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

[tomas]

They will make big, rash, unrealistic promises to the people and put forward challenges to the legislature to act, but the legislature will remain tangled, nothing will pass, and they will blame the legislature for the delay in implementing their big, rash, unrealistic plan... hey, wait a sec... what are we talking about?

Actually, I predict a theory 6: they will progressively deliver features under the FSD banner beginning in a couple of years. But since Level 5 autonomy will not be legal for decades, they will not be able to deliver all of the things Musk has hinted at. Some owners will feel shorted, there will be a class action suit. Tesla will maintain that they delivered what they could, and the contract details protect them, but settle for refunding a small portion of the $3k... say $1k to the plaintiffs. The lawyers, of course, will take half. Tesla will stop selling the future option. This will briefly be a small dent in Tesla in the media, but shortly after, they will announce the generation 3 roadster, which will have sub 1 second 0-60 and limited flight capability coming at a future date for $10k... and we will be off to the next thing.

 

[JeffK]

If I wanted FSD, I would buy a Nissan Leaf.

Current Nissan Leaf isn't FSD... just so you know. Also, there are some great bloopers from that test vehicle on youtube.

One the website Tesla does say FSD and not SAE level 5 autonomous driving. There's actually a large difference here. Elon has suggested that he believes it will be level 4 or 5 in the future, but that's not a promise on the website when you go to order.

It can easily be qualified as FSD only when lanes are clearly marked and only in good weather. It might be twice as safe as a human when these conditions are met. Who knows?

At the end of the day I'm confident they will offer something that stops at stop lights, makes turns, parks itself, is capable of driving without a human in certain conditions, etc. However, how dynamic this ends up being, remains to be seen.

 

[LosAltosChuck]

It would be nice. But I'm seriously doubting that we'll even see L3 within the next decade. Other than a few staged videos there's not much evidence that they're making a lot of progress.

As of right now, the best system on the market (Tesla) is just a lane follower. And even that has had a number of struggles over the last year. The systems from the other manufacturers are lane followers with a mandatory ping-pong feature. None of these systems make any decisions about how to interact with the traffic or the nav system.

Technology always follows a trajectory. If FSD were in the offing we would be starting to see progressions. Maybe we'd see car initiated lane switching. Or nav triggered off-ramp/on-ramp highway transfers. Or automated stop light/stop sign negotiation. At a minimum there would be rumors of such features being trialed in Tesla's fleet of beta-testers. But nothing. Right now the trajectory for auto-driving tech points to a much longer, slower development process than was expected a couple of years ago.

I'm not particularly upset by that. Technology takes time, and I'm thrilled that Tesla is pushing hard. But I expect that they're going to have to roll back their promises on FSD fairly soon.

I'm not as pessimistic, perhaps because I have a pretty deep inside view of multiple players. Advances are happening rapidly with L3 (Actually L4 systems with close supervision). It is L4 that will take substantial time to fully deploy with the driver out (of the loop). What we'll see are L4 systems in professional environments (car and truck fleets) operated as L3 systems with human safety/supervisory drivers, some operating commercially, and in this decade.

I completely agree with your pessimism for the non-professional market. It will be a while before anyone but Tesla puts untrained civilians in an L4 vehicle. And my guess (based on no inside information, just intuition), is that what Tesla will release will require L3 attention, while perhaps being marketed as L4. (That in my opinion would be a dangerous combination, BTW.)

Tesla's AP system may be the best *on the market*, but they are far from the best technology actually operated (unreleased, under test) by others. Within the development community, Tesla is generally recognized for its aggressiveness (and its ability to collect data from its fleet) but not for the state of its tech. Tesla's ability to collect data from its deployed customer base could turn this around rapidly. The value of that advantage is unmeasurably huge.

I don't mind that Tesla is taking longer than Elon promised. That means someone inside is being careful, and that's a good thing.

 

[JeffK]

is that what Tesla will release will require L3 attention

That's just it, L3 requires a longer transition time between the car and the driver. I think we might actually see more of an L2 "FSD" before anything else.

 

[jimmy_d]

I’m still baffled why anyone checked that box when they ordered their car. Tesla has been very clear that FSD is not yet available and the legalities are yet to be determined in each state. So, I don’t know, they checked the box to save $1,000 some time in the future?

I bought it. On two vehicles already. I'm sending a signal to them that I want whatever they can give me with that sensor suite, and that I'm willing to pay for it. I like the model where they deliver what they can safely provide, when they can safely provide it. I greatly prefer it over the "nothing at all until the whole thing is perfect" approach that every other player seems to be pursuing. Requiring the system to be essentially perfect before anyone can use it is a recipe for postponing the potential benefits into the distant future.

