Enhanced Autopilot 2 may never happen before Full Self Driving is ready!

[eloder]

No. Even with WAAS, no. Not granular enough.

GPS alone can narrow it down to within a three lane width, but if only Tesla had some sort of hardware equipped on all its cars, along with high-resolution map building that has been happening across its fleet for over a year now, that could help fill in the gaps...

Tesla will never have such magical devices though, it's impossible. Let's cancel autopilot, Elon Musk is not intelligent enough to figure out how to have its autopilot system remember which lane it's in using high-resolution maps and multiple cameras. Despite the fact that he can make self-driving cars this obstacle is just way too much.

Why not bring up an actual challenge Elon and his team has to face, for example driving in ice, rather than one of the most trivial out there? It's like saying autopilot can't go forward because they haven't designed a UI for it that the human watches.

 

[andrewket]

There is also GLONASS, which is supposedly good within 2 meters. Using both GLONASS and GPS I believe also has some advantages, and many modern chipsets contain both.

 

[brkaus]

Yes, GPS is accurate to a few meters. The camera can (and does) center the car in the lane. The camera can also identify if it is in a left, center, or right lane based on line coloring and other visual cues

The navigation will be able to identify if the exit is ahead 1/2 mile and move to the right.

 

[Dstrohl]

Ok, let me try to address some of these points without name calling...

1. Automatic Lane Changes:

• You are correct that the car would need to know what lane it is in, and what lane to be in... and you are correct in the inaccuracies of GPS. In general, GPS can get to as low as 0.01m in accuracy with post processing and correcting technologies, but without those, the best that is theoretically possible is about 2m, and 4m is more generally accepted as the accuracy goal. Even with the 2m best case, that is a lateral accuracy of 6.5 feel, or clear in another lane. However, the cameras should be able to tell if you are in a middle lane, or a left edge/right edge one... and therefore, they should be able to decide to move to the left or right as needed for the exit.
• The information about if an exit is a left or right exit, and how many lanes there are are available is normally in the GPS database, as well as advertised with lane markers and street signs, if the roads follow the NHTSA rules *(in the US) this should work for the majority of cases. (limiting this to talking about highways now). So, if the car knows that an exit is coming up on the left, it can start moving left at least x minutes early, with x defined based on traffic.
• In terms of the problems with decisions like AP1 has... well, we will have to see... this has 3 more cameras than AP1 did (active at least in the EAP option set), along with better ultrasonic sensors, so I suspect that it will be able to do much better at navigating lanes, and I would not be surprised if there were more cases of missed exits, closer calls, etc... but again, the owners will have to learn right along with the car, what works, what doesn't.
• I think that the places that are most likely to cause problems are roads that are ones that some might call highways, others would not. Divided, limited access, interstate roads will probably work really well, but when you get off of those to state highways, some are going to say they are freeways and the system should work perfectly there, and on some of them, it will, on others, especially ones without well marked lanes and exists, there will be problems.

2. Exit freeway and hand nav back to driver:

• You are correct in identifying this as a challenge. I don't know how this will work, nor does anyone else other than Tesla engineers at this point. However, just because you and I don't know how it will work, doesn't mean it is impossible. In the end, I suspect it will start poking at the driver a few miles prior to the exit, and if it does not get some kind of answer (holding the wheel, pressing an "accept" button, whatever), it will decide to do "something". Personally, I would expect it to simply exit as planned and pull over at the next available place to pull over (shoulder, parking lot, whatever. not an ideal situation, but better than just letting the car crash after exiting. It could even work exactly as the Volvo system you mentioned.
• Remember that these cars have ALL of the full self driving hardware on-board, so it is certainly possible for Tesla to decide that in an emergency, the car should utilize everything to safely drive in town until it can find a place to pull over... it could even call 911 at that point if needed.

