Let's Speculate - What will happen if FSD isn't possible on cars sold with that option?

[ZeApelido]

Simple question: I know of a self-driving device. It does not have lidar. It has only two cameras that are quite close together, so parallax is limited for distance sensing. Its reaction times are, in robotic terms, glacially slow. When it is fully engaged it does a pretty good job of making judgments, at least when the software is appropriate to the task and when judgment is used to anticipate problems.

The device is a human driver. Yes, there are lots of accidents in automobiles. But the accident rates would be much lower without performance degrading factors like fatigue, chemicals and distractions.

Is this not a demonstration that a camera-based self-driving device is possible, especially when there are 8 cameras, a very powerful computer and radar on board? Isn't it therefore largely a question of software development?

Progress in software development is notoriously hard to predict, because it depends to a great degree upon breakthroughs that are impossible to anticipate.

Landing a rocket booster tail first was thought by experts to be impossible, until it was done--by means of breakthroughs in guidance technology--software, in other words.

Should be interesting to watch...

Imagine wearing glasses that pixelate and lower the re solution of what your can see.

 

[bigdaddie]

I think that Tesla is developing two software versions now: one to keep us pacified and the FSD version. IMHO there are probably Teslas driving themselves all over Fremont, be it supervised. We are only getting the branch of the code that improves the lane keeping capabilities. My guess is that there will be 3-4 major releases over the next couple of years that will make us go WOW! Elon is not going to be tech Trumped (maybe a pun) by Cadilac, Nissan, or Audi.

 

[calisnow]

There are numerous threads about whether or not Tesla can deliver on the promise of FSD on cars sold with that option, and while there are occasionally posts about what they might do if they can't deliver I have been unable to find a thread specifically for speculating about that.

Who's got the baseball bats?

 

[Bladerskb]

Well, you can assert this but please read the SAE definition of L3 and I think you'll see this is not the case. The SAE definition of L3 explicitly defines the nature of disengagement, the responsibility of the system, and the responsibility of the driver in the face of automation and vehicle failures, some of which *can't* be detected by the automation but *must* still be detected by the driver. That won't happen if the driver isn't paying attention.

OEMs can demo whatever they want, it's what they ship that matters, and shipping L3 with the message that it will "NEVER" disengage is downright dangerous and inaccurate.

The SAE standard does not define a specific takeover time, it leaves that to the implementor and will obviously be implementation AND domain sensitive. I can easily imagine a L3 traffic jam assist that could allow for 10-15 second takeovers. Nothing happens (outside of meteors) inside a traffic jam that an automated system can't handle well. I have a hard time imagining an L3 highway speed autopilot that would offer 10-15 second takeover time. That would be roughly 1000-1500 feet, which is, by the way, beyond the range of existing radar, lidar, and camera perception systems. I would guess an L3 highway autopilot would give you on the order of 3 seconds max to respond, and even that would be best case. Existing sensors (of all types) don't reach beyond 3-500 feet, and at highway speeds you're traveling on the order of 100 feet per second. Without an external stimulus the vehicle won't even perceive the emergency until you're within a very small number of seconds.

3. At L3, the driver must be "fallback-ready". This means they need to be ready to make strategic decisions quickly (should I swerve, stop, or accelerate, or some combination). A driver "doing something else" is not "fallback-ready". So you don't get to text or watch a movie at L3. The confusion comes from the SAE standard that the Dynamic Driving Task (DDT) must continue for several seconds. This means that at L3, the automated system can't stop doing longitudinal or lateral control or object and event response execution. This is to allow a sufficiently graceful transition to the user. Note that even for an alert, engaged driver, it can take 1.5 seconds or more to begin to react to an event. At L3, "several seconds" is to account for transition to an engaged user, not a distracted user. Small things, like swerving to avoid a skunk while alerting the driver to take over, are the sorts of things an L3 system may be able to handle gracefully (but see below). Dealing with, for example, a suddenly jackknifing 18 wheeler on the car's right front bumper is likely something an L3 system is not going to handle for "several seconds" while a distracted driver gets their wits back about themselves to take over.

This is not how L3 works. A l3 won't alert the driver to take over if it detects a skunk and has to swerve for it. A L3 will handle a "suddenly jackknifing 18 wheeler on the car's right front bumper" and never disengage or request to take over. These are easy scenarios.

