Full Self-Driving - feels like a long way off to me...

[Theflash95]

When it comes to freeway driving many many (and maybe you) use it for 95% of the time without taking it out of AP Mode. It is that way for me. I could go hands free except for the nags and my own personal wanting to be extra safe. But I believe EAP is primary a freeway driver assist (or full assist 95% of the time) and we need much more for surface street driving and that is where I believe FSD comes in. Remember EAP means On-Ramp to Off-Ramp without driver input which Elon actually said he used the night before to get home when talking about v9 in August. Elon seems to believe EAP for that purpose is close.

EAP isn't flawless on the highway... I've had phantom braking events and they haven't all been consistent. Also... if traffic is at a standstill up ahead EAP isn't going to work and bring me to a nice slow stop. It rained like crazy the other day (I know... it's an "edge event") and I couldn't see the lines in the road nor could EAP (they weren't shown on the IC).

In my opinion... FSD will be building upon what EAP is capable of now. I'm sure Tesla is internally testing a version that is light years ahead of what we've got now, and maybe that can do some other stuff pretty flawlessly... but I'm guessing it's only in ideal conditions and isn't 99% safe... otherwise we'd have seen it already.

Tesla sure has been improving EAP all along and I look forward to continued improvements. If I don't see FSD anytime soon it's no biggie to me... I've always expected it to be at least 10 - 15 years down the road. If it gets here sooner I will be pleasantly surprised that my personal estimate was way off. I do agree with others on here who have said Tesla should have backed off with all the FSD talk... they've got a great car already... EAP & FSD is just icing on the cake... if you promise the icing within a few years and don't deliver it's a let down... if you say you're working towards the FSD goal with no definitive ETA (or a more broad one) and then slowly roll out features as they're tested and reliable that's a better way to go in my opinion.

 

[Electroman]

does ur city or freeway exits have dotted lines across? Not all cities have this and this is where problems start - there's no standardization to street lines.

No the ones I am referring to do not have any lines. This one is curved with no lines. It centers perfectly before and after the Xn

 

[cybergates]

No the ones I am referring to do not have any lines. This one is curved with no lines. It centers perfectly before and after the Xn

View attachment 322004

90% of the time I can go thru an intersection like that with no issue so it's better than it used to be. But tehre's two where the car unexpectedly jerks into the other lane across the intersection - that's scary and there's no rhyrme or reason. I've tried reporting these using the feedback so hopefully they get fixed! They have to be 100.00% accurate in this.

 

[diplomat33]

Could you clarify? Seems like you and @diplomat33 believe that EAP will have some changes going forward but there will be pretty much New Software loaded on cars that have purchased FSD that will replace (not added too) the current software being used for EAP? Like the freeway driving on AP will have EAP Firmware for the cars without FSD and FSD Firmware for the cars that have FSD and the software is not the same code base for AP? As I said I personally do not think this will be the case but of course I could be wrong.

I think you represented my view correctly. While EAP and FSD will most likely share some common code, I do think FSD will be mostly new code. Think about it. FSD will need to do so much more than EAP. If you think about it, EAP basically just does 2 things: it uses the front radar to track the speed and distance of the car in front and modulates the accelerator appropriately in order to maintain a given distance and it uses the front camera to recognize lane markings and control the steering appropriately to maintain the car in the center of the lane. So FSD will certain use some of that too of course but FSD will need to be mostly camera vision based because it will need to see all cars and objects around the car, track them, recognize speed limit signs and stop signs, recognize traffic lights, turn at intersections etc... So based on what FSD will need to do, I think it makes sense that it will use a ton of new code that is not shared with EAP. Which is the main reason I don't think we can use the current state of EAP to gauge the future performance of FSD.

Moreover, it is possible that FSD will actually replace how it does things compared to EAP. For example, right now, TACC uses the front radar to track the car in front but FSD will probably use the front camera instead. By using the cameras, FSD will see a stationary vehicle on the side of the road and be able to avoid it, something that TACC struggles with because it only uses the radar. So there may be areas like this where we cannot use EAP as a measure for how FSD will do it since they use different methods entirely.