3. Smart Summon:

• the definition of "smart summon" (from their web site) is "With Smart Summon, your car will navigate more complex environments and parking spaces, maneuvering around objects as necessary to come find you." and from the sales page; "self park when near a parking spot and be summoned to and from your garage."... nothing here talks about parking lots and the "complex environments" could be simply a garage with a curved driveway (see the picture)... my read of this means that I can get out of the car when it is in the lane next to an open spot and walk away and it will park itself (AP1 can basically do that already). and that I can summon it from my garage even it it has to turn to get in / out and get to the door. If the driveway is near a busy street, it will be up to the owner to know its limits (as it is today for many features). I don't take this to mean it will drop me off at the door of the mall and wonder off to find a spot.

4. TLDR:

• I suspect that FSDC is pretty close to being able to do 90% of what it needs to do, they are now working on that last hard 10%. Then their real job starts, trying to convince all of the various rule making bodies around the world that it is safe to be on their streets. What this means is that I think they are already at an advanced stage that will be able to use pieces of it in the EAP (AP2) stuff.
• Having Telsa's map highways helps, but (at least on highways), the cameras that do exist should be able to figure out where exists are and how to handle them...We cant really tell if adding cameras or other sensors is needed for safety until we know more about the systems. Sure, more cameras are good, but are they needed? No way of telling at this point.

Finally, in terms of the cameras, I would not be surprised if they stop talking about the numbers of cameras in each option, and instead just start talking about what each option can do. In the end, if having 4 cameras is all that is needed to drive safely on highways, fine, but I hope that if they determine that 5 cameras are needed (or 8), they will just start using those other cameras (again, they all exist, so its simply a matter of updating the software). I don't see the part of the description anymore that talks about which cameras are live for EAP and which ones are added for FSDC

 

[Electricfan]

If, three months from now, the new Tesla's with the AP2 hardware still have no AP functionality the impact on sales would be devastating. Tesla will never let that happen. It would be the end of Tesla. AP on the new cars will be working before the end of January, my prediction.

 

[Dstrohl]

I am new to Tesla owner ship and trying to catch up.. I bought a S in Sept and am getting a X for my wife next week.

however I own a very nice airplane.. and our GPS system in that is only good to 10 feet o so.. that's called WAAS wide area augmentation system. now its nice when you break out of the cloud and the run way is right there.. but from what I know only the military as dead accurate GPS for navigatation.. but then again I could be behind the times.
Also how do I know if my S is able to accept the up date for version 2 full autonomous drive.. ?

If bought in September, you are out of luck... if you want to be sure, if you have cameras on the side (where the Tesla logo's are right in front of the front doors), then you car has AP2 hardware and can handle full self driving option if you want. if you don't (and since it was announced mid-October... you probably don't), you can only ever do AP1 features.

In terms of aviation comparison, GPS with WAAS is only good (by spec at least) to 7.6m laterally and vertically (though in actual use it is much better)... but when landing a plane, even in full IFR to minimum visibility (say standard GA CAT I landing) you still have 200' of height and are, at worst, 1210' from the runway. lots of time to move 25; to the left or right. Compare that with a normal highway, the speeds are (usually) slower, but you must keep in the middle of a lane that is only a few feet wider than the car, and often are making decisions with much less than a 100' of distance between cars (often MUCH less). My only point here is that they are not really comparable. Aviation uses GPS units that are often more accurate than the ones that are in cars, but driving a car is a MUCH more demanding and more dangerous endeavor. With decision times that are normally shorter, and come up WAY more often, and normally much higher accuracy requirements. (*Not including high performance flying, bush flying, or the occasional student pilot that gives flight instructors heart attacks)

 

[Dstrohl]

If, three months from now, the new Tesla's with the AP2 hardware still have no AP functionality the impact on sales would be devastating. Tesla will never let that happen. It would be the end of Tesla. AP on the new cars will be working before the end of January, my prediction.

I agree that AP2 will be working AT THE LEVEL of AP1 (in terms of features at least). I don't expect it to do everything that it is talking about by Jan (I would be very happy to be wrong here!)