Well, you can assert this but please read the SAE definition of L3 and I think you'll see this is not the case.

No YOU are the one asserting BS and not what the SAE says. stop saying read the SAE when you are the one who can't understand.

unless you are reading a different document than the rest of us humans.

https://www.sae.org/misc/pdfs/automated_driving.pdf

"A key distinction is between level 2, where the human driver performs part of the dynamic driving task, and level 3, where the automated driving system performs the entire dynamic driving task."

"the driving mode-specific performance by an automated driving system of all aspects of the dynamic driving task with the expectation that the human driver will respond appropriately to a request to intervene."

"Dynamic driving task includes the operational (steering, braking, accelerating, monitoring the vehicle and roadway) and tactical (responding to events, determining when to change lanes, turn, use signals, etc.) aspects of the driving task"


Here is the SAE levels:

SAE automated vehicle classifications:

• Level 0: Automated system issues warnings but has no vehicle control.
• Level 1 (”hands on”): Driver and automated system shares control over the vehicle. An example would be Adaptive Cruise Control (ACC) where the driver controls steering and the automated system controls speed. Using Parking Assistance, steering is automated while speed is manual. The driver must be ready to retake full control at any time. Lane Keeping Assistance (LKA) Type II is a further example of level 1 self driving.
• Level 2 (”hands off”): The automated system takes full control of the vehicle (accelerating, braking, and steering). The driver must monitor the driving and be prepared to immediately intervene at any time if the automated system fails to respond properly. The shorthand ”hands off” is not meant to be taken literally. In fact, contact between hand and wheel is often mandatory during SAE 2 driving, to confirm that the driver is ready to intervene.
• Level 3 (”eyes off”): The driver can safely turn their attention away from the driving tasks, e.g. the driver can text or watch a movie. The vehicle will handle situations that call for an immediate response, like emergency braking. The driver must still be prepared to intervene within some limited time, specified by the manufacturer, when called upon by the vehicle to do so.
• Level 4 (”mind off”): As level 3, but no driver attention is ever required for safety, i.e. the driver may safely go to sleep or leave the driver's seat. Self driving is supported only in limited areas (geofenced) or under special circumstances, like traffic jams. Outside of these areas or circumstances, the vehicle must be able to safely abort the trip, i.e. park the car, if the driver does not retake control.
• Level 5 (”wheel optional”): No human intervention is required. An example would be a robot taxi.

You are so misinformed is not even funny. it looks like its the run of the mill around this forum. A whole bunch of people who take their personal opinion above mountains of evidence and facts.

Basically you are saying all automakers and all self driving tech researchers. All hundreds of them are wrong and you are the only singular human on earth who has it right.

Existing sensors (of all types) don't reach beyond 3-500 feet

lmao tesla themselves has cameras that see up to 820 ft.
google's lidar and cameras can see well beyond 1100ft.
radar also is well beyond 500ft.

you're clearly out of your elements and don't know what you are talking about.

 

[JeffK]

google's lidar and cameras can see well beyond 1100ft.

Do you have a source or is that range just the cameras? Even Velodyne's $70K+ HDL-64E only has a range of 120 m (393 ft)
HDL-64E

The smaller, cheaper, models have a range of 100 m.
The startup Luminar claims 200 m (656 ft) with theirs.

 

[LosAltosChuck]

This is not how L3 works. A l3 won't alert the driver to take over if it detects a skunk and has to swerve for it. A L3 will handle a "suddenly jackknifing 18 wheeler on the car's right front bumper" and never disengage or request to take over. These are easy scenarios.



No YOU are the one asserting BS and not what the SAE says. stop saying read the SAE when you are the one who can't understand.

unless you are reading a different document than the rest of us humans.

https://www.sae.org/misc/pdfs/automated_driving.pdf

"A key distinction is between level 2, where the human driver performs part of the dynamic driving task, and level 3, where the automated driving system performs the entire dynamic driving task."

"the driving mode-specific performance by an automated driving system of all aspects of the dynamic driving task with the expectation that the human driver will respond appropriately to a request to intervene."