 

[Theflash95]

Moreover, it is possible that FSD will actually replace how it does things compared to EAP. For example, right now, TACC uses the front radar to track the car in front but FSD will probably use the front camera instead. By using the cameras, FSD will see a stationary vehicle on the side of the road and be able to avoid it, something that TACC struggles with because it only uses the radar. So there may be areas like this where we cannot use EAP as a measure for how FSD will do it since they use different methods entirely.

I thought EAP actually did use a front camera or two, no? Regardless someone should ask Elon about this... are FSD features a bolt on / extension to what EAP already does or is it mostly new code? My opinion is that the FSD features that are coming and eventually full FSD itself will use what they've got so far for EAP and build on it. My guess is that if someone asked Elon this and that was his answer a lot of people, frustrated already with the progress of EAP, would be pretty upset. If it's something mostly new that when they start rolling it out is amazing I'll be pleasantly surprised.

 

[emmz0r]

I think it is a mistake to use EAP to judge what FSD will or won't be able to do. EAP is not the FSD software. EAP is just the driver assist package that Tesla created to duplicate the Autopilot software that MobilEye had made. It is not designed to do any self-driving stuff. We should wait until Tesla does release the FSD features and then judge if they are capable of doing FSD or not.

It makes no sense that EAP is a completely different software path than FSD. It would be nuts not to have a EAP as a part/sub functionality of FSD.
Also it would be nuts to release EAP in this state if they secretly have a much better version in the FSD source tree. Also it would be unnecessary terrible PR since EAP is supposed to be a showcase for what FSD will do in the future.

So no - I hold that EAP is the actual state of the system and they don't have a vastly better version secretly in development, then it would be another rewrite in that case. Elon said some FSD features, which I guess would mean intersections using traffic lights etc - on top of EAP.

 

[brendon1970]

I tend to agree from s technical point of view. However what if in the near term the EAP only cars are NOT getting the good stuff in order to encourage upgrades?

 

[bob_p]

The current EAP software is likely just a continuation of the software developed for AP 1 hardware, adapted to the new AP 2.x hardware.

As Tesla gets more of the FSD capabilities working (using machine learning), it's likely EAP will be shifted onto the FSD code base, imposing the "hands on wheel" restriction.

It's possible FSD approval may not be an "all or nothing" situation, and that vehicles might be approved for FSD operation in certain conditions, while require operating in driver assist mode everywhere else. High occupancy lanes would be an obvious first place to approve use of FSD - and then followed by approval for use on (most) limited access highways.

One of the posts above mentioned an interesting side effect of FSD - obeying the traffic laws. It's very possible vehicles operating under FSD will be required to strictly obey all traffic laws. Which means driving at or below the posted speed limit, coming to a complete stop at intersections, parking only in valid parking spaces, … While FSD could allow safer driving at higher speeds with denser traffic, in the short term, when FSD vehicles are sharing the road with non-FSD vehicles, we could see traffic actually slow down.

Besides the technical obstacles for FSD, another major hurdle is liability. Who is responsible when an FSD breaks the law or is in an accident. Is it the owner of the vehicle? Or is it the driver, who is being supplied by the manufacturers? While the manufacturers will want to push liability onto the owners - when there isn't anyone even inside the vehicle and it's driving itself, how can an owner be held responsible for what will likely be viewed as flaws in the FSD implementation, which should be the responsibility of the manufacturers.

That said, we're still satisfied with the FSD purchase we've made with our S and new X - because even if FSD isn't approved for unattended use, the software should start handling more driving situations better, which could start with the Version 9 release.

 

[diplomat33]

I thought EAP actually did use a front camera or two, no? Regardless someone should ask Elon about this... are FSD features a bolt on / extension to what EAP already does or is it mostly new code? My opinion is that the FSD features that are coming and eventually full FSD itself will use what they've got so far for EAP and build on it. My guess is that if someone asked Elon this and that was his answer a lot of people, frustrated already with the progress of EAP, would be pretty upset. If it's something mostly new that when they start rolling it out is amazing I'll be pleasantly surprised.