 

[Electricfan]

I agree that AP2 will be working AT THE LEVEL of AP1 (in terms of features at least). I don't expect it to do everything that it is talking about by Jan (I would be very happy to be wrong here!)

How could it not be better than AP1 in its first iteration? It will have more info and better CPU - it will beat AP1 on its first go. Now it may not be perfect on its first go, but it has to be better than what we have now - hands on wheel ready to take over at any moment quality level (some are happy with, I am not).

 

[Dstrohl]

How could it not be better than AP1 in its first iteration? It will have more info and better CPU - it will beat AP1 on its first go. Now it may not be perfect on its first go, but it has to be better than what we have now - hands on wheel ready to take over at any moment quality level (some are happy with, I am not).

I do expect it to work better than AP1 initially, (at least I REALLY HOPE SO!), for all the reasons you mentioned. I was simply saying that I don't expect it to handle things like on-ramp/off-ramp, and smart-summon right away. I expect that in Jan we will see AP2 cars that will have the same feature set as AP1 cars... (lane keeping and auto-park) though hopefully they will work better... then shortly (hopefully) after that, we will start to see the enhancements that will take it beyond AP1.

 

[cirrus282cd]

If bought in September, you are out of luck... if you want to be sure, if you have cameras on the side (where the Tesla logo's are right in front of the front doors), then you car has AP2 hardware and can handle full self driving option if you want. if you don't (and since it was announced mid-October... you probably don't), you can only ever do AP1 features.

In terms of aviation comparison, GPS with WAAS is only good (by spec at least) to 7.6m laterally and vertically (though in actual use it is much better)... but when landing a plane, even in full IFR to minimum visibility (say standard GA CAT I landing) you still have 200' of height and are, at worst, 1210' from the runway. lots of time to move 25; to the left or right. Compare that with a normal highway, the speeds are (usually) slower, but you must keep in the middle of a lane that is only a few feet wider than the car, and often are making decisions with much less than a 100' of distance between cars (often MUCH less). My only point here is that they are not really comparable. Aviation uses GPS units that are often more accurate than the ones that are in cars, but driving a car is a MUCH more demanding and more dangerous endeavor. With decision times that are normally shorter, and come up WAY more often, and normally much higher accuracy requirements. (*Not including high performance flying, bush flying, or the occasional student pilot that gives flight instructors heart attacks)

 

[cirrus282cd]

thank you for your detailed response.. .. that's why I bought the plane with the Chute.. if all else fails. pull early pull often. and I usually only scare the S$$$ out of my self.
I was just confused thinking GPS alone would allow self driving cars.

 

[mikerathbun]

Does Tesla state anywhere that changing lanes is dependent on GPS? I change lanes all the time with my eyeballs and mirrors. I am confident that a car with four or eight active cameras, radar, sonar, and an advanced dedicated computer can perform something that a sixteen year old learning to drive can execute.

Recognizing traffic lights seems to be a solved problem in the official videos that Tesla has been putting out showcasing the new features that are soon to be available to the people with AP2 hardware. Abilities will only get better as they train the neural network more with all the miles they are collecting.

If the OP is correct and EAP or FSD isn't possible or practical with the hardware we have, then that will probably be the end of Tesla and could set back the confidence in the industry. The brightest minds in the field said it would work and many of us purchased the relatively expensive upgrades on the faith of their conclusions. I can't fathom that so many engineers and a tech leader with billions in government contracts are in on a scam to promise something that they know won't work.

 

[Ben W]

Finally, in terms of the cameras, I would not be surprised if they stop talking about the numbers of cameras in each option, and instead just start talking about what each option can do. In the end, if having 4 cameras is all that is needed to drive safely on highways, fine, but I hope that if they determine that 5 cameras are needed (or 8), they will just start using those other cameras (again, they all exist, so its simply a matter of updating the software).

This. It's a tail-wagging-the-dog argument to assume that "4 cameras" is set in stone, particularly if Tesla finds that using the additional cameras would add safety to AP2. In fact, Tesla would be remiss (if not borderline negligent) NOT to use the extra cameras in that case. I presume that all new cars will ship with all eight cameras physically installed and operational, if for no other purpose than to gather driving data for Tesla.