"Dynamic driving task includes the operational (steering, braking, accelerating, monitoring the vehicle and roadway) and tactical (responding to events, determining when to change lanes, turn, use signals, etc.) aspects of the driving task"


Here is the SAE levels:

SAE automated vehicle classifications:

• Level 0: Automated system issues warnings but has no vehicle control.
• Level 1 (”hands on”): Driver and automated system shares control over the vehicle. An example would be Adaptive Cruise Control (ACC) where the driver controls steering and the automated system controls speed. Using Parking Assistance, steering is automated while speed is manual. The driver must be ready to retake full control at any time. Lane Keeping Assistance (LKA) Type II is a further example of level 1 self driving.
• Level 2 (”hands off”): The automated system takes full control of the vehicle (accelerating, braking, and steering). The driver must monitor the driving and be prepared to immediately intervene at any time if the automated system fails to respond properly. The shorthand ”hands off” is not meant to be taken literally. In fact, contact between hand and wheel is often mandatory during SAE 2 driving, to confirm that the driver is ready to intervene.
• Level 3 (”eyes off”): The driver can safely turn their attention away from the driving tasks, e.g. the driver can text or watch a movie. The vehicle will handle situations that call for an immediate response, like emergency braking. The driver must still be prepared to intervene within some limited time, specified by the manufacturer, when called upon by the vehicle to do so.
• Level 4 (”mind off”): As level 3, but no driver attention is ever required for safety, i.e. the driver may safely go to sleep or leave the driver's seat. Self driving is supported only in limited areas (geofenced) or under special circumstances, like traffic jams. Outside of these areas or circumstances, the vehicle must be able to safely abort the trip, i.e. park the car, if the driver does not retake control.
• Level 5 (”wheel optional”): No human intervention is required. An example would be a robot taxi.

You are so misinformed is not even funny. it looks like its the run of the mill around this forum. A whole bunch of people who take their personal opinion above mountains of evidence and facts.

Basically you are saying all automakers and all self driving tech researchers. All hundreds of them are wrong and you are the only singular human on earth who has it right.



lmao tesla themselves has cameras that see up to 820 ft.
google's lidar and cameras can see well beyond 1100ft.
radar also is well beyond 500ft.

you're clearly out of your elements and don't know what you are talking about.

You are embarrassing yourself. The specs for all these sensors are readily available. And I'm so far out of my element that for some bizarre reason, leading autonomous vehicle companies hire me to lead their production engineering projects.

Some simple facts: an automated vehicle with today's sensors don't perceive reliably beyond about 200 meters. A camera can "see" line of sight to the horizon, but the pixel density of the sensor doesn't allow it to interpret what it is seeing beyond on the order of 200 meters. Objects occupy one pixel or less at the longer distances. One could fuse in a narrow FOV camera to detect at longer distances, but the limited FOV would limit the sorts of threats that could be detected (eg, no side threats with a forward looking narrow camera). One could add a *bunch* of narrow FOV cameras, covering 360 degrees, but then you'd have a trunk full of hardware to process all those feeds. And the wider the view, the shorter the range of the camera. Because pixels.

Lidar doesn't help. Lidar goes about 100 meters. Some go 150 meters. Many go 80 meters. I'd love to hear about a production Lidar that is useful well beyond 1100 feet. Please enlighten me (with a link).

Automotive radar goes about 200 meters at best.

I keep wasting my breath arguing with folks like you because it is such attitudes that will cause accidents with L3 systems and will set back the entire industry. L3 does not allow the driver to be a passenger. Passengers can watch movies (or sleep). "Receptive Users" stay aware of the vehicle systems and environment. L3 requires a receptive user. (Look it up.) Receptive users are expected to take over *even if the L3 system doesn't request a takeover* if needed. (Look it up.) One can't meet that expectation while watching a movie.

By the way, the document you linked is a summary. If you really want to understand SAE J3016 you should read the actual, full document. (Latest: J3016-201609.) There's something else you should know. This is all a moving target. The SAE document is a recommended practice document that is evolving as we learn more about these systems. I expect this to all tighten up as we gain practical experience.

I hope we never see L3 systems in the wild for non-professional drivers. I personally only work on L3 systems if they are targetting to professional drivers. I hope Tesla avoids L3 and produces an honest L4 FSD.