Whatever label you chose to use for the capability, if Tesla can't provide substantial value with the hardware in AP2, then you can assume that truly driverless cars aren't going to happen. Because going 100% driverless is a much, much higher bar than going 90% driverless, much less 50% driverless. What do I think they can deliver? Eventually I think they can deliver a lot. Better safety and a better driving experience by a significant margin over AP1 cars; something that is easily worth a few $K to me.

In all honesty I really do believe that capability worthy of the name FSD is possible with the AP2 hardware suite, though I won't be surprised if it takes years for it to arrive in a form where it can be safely deployed to all the owners of AP2 vehicles.

And if I never get FSD from my AP2 car I still won't feel bad about checking that box, because the lost human potential arising from the "impossibility" of driverless cars will sadden me far beyond the loss of the money.

To me checking that box is a vote for a better future. If you can't afford the ticket or don't believe in the dream then you should just wait until something concrete appears. But to me, it's worth it.

 

[sandpiper]

I'm not as pessimistic, perhaps because I have a pretty deep inside view of multiple players. Advances are happening rapidly with L3 (Actually L4 systems with close supervision). It is L4 that will take substantial time to fully deploy with the driver out (of the loop). What we'll see are L4 systems in professional environments (car and truck fleets) operated as L3 systems with human safety/supervisory drivers, some operating commercially, and in this decade.

I completely agree with your pessimism for the non-professional market. It will be a while before anyone but Tesla puts untrained civilians in an L4 vehicle. And my guess (based on no inside information, just intuition), is that what Tesla will release will require L3 attention, while perhaps being marketed as L4. (That in my opinion would be a dangerous combination, BTW.)

Tesla's AP system may be the best *on the market*, but they are far from the best technology actually operated (unreleased, under test) by others. Within the development community, Tesla is generally recognized for its aggressiveness (and its ability to collect data from its fleet) but not for the state of its tech. Tesla's ability to collect data from its deployed customer base could turn this around rapidly. The value of that advantage is unmeasurably huge.

I don't mind that Tesla is taking longer than Elon promised. That means someone inside is being careful, and that's a good thing.

Interesting. I'd love to ask you to elaborate on your first paragraph, but clearly you're not able to do so. If the L3/L4 systems are coming close, why don't we see evidence of some aspects of it moving into the market? For the sake of safety, there should be a slow roll-out of the L3/4 capabilities, into L2 systems. There's incredible advantage to slowly acclimating drivers, rather than doing a shock roll out. Plus, companies need time to tune the features, to make them usable. Yet we see none of this.

 

[d21mike]

Autopilot is the most import reason I ordered a Tesla 9 days ago and check the box for FSD. And before I did I watched a lot of Youtube Videos on the current state of EAP. I decided to go with Tesla mainly for the future of Autopilot. I was considering a Mercedes with their Version (my wife has a 2016 GLC without it) and Cadillac's CT6 Super Cruise which is coming out soon (which is only for freeways). But I felt the most confidence in Tesla's Vision with OTA Updates. Also, for me spending around $90k for a car (which I normally would not do) the $3k was simply not a concern. Since I fully expect FSD to come out in drops over time, like what is happening with EAP, I would not no when to pull the trigger on the $4K in the future. But now I have it so no worries. I fully expect in 2018 we will get some of the features. Based on what I have seen on the videos I think stopping at red lights and stop signs would be a huge help and I think it is a FSD Only Feature and not a EAP Feature. EAP is supposed to get you from on ramp to off ramp at some point but that to me is less important then the stopping on side streets. If it turns out I am completely wrong it will not be the first time. But I am betting that we will see progress on FSD but I expect will never be L5 while I still own this car.

 

[zmarty]

1. Tesla will release a door to door L3 (NOT 4 or 5) system that will follow a navigation plan, will require an alert driver (possibly the same hands on wheel required now, unless they retrofit an inward facing camera for driver monitoring.

L3 is a standard and means something very specific. Tesla cannot make up the rules:

1. The driver does NOT have to keep their hands on the steering wheel
2. The driver is NOT required to monitor the flow of traffic (they can do something else).
3. The car needs to give a reasonable amount of time to the driver (several seconds) to take over. That automatically means the car needs to be able to handle any short term emergencies.

I would be fine with AP2.X achieving L3 only actually. I researched the current sensor suite for a long time and I doubt it will ever get to L4. I expect the next version of the suite to have solid state LIDAR in addition to the rest of the current sensors. I am also concerned about the height of the front side cameras, they should be on the mirrors or higher.

