "summon the car across country". If Tesla can achieve a coast to coast demo with very minimal driver interventions (say 0-5), that would still be a huge intermediary success!

So I think @willow_hiller 's metric makes sense. It gives us a specific way to measure how close or how far Tesla is to the "summon the car across country" goal you describe.

Your cross country example consists of 99.5% freeways though...
0.5% of the miles will actually be city streets.
Not at all convincing nor a representation of what Tesla is selling which is a system that works on all road types and conditions.

Now, next year or the year after that, my standards for a coast to coast demo will be higher because by then, I will expect FSD to be better than "feature complete".

Here's Elon Musk's response to you.

“We expect to be feature complete in self driving this year, and we expect to be confident enough from our standpoint to say that we think people do not need to touch the wheel and can look out the window sometime probably around … in the second quarter of next year. And we expect to get regulatory approval, at least in some jurisdictions, for that towards the end of next year. That’s roughly the timeline that I expect things to go on.”

 

[Electroman]

This is my definition of Coast to Coast demo. Three different driving phases in this demo:

- the stretches that the car drives autonomously (FSD)
- the stretches that the car drives itself but needs a human supervision
- the stretches that a human has to drive.

Publish those stretches, from NY to LA. Any sections humans need to take over, or humans need to supervise has to be clearly well communicated by the car to the driver, well ahead of time.

If we get to around 90%+ of miles in favor of full autonomous driving for a coast to coast route, and the rest between human supervision and manual driving, to me that is an unqualified success for the FIRST attempt.

100% can wait.

 

[diplomat33]

I'm being totally realistic, because Elon has been 100% consistent about what we're going to get for years now. This is not some arbitrary bar that I came up with for a forum post, but a consistent message from the CEO (and now head of Autopilot team) of Tesla. It has been a core part of Tesla's marketing and sales flow for years now.

Well yes, but it also matters where Tesla is at in their FSD development because that will determine what we can realistically expect to get now. Elon has said that what they have right now is "feature complete". Of course, we should expect Tesla to deliver everything that they promise but demanding or expecting that they give us something that they don't have yet, does not make any sense.

Your cross country example consists of 99.5% freeways though...
0.5% of the miles will actually be city streets.
Not at all convincing nor a representation of what Tesla is selling which is a system that works on all road types and conditions.

True. A coast to coast demo probably won't show a lot of what Tesla's city street FSD is capable of. I think the reason people like the idea of a coast to coast demo is because it feels like a milestone. It feels like an achievement to say that a car can self-drive the length of the entire county. And remember the love that Americans have for long trips. So there is a certain appeal in the idea that the car could self-drive you all the way from one coast to another if you wanted to. And if Tesla's coast to coast demo, even if it is mostly highway driving, is successful, then it will at least prove that Tesla has achieved true highway self-driving.

Here's Elon Musk's response to you:
“We expect to be feature complete in self driving this year, and we expect to be confident enough from our standpoint to say that we think people do not need to touch the wheel and can look out the window sometime probably around … in the second quarter of next year. And we expect to get regulatory approval, at least in some jurisdictions, for that towards the end of next year. That’s roughly the timeline that I expect things to go on.”

I am well aware of Elon's timeline. It is precisely why I wrote what I wrote. A coast to coast demo now that is feature complete but is much better next year or next year is completely consistent with Elon's timeline.