 

[aWalkingShadow]

I keep wasting my breath arguing with folks like you because it is such attitudes that will cause accidents with L3 systems and will set back the entire industry. L3 does not allow the driver to be a passenger. Passengers can watch movies (or sleep). "Receptive Users" stay aware of the vehicle systems and environment. L3 requires a receptive user. (Look it up.) Receptive users are expected to take over *even if the L3 system doesn't request a takeover* if needed. (Look it up.) One can't meet that expectation while watching a movie.

What's the difference between L2 and L3? If the human is supposed to take over even when not alerted by the L3 system, how is this different than L2? It seems like the human has to monitor the environment in the L3 just as much as in a L2.

 

[calisnow]

You are embarrassing yourself. The specs for all these sensors are readily available. And I'm so far out of my element that for some bizarre reason, leading autonomous vehicle companies hire me to lead their production engineering projects.

Some simple facts: an automated vehicle with today's sensors don't perceive reliably beyond about 200 meters. A camera can "see" line of sight to the horizon, but the pixel density of the sensor doesn't allow it to interpret what it is seeing beyond on the order of 200 meters. Objects occupy one pixel or less at the longer distances. One could fuse in a narrow FOV camera to detect at longer distances, but the limited FOV would limit the sorts of threats that could be detected (eg, no side threats with a forward looking narrow camera). One could add a *bunch* of narrow FOV cameras, covering 360 degrees, but then you'd have a trunk full of hardware to process all those feeds. And the wider the view, the shorter the range of the camera. Because pixels.

Lidar doesn't help. Lidar goes about 100 meters. Some go 150 meters. Many go 80 meters. I'd love to hear about a production Lidar that is useful well beyond 1100 feet. Please enlighten me (with a link).

Automotive radar goes about 200 meters at best.

I keep wasting my breath arguing with folks like you because it is such attitudes that will cause accidents with L3 systems and will set back the entire industry. L3 does not allow the driver to be a passenger. Passengers can watch movies (or sleep). "Receptive Users" stay aware of the vehicle systems and environment. L3 requires a receptive user. (Look it up.) Receptive users are expected to take over *even if the L3 system doesn't request a takeover* if needed. (Look it up.) One can't meet that expectation while watching a movie.

By the way, the document you linked is a summary. If you really want to understand SAE J3016 you should read the actual, full document. (Latest: J3016-201609.) There's something else you should know. This is all a moving target. The SAE document is a recommended practice document that is evolving as we learn more about these systems. I expect this to all tighten up as we gain practical experience.

I hope we never see L3 systems in the wild for non-professional drivers. I personally only work on L3 systems if they are targetting to professional drivers. I hope Tesla avoids L3 and produces an honest L4 FSD.

LOL - sure buddy. You're a production engineer for "leading autonomous vehicle companies" who spends his time arguing on the internet and getting in flame wars to educate the people. Right...

 

[LosAltosChuck]

[

LOL - sure buddy. You're a production engineer for "leading autonomous vehicle companies" who spends his time arguing on the internet and getting in flame wars to educate the people. Right...

Um, I'm a VP. Leading production engineering for a company I choose not to mention in the field of autonomous vehicles. I happen to own a Tesla and enjoy this forum. Discussions related to autonomy are a special interest. So shoot me for participating. Or block me I won't be offended.

 

[JeffK]

Um, I'm a VP. Leading production engineering for a company I choose not to mention in the field of autonomous vehicles. I happen to own a Tesla and enjoy this forum. Discussions related to autonomy are a special interest. So shoot me for participating. Or block me I won't be offended.

Peloton?

 

[agtdDelirium]

Well, whatever L3 means, it's the level that I would consider FSD an acceptable option for charging money for. I payed my $3K upfront, knowing that I would not see it for some time (maybe several years). I looked at the 3% premium on the car as an investment into a company that I want to see succeed, not only as self driving technology, but a company that is pushing transportation away from fossil fuels. Without Tesla, I don't think we would be seeing multiple automakers making statements that their fleet will be electrified to at least to some extent within the next several years.