 

[MP3Mike]

I am also concerned about the height of the front side cameras, they should be on the mirrors or higher.

Do you mean the side front looking cameras? If so those are already above the mirrors, on the B pillars.

 

[LosAltosChuck]

Interesting. I'd love to ask you to elaborate on your first paragraph, but clearly you're not able to do so. If the L3/L4 systems are coming close, why don't we see evidence of some aspects of it moving into the market? For the sake of safety, there should be a slow roll-out of the L3/4 capabilities, into L2 systems. There's incredible advantage to slowly acclimating drivers, rather than doing a shock roll out. Plus, companies need time to tune the features, to make them usable. Yet we see none of this.

The traditional automakers are extremely conservative, so expect their rollout pace to be slow, and tied to model years. Tesla takes a different approach, and specifically does not tie releases to model years. Plus they are the most aggressive at crowdsourcing their training data.

I think many (including me) are very concerned with L3 automation. L3 requires that a driver stay alert and prepared for immediate takeover. That's a tough ask, given all the distractions available to drivers who think they can trust the system. A well respected NASA researcher on human factors in automation compared aircraft automation to car automation. As he put it, if he literally tossed you out the door of an airliner at altitude, you have 3 minutes to decide what to do next. There is almost always time to intervene in an airplane, even if the pilot is distracted. With a car, you have to react in less than a second. It is a huge ask, and no one has shown that such a reaction can be depended upon every time.

So even adding a "simple" stoplight/stop sign recognition and response system would be a great addition - as long as it works. But a single failure requiring instant human response is a terrifying concept, because there *will* be failures, and they *will* happen with a distracted driver.

This is what keeps the automakers from dribbling L3/4 features into L1/L2 environments. They want to be very sure the stuff works before inflicting it on a non-professional driving public, with all the liabilities that will acrue.

 

[zmarty]

Do you mean the side front looking cameras? If so those are already above the mirrors, on the B pillars.

No, the other two side cameras. The ones in the logos that look backwards. By front I meant they are in front, not that they look forward

 

[zmarty]

I think many (including me) are very concerned with L3 automation. L3 requires that a driver stay alert and prepared for immediate takeover.

See my post above above. People here seem to misunderstand what L3 means because of their experience with the current Autopilot. In L3 the driver needs to be lucid but does not have to actively monitor the traffic or what's going around the car all the time. L3 is not like the current Autopilot where you have to take over in hundreds of milliseconds because the car decided to swerve into a barrier. There is a large divide between the capabilities of L2 and L3 systems. I would say L3 is probably two orders of magnitudes harder than L2.

 

[Bebop]

It's really simple actually and not that complicated.

Tesla will retrofit any and all additional required hardware to make FSD possible. Otherwise, like many people here would, will sue the s*** out of Tesla for selling me something that isn't.

They keep advertising FSD on the new order page still to this day. Even about a year later. I'm not concerned at all. No reason to be speculating what if.

 

[oktane]

It's really simple actually and not that complicated.

Tesla will retrofit any and all additional required hardware to make FSD possible. Otherwise, like many people here would, will sue the s*** out of Tesla for selling me something that isn't.

They keep advertising FSD on the new order page still to this day. Even about a year later. I'm not concerned at all. No reason to be speculating what if.

Well you should be concerned, because one year after enhanced auto pilot was supposed to launch they still can't stay with in a lane or have auto dimming headlights that work properly or have rain sensing wipers that work or have summon work without crashing into a garage.

The more time that passes, the less likely it is that they will ever be able to get these features working.

 

[alcibiades]

Well you should be concerned, because one year after enhanced auto pilot was supposed to launch they still can't stay with in a lane or have auto dimming headlights that work properly or have rain sensing wipers that work or have summon work without crashing into a garage.

I think you're getting too caught up in the details!

 

[Canuck]

It is a misconception that "the regulators" have anything to do with the implementation of an autonomous system. There is no close federal (or state) oversight that requires certain redundancies, etc.

That's a very strange statement you make. We can't have a misconception about how self-driving vehicles are regulated since they haven't drafted the regulations yet -- hence Tesla's caveat regarding FSD.

Congress Finally Tackles the Murky World of Robocar Rules

To suggest that the regulators won't require redundancy, etc. is something no one knows at this time. I suggest it's highly likely the regulations will require things like redundancy since we know how safety regulations for vehicles currently work and they do deal with particulars, especially with regard to safety issues.