It makes no sense that EAP is a completely different software path than FSD. It would be nuts not to have a EAP as a part/sub functionality of FSD.
Also it would be nuts to release EAP in this state if they secretly have a much better version in the FSD source tree. Also it would be unnecessary terrible PR since EAP is supposed to be a showcase for what FSD will do in the future.

So no - I hold that EAP is the actual state of the system and they don't have a vastly better version secretly in development, then it would be another rewrite in that case. Elon said some FSD features, which I guess would mean intersections using traffic lights etc - on top of EAP.

Ok, here is an analogy for EAP v FSD that might help. EAP is like a runner and FSD is like a soccer player. The skeptics are saying "all I've seen the person ever do is run. And they can run ok on most roads but they still have some difficulty running on other roads. Based on this, I don't see how this runner is ever going to be a soccer player, let alone a good one, if all I have ever seen them do is run and be an average runner at that." I am saying "running and soccer are very different. You can't look at someone running a 5K and use that to judge them on how they play soccer. They are not trying to play soccer. Since we have not seen the runner try to play soccer, it is not fair to try to judge them on that." EAP is not trying to be self-driving so how can we judge it on that?

To continue my analogy, running is a part of soccer but just a part. A good soccer player needs to do a lot more than just run. a good soccer player also needs to be good at dribbling, passing and shooting. They also need to know the rules of the game as well as the tactics and strategies of the game. You would probably teach someone soccer from the ground up. You wouldn't just take a 5K runner and try to add some extra skills like dribbling, passing and shooting and try to make them into a soccer player! Likewise, I would guess that FSD might contain some EAP code but will be mostly be new code as FSD will need to do a lot more than what EAP does.

 

[emmz0r]

I strongly doubt it's like that. To follow your analogy I think EAP is a soccer player that only knows how to run and avoid other players.
EAP is a subset of code in FSD, not an entirely separate thing. That would be incredibly ineffective and useless from a engineering / coding perspective (yes I am one, and yes I have worked on computer vision)

 

[diplomat33]

I strongly doubt it's like that. To follow your analogy I think EAP is a soccer player that only knows how to run and avoid other players.
EAP is a subset of code in FSD, not an entirely separate thing. That would be incredibly ineffective and useless from a engineering / coding perspective (yes I am one, and yes I have worked on computer vision)

Well, I do think my analogy is a good way of describing the different points of view in this thread. You see EAP as a soccer player who only knows how to be a runner whereas I see EAP as a runner who has not been taught how to play soccer yet. But it does explain your skepticism. If I saw a soccer player who only knew how to run and avoid other players and moreover, had not learned any new soccer skills in the past 2 years, I would probably doubt that the player could ever be a good soccer player too.

 

[Theflash95]

I've got to agree with emmz0r, to me EAP is the stepping stone that FSD features are going to be built on. In order to become that great soccer player you need to develop your motor skills and endurance before learning more advanced skills like ball control, defense and kicking in goals. Why work on an EAP system that does lane keeping, TACC and a few other features and then have a separate team not using any of that and doing their own thing to have a vehicle that does lane keeping, TACC AND stopping at traffic lights, stop signs, completing turns through unmarked intersections, etc.?

Tesla may have different teams working on individual situations (making a left turn on a divided road with no lane guides, slowing for a yellow light and stopping at a red, waiting for a green light before proceeding, stopping at a stop sign, etc.), but I've got to think it's all being built upon what they've got so far.

Again, maybe I'm wrong and before you know it the folks who paid for FSD will get an update that makes EAP look like my 2 year old trying to drive the Model S for kids (I know it says ages 3 to 8, but he's a big kid and can reach the go pedal... just need to work on his "lane keeping" skills)... we'll just have to wait and see.

 

[diplomat33]