Will I be upset if I don't at least see a divergence in abilities between EAP and FSD cars within the next year? Yeah, I will. I would like to see something for my investment, even if it's just safety related. One of the driving factors in my mind at the time of purchase was why would I only use 4 cameras when the car has eight? Even if the difference becomes alerting me to a bad lane change and preventing one accident, it would be worth the $3000. I'll be happy if it only ends up being a safety advantage rather than being able to take me to work and back. But with that said, I want it to pick me up at my front door and take me to work, all the while allowing me to check my emails on the way.

 

[verygreen]

So I was just rereading what Tesla says NOW when ordering FSD and I realized a few things:

The system is designed to be able to conduct short and long distance trips with no action required by the person in the driver’s seat.
...
When you arrive at your destination, simply step out at the entrance and your car will enter park seek mode, automatically search for a spot and park itself. A tap on your phone summons it back to you.
...
Please note also that using a self-driving Tesla for car sharing and ride hailing for friends and family is fine, but doing so for revenue purposes will only be permissible on the Tesla Network

So what we glean from this? short and long distance trips are only promised with somebody in the driver's seat, but no promises of truly driver-less operations. Nowhere do they make any claims about level of attention required at that, just that no action should be required.
BUT! They don't promise you any of that! They say the system is designed to be able, but no word on whenever the actual implementation ever gets there.
The only direct reference for driverless operations is a park-seek mode (which is supposedly a very low speed in some very restricted conditions).

The car sharing thing I now believe is a red-herring, perhaps they only mean car sharing same as what we have now: contact the site, get the car location, walk up to it and drive off, drop the car when done and the next person would pick it from there.
Similar for friends and family: they take your car and it drives them whereever and then they return the car back to you. No driver-less operations.

So with this in mind I think they can announce FSD arrival much sooner than anybody anticipates, but the actual list of features would be rather underwhelming.

 

[Bladerskb]

So I w
as just rereading what Tesla says NOW when ordering FSD and I realized a few things:


So what we glean from this? short and long distance trips are only promised with somebody in the driver's seat, but no promises of truly driver-less operations. Nowhere do they make any claims about level of attention required at that, just that no action should be required.
BUT! They don't promise you any of that! They say the system is designed to be able, but no word on whenever the actual implementation ever gets there.
The only direct reference for driverless operations is a park-seek mode (which is supposedly a very low speed in some very restricted conditions).

The car sharing thing I now believe is a red-herring, perhaps they only mean car sharing same as what we have now: contact the site, get the car location, walk up to it and drive off, drop the car when done and the next person would pick it from there.
Similar for friends and family: they take your car and it drives them whereever and then they return the car back to you. No driver-less operations.

So with this in mind I think they can announce FSD arrival much sooner than anybody anticipates, but the actual list of features would be rather underwhelming.

Well they purposely left it up to Elon to do the hyping. Elon has been hyping up driverless summon from across the country for a while now and he goes around claiming level 5 FSD and sleeping in the car.

January 2016: "In ~2 years, summon should work anywhere connected by land & not blocked by borders, eg you're in LA and the car is in NY"

https://twitter.com/elonmusk/status/686279251293777920

 

[verygreen]

Well they purposely left it up to Elon to do the hyping. Elon has been hyping up driverless summon from across the country for a while now and he goes around claiming level 5 FSD and sleeping in the car.

January 2016: "In ~2 years, summon should work anywhere connected by land & not blocked by borders, eg you're in LA and the car is in NY"

https://twitter.com/elonmusk/status/686279251293777920

Well, clearly he means that while now summon requires you to be near the car (either proximity by the keyfob) or geolocation of your phone.
You would be able to move your car out of the garage/whatever so your wife can clean the floor while you are on your business trip to LA, nothing more than that!

 

[LosAltosChuck]

Peloton?

No.

 

[McRat]

Sounds like the common person will soon find out what the status of autonomy is:

Lyft to offer rides in self-driving cars in San Francisco area

It's probably not the Bolt, even though the Bolts have been mapping SF streets.

 

[McRat]

No.

The best course of action for folk with NDAs is not to even discuss particulars. The interwebbythingy is persistant.

 

[Bladerskb]

Some simple facts: an automated vehicle with today's sensors don't perceive reliably beyond about 200 meters.

Wait you said 3-500 feet (152 meters) before, now you are saying 200 meters (656 ft.) So which is it?

Stop goal shifting.