 

[alcibiades]

That's a very strange statement you make. We can't have a misconception about how self-driving vehicles are regulated since they haven't drafted the regulations yet -- hence Tesla's caveat...

Congress Finally Tackles the Murky World of Robocar Rules

To suggest that the regulators won't require redundancy, etc. is something no one knows at this time. I suggest it's highly likely the regulations will require things like redundancy since we know how safety regulations for vehicles currently work and they do deal with particulars, especially with regard to safety issues.

While not having a strict designation on the Autonomy Level Scale, why couldn't a car manufacturer have a feature called "Superduper Autopilot" which follows routes and obeys traffic laws and doesn't run people over BUT requires a driver to be ready to take over? Is that regulated already?

 

[Bebop]

Well you should be concerned, because one year after enhanced auto pilot was supposed to launch they still can't stay with in a lane or have auto dimming headlights that work properly or have rain sensing wipers that work or have summon work without crashing into a garage.

The more time that passes, the less likely it is that they will ever be able to get these features working.

Being negative doesn't solve any problems. We just have to be optimistic about the future and I have full faith in Elon on delivering. Let's wait for 9.0 before we jump ship here. I expect major features in 9.0.

 

[Doug_G]

The term "Full Self Drive" is ambiguous enough that Tesla can move the goal posts. They didn't promise a specific automation level. That will probably make it difficult to sue them.

If they really pull it off then at best they will achieve a Level 3 system (human override is the fallback). It's far more likely they'll achieve Level 2 (human has to constantly monitor).

I think I can safely predict that they will NOT achieve a Level 5 system in the next four years. Level 4 in that time frame would be an astounding accomplishment. And whether they achieve Level 3 before I trade in my brand-new car... questionable. That is why I didn't select the FSD option.

Elon may be correct that the current sensor suite is sufficient for FSD. But there is no way to know that for sure until you actually get a system fully working.

Software is the huge hurdle. Getting this to work under ideal conditions will be challenging. The edge cases are killer.

What happens when you approach a construction zone, with poorly-positioned cones directing you across the lane markings and a lazy guy kinda-sorta ambiguously holding up a stop/slow sign (like what I encountered on my commute today)? What happens when it snows and you can't see the lanes? etc.

Also it is completely unknowable at this point how much processor power is required.

So... in summary, they will achieve some as-yet-unknown level of automation, move the goalposts, and declare victory.

 

[LosAltosChuck]

L3 is a standard and means something very specific. Tesla cannot make up the rules:

1. The driver does NOT have to keep their hands on the steering wheel
2. The driver is NOT required to monitor the flow of traffic (they can do something else).
3. The car needs to give a reasonable amount of time to the driver (several seconds) to take over. That automatically means the car needs to be able to handle any short term emergencies.

I would be fine with AP2.X achieving L3 only actually. I researched the current sensor suite for a long time and I doubt it will ever get to L4. I expect the next version of the suite to have solid state LIDAR in addition to the rest of the current sensors. I am also concerned about the height of the front side cameras, they should be on the mirrors or higher.

Weeeellll, I'll have to beg to differ on your interpretation.

To your numbered points:

1. Keeping hands on the wheel is not for vehicle control, it is the (currently) only way Tesla can test that the driver is actually (mentally) present. That isn't something specific to the SAE level, it is a safety mitigation technique. If Tesla retrofits a driver monitoring camera, they could possibly eliminate the need for hands on wheel, assuming the camera system can assure that the driver is present (or in SAE-lingo, the driver is "fallback-ready").

2. The standard says nothing of the sort. It says the driver must be "fallback-ready". That inherently means not distracted. At L3, the driver doesn't need to *react* to the flow (objects and events in SAE terms), but they need to be aware of them.

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.

Perhaps another way to understand the standard is this, which isn't reflected in the chart: At L3, the system is required to perform the DDT as long as the vehicle operates within the expected environment (SAE calls this the Operational Design Domain (ODD)). It alerts the driver when it is no longer within the expected environment. The maker of the system determines the ODD, not the SAE. At L3, it is the *driver's* responsibility to return the vehicle to a safe state (the "minimal risk condition") when the vehicle exits from the ODD. This is distinguished from L4, where it is the *system's* responsibility to do this, and L4 doesn't depend on a fallback-ready driver. The "several seconds" at L3 is *only* to cover the transition time from automation to the alert driver, with a small buffer to account for the range of response times of alert drivers.

So the bottom line is this: At L3, according to SAE standards and common sense, the driver *can't* be distracted. The vehicle can do all the actual work of driving door to door, but the driver *must* be ready to take over immediately.