I find it much more likely that AP2 will primarly use the four cameras for most typical tasks, with occasional added input from the additional cameras when dealing with edge cases. And once full autonomy is achieved, it makes far more sense to use all eight cameras exactly the same whether or not "full autonomous" is enabled, and simply restrict the high-level feature set in software. There's no sense having to write (and test and validate and debug) the onramp-to-offramp subroutines twice: once for 4 cameras, once for 8 cameras.

 

[Ben W]

Does Tesla state anywhere that changing lanes is dependent on GPS? I change lanes all the time with my eyeballs and mirrors. I am confident that a car with four or eight active cameras, radar, sonar, and an advanced dedicated computer can perform something that a sixteen year old learning to drive can execute.

GPS does not have nearly the precision or reliability required for lanekeeping. It can be used to give the car a more or less accurate idea where it is on the map, but fine-grained positioning must rely on the cameras only (+ radar + ultrasonic). Driving through cities with skyscrapers, or through tunnels, GPS tends to jump all over the place or cut out entirely, and the car (particularly with full-autonomy) will have to be robust to that.

Bayesian statistics comes into play with these sorts of things. It's a form of probability that takes a prior estimation into account. For instance, if you see on the map (using GPS) that you're approaching an intersection, and you see something out the window that looks sort of like a stop sign, then it's a pretty safe assumption that it really is a stop sign. But if you're driving on a road that has no stop signs mapped, and no other cars have ever reported a stop sign there, and you see something that momentarily looks like a stop sign, then it's quite likely that it's not actually a stop sign. (But it still might be, so drive cautiously and try to confirm.) If the road is completely unmapped, then the probability is somewhere in the middle.

The goal is to reduce both false positives and false negatives as far as possible: you obviously don't want to run a stop sign, but you also don't want to be constantly slamming on your brakes for nothing. If your prior estimation is that there's a 99.999% likelihood of finding a stop sign at a given location, then your behavior will be different from if your prior estimation is 0.001%, even if the camera sees exactly the same thing in both cases.

 

[JRMW]

As others have pointed out, all cars have 8 cameras

Tesla is marketing EAP as having 4 "active cameras" and FSD having 8.

But all cars have all 8. I guarantee all cars actually have all 8 active, we just don't have access to then. But Tesla is using them to collect data..

In the future they will continue to roll out functionality as it becomes validated. If the functionality takes 2, 4 or 8 cameras is immaterial. They'll use what is needed and software limit the rest. They may advertise that EAP only has 4 cameras but that doesn't mean it's true

It's like my 60D. It has a 75 battery
They advertise it as a 60 though. My 60 is software limited but I still have some of the advantages of a 75 (eg slower taper, ability to charge to 100% more safely)

Likewise EAP will not need FULL FSP functionality before EAP can roll out

At worst they could role out an AP1.0 product fairly quickly

 

[busaman]

i agree with OP its not ready yet and wont be for a while.

 

[Bladerskb]

Few points...

- I don't think they need HD maps to determine what lane you're in based on what I've seen about the new system... it can determine your lane without any lane markings whatsoever, so determining the correct lane on a clearly marked highway should be a much easier task

- It's common knowledge that Tesla has been developing HD maps for years already. Combined with the above, I don't think that they'll need years more data to be able to complete functionality

- I don't think having access to all 8 cameras with FSD would change any of the points you've brought up, so regarding this functionality being unavailable before FSD rolls out I think doesn't really make sense

- Off-ramp transition to driver could literally be as simple as a dash countdown from the time the car is in the correct lane to the time it takes an exit (10 seconds?) in which the car transitions control back to the driver instantly once the exit has been taken. Problem solved, no additional cameras or sensors needed and promise fulfilled. Pretty simple.​

I like the way you're thinking, but I'm pretty sure the flagship functionality will definitely be rolled out in a reasonable time frame. It would be silly for them to promise the three or four main features on the Tesla website then decide it was too hard to do after people bought cars. I understand Elon made promises about AP1 he didn't fulfill, but those were fringe use cases not flagship functionality promises.

HD map is for determining your position/lane in relation to the world. And yes I agree that Tesla have billions of miles of AP data and hence have alot of HD map data already.

I'm not discounting that. I'm just stating the necccesity of it.

But don't you think 10 seconds before the system disengages is dangerous? Especially by which time other competitors will have systems for full L4 highway automation. And this is why I posted the Volvo video. You would think Tesla will take that route. Why not have a L4 highway autonomy like mode? A mixed system still leads to an even more dangerous AP? Why t

Ok, let me try to address some of these points without name calling...

1. Automatic Lane Changes:

• You are correct that the car would need to know what lane it is in, and what lane to be in... and you are correct in the inaccuracies of GPS. In general, GPS can get to as low as 0.01m in accuracy with post processing and correcting technologies, but without those, the best that is theoretically possible is about 2m, and 4m is more generally accepted as the accuracy goal. Even with the 2m best case, that is a lateral accuracy of 6.5 feel, or clear in another lane. However, the cameras should be able to tell if you are in a middle lane, or a left edge/right edge one... and therefore, they should be able to decide to move to the left or right as needed for the exit.
• The information about if an exit is a left or right exit, and how many lanes there are are available is normally in the GPS database, as well as advertised with lane markers and street signs, if the roads follow the NHTSA rules *(in the US) this should work for the majority of cases. (limiting this to talking about highways now). So, if the car knows that an exit is coming up on the left, it can start moving left at least x minutes early, with x defined based on traffic.
• In terms of the problems with decisions like AP1 has... well, we will have to see... this has 3 more cameras than AP1 did (active at least in the EAP option set), along with better ultrasonic sensors, so I suspect that it will be able to do much better at navigating lanes, and I would not be surprised if there were more cases of missed exits, closer calls, etc... but again, the owners will have to learn right along with the car, what works, what doesn't.
• I think that the places that are most likely to cause problems are roads that are ones that some might call highways, others would not. Divided, limited access, interstate roads will probably work really well, but when you get off of those to state highways, some are going to say they are freeways and the system should work perfectly there, and on some of them, it will, on others, especially ones without well marked lanes and exists, there will be problems.

2. Exit freeway and hand nav back to driver:

• You are correct in identifying this as a challenge. I don't know how this will work, nor does anyone else other than Tesla engineers at this point. However, just because you and I don't know how it will work, doesn't mean it is impossible. In the end, I suspect it will start poking at the driver a few miles prior to the exit, and if it does not get some kind of answer (holding the wheel, pressing an "accept" button, whatever), it will decide to do "something". Personally, I would expect it to simply exit as planned and pull over at the next available place to pull over (shoulder, parking lot, whatever. not an ideal situation, but better than just letting the car crash after exiting. It could even work exactly as the Volvo system you mentioned.
• Remember that these cars have ALL of the full self driving hardware on-board, so it is certainly possible for Tesla to decide that in an emergency, the car should utilize everything to safely drive in town until it can find a place to pull over... it could even call 911 at that point if needed.

3. Smart Summon:

• the definition of "smart summon" (from their web site) is "With Smart Summon, your car will navigate more complex environments and parking spaces, maneuvering around objects as necessary to come find you." and from the sales page; "self park when near a parking spot and be summoned to and from your garage."... nothing here talks about parking lots and the "complex environments" could be simply a garage with a curved driveway (see the picture)... my read of this means that I can get out of the car when it is in the lane next to an open spot and walk away and it will park itself (AP1 can basically do that already). and that I can summon it from my garage even it it has to turn to get in / out and get to the door. If the driveway is near a busy street, it will be up to the owner to know its limits (as it is today for many features). I don't take this to mean it will drop me off at the door of the mall and wonder off to find a spot.

4. TLDR:

• I suspect that FSDC is pretty close to being able to do 90% of what it needs to do, they are now working on that last hard 10%. Then their real job starts, trying to convince all of the various rule making bodies around the world that it is safe to be on their streets. What this means is that I think they are already at an advanced stage that will be able to use pieces of it in the EAP (AP2) stuff.
• Having Telsa's map highways helps, but (at least on highways), the cameras that do exist should be able to figure out where exists are and how to handle them...We cant really tell if adding cameras or other sensors is needed for safety until we know more about the systems. Sure, more cameras are good, but are they needed? No way of telling at this point.

Finally, in terms of the cameras, I would not be surprised if they stop talking about the numbers of cameras in each option, and instead just start talking about what each option can do. In the end, if having 4 cameras is all that is needed to drive safely on highways, fine, but I hope that if they determine that 5 cameras are needed (or 8), they will just start using those other cameras (again, they all exist, so its simply a matter of updating the software). I don't see the part of the description anymore that talks about which cameras are live for EAP and which ones are added for FSDC

Thanks for your detailed feedback.
I believe there are two teams for AP. One team working on EAP and training the NN using 4 cameras and another team working on FSD and training the NN using 8 cameras.

So it won't be easy to abruptly add a camera. And the only camera they can add without throwing off the NN for EAP is the forward main camera. But they could be lying and already have it on.

However, I like your freeway exit transition scenario. It fits better with the rest of the industry. But that requires a level of FSD to be complete and be turned on during exits.

If tesla put a timer 10 seconds before exit as someone suggested we will simply get the same type of accidents as AP1. And isn't AP2 purpose to eliminate those quirks? It will introduce more dangerous quirks. I think I speak for everyone when I say. We don't want more parlor trick that we have to heavily supervise. We can something relatively safe.

If Tesla pulls a 10 seconds before exit and disengages. What could they pull under FSD?

I think this is the time to figure out how to make a 100% safe system in a particular scenario like an official highway.

Or there will be even more accidents and ppl being blamed for not being vigilant.

Why not also have a L4 highway autonomy mode that acouple other company will be releasing 2017 onwards?

EAP is basically still L2 autonomy.

 

[182RG]

How could it not be better than AP1 in its first iteration?

It would be easy for it not to be better. They are doing it without Mobileye, and it's clear that the Mobileye divorce happened sooner than desired and accelerated the timeline. They are behind the curve, using new hardware, hence the smokescreen of "fleet learning". They'll slap "beta" on it, and the usual apologists will be here to explain it away.

 

[Bladerskb]

How could it not be better than AP1 in its first iteration? It will have more info and better CPU - it will beat AP1 on its first go. Now it may not be perfect on its first go, but it has to be better than what we have now - hands on wheel ready to take over at any moment quality level (some are happy with, I am not).

I hate to break it to you but hands on ready to take over will still exist and actually more than ever. You will need to be even more vigilant. Why? Because EAP is actually still L2 automation because there doesn't seem to be any safeguards, plus there are blindspots based on the 4 camera configurations. One in the left and right front and one at the back.

Think about those freeways where u come from the on ramp and immediately you have to get over to the left side, 3-4 lanes away so you can catch your exit or transition to another freeway. Now your car has to negotiate with wreckless speeding human drivers to get over three lanes.

Or even negotiating to enter the freeway from an on ramp so you don't get hit by the oncoming exit on the lane you are in.

Let's just say your hands will be alot more sweaty than simply driving straight. Again the car has to do with with numerous blindspots and relying on its ultrasonic. That's just unsafe.
I see the accident rate increasing from AP1 not decreasing unless tesla focus on making it safe

 

[Vger]

To me, it is comical to see all this negativity now about HW2 and EAP, when the doubters (myself included) were so wrong about AP1. No it will not be perfect, especially on day 1 (and day 1 is very soon), but I fully expect that each feature enabled on HW2 will exceed the performance of it equivalent on MobileEye..