Lidar doesn't help. Lidar goes about 100 meters. Some go 150 meters. Many go 80 meters. I'd love to hear about a production Lidar that is useful well beyond 1100 feet. Please enlighten me (with a link).

Google's has a custom built lidar that goes over 1100 ft (over 300 meters), infact they say they "developed a first-of-its-kind long range LiDAR that’s capable of quickly zooming into objects on the road. This powerful LiDAR lets us see a football helmet two full football fields away."

Introducing Waymo’s suite of custom-built, self-driving hardware

Audi A8 2018 production car has lidar that has a 200 meters range.

Hands on -- and off -- with the Audi A8's Level 3 self-driving system

Automotive radar goes about 200 meters at best.

what happened to the statement "Existing sensors (of all types) don't reach beyond 3-500 feet" that you made?

Nevertheless, The audi A8 2018 production car has radar with 240 meters range.

Hands on -- and off -- with the Audi A8's Level 3 self-driving system

lmao now go kick rocks!

Its so funny how you started out with 150 meters MAX but after getting called out on your bs, shifted to "200 meters" and still got 0wned!

bahahaha

You are embarrassing yourself. The specs for all these sensors are readily available. And I'm so far out of my element that for some bizarre reason, leading autonomous vehicle companies hire me to lead their production engineering projects.

lol so pathetic. If you are VP of Production Engineering, then i'm President Trump Lmao!

 

[JeffK]

Google's has a custom built lidar that goes over 1100 ft (over 300 meters), infact they say they "developed a first-of-its-kind long range LiDAR that’s capable of quickly zooming into objects on the road. This powerful LiDAR lets us see a football helmet two full football fields away."

Two football fields are about 183 m or 600 ft.

Do you have a link to a technical doc about their lidar? I want to read if they have a similar approach to Luminar's 200 m unit.

 

[Bladerskb]

I keep wasting my breath arguing with folks like you because it is such attitudes that will cause accidents with L3 systems and will set back the entire industry. L3 does not allow the driver to be a passenger. Passengers can watch movies (or sleep). "Receptive Users" stay aware of the vehicle systems and environment. L3 requires a receptive user. (Look it up.) Receptive users are expected to take over *even if the L3 system doesn't request a takeover* if needed. (Look it up.) One can't meet that expectation while watching a movie.

By the way, the document you linked is a summary. If you really want to understand SAE J3016 you should read the actual, full document. (Latest: J3016-201609.) There's something else you should know. This is all a moving target. The SAE document is a recommended practice document that is evolving as we learn more about these systems. I expect this to all tighten up as we gain practical experience.

I hope we never see L3 systems in the wild for non-professional drivers. I personally only work on L3 systems if they are targetting to professional drivers. I hope Tesla avoids L3 and produces an honest L4 FSD.

Since i'm a nice guy, i'm gonna help our VP of production engineering friend here. We wouldn't want him to lose his job because of incompetency now would we?

What our friend fail to understand is that the 3.12 page 12 of the SAE document is referring to vehicle failure such as your tire rod popping off , tire blowout or brake failure not the ADS failing, because in level 3 the ADS CANNOT fail, but there are vehicle conditions that can affect its performance which is why you can't sleep in a L3 but you also don't have to monitor the system or the road either. Only to be aware in case of a vehicle failure.

https://wiki.unece.org/download/attachments/40009763/(ITS_AD-10-08) SAE_J3016_Taxonomy and Definitions for Terms Related to Driving Automation Systems.pdf?api=v2

it says:

1. In level 3 driving automation, a DDT fallback-ready user is considered to be receptive to a request to intervene and/or to an evident vehicle system failure, whether or not the ADS issues a request to intervene as a result of such a vehicle system failure.

It then gives examples:

• EXAMPLE 1: While a level 3 ADS is performing the DDT in stop-and-go traffic, the left-front tie rod breaks. The DDT fallback-ready user feels that the vehicle has pulled dramatically to the left and intervenes in order to move the vehicle onto the road shoulder. EXAMPLE 2: While a level 3 ADS is performing the DDT on a free-flowing highway, the left side mirror glass falls out of the housing. The DDT fallback-ready user, while receptive, does not and is not expected to notice this failure, because it is not apparent.

@LosAltosChuck I just saved you... your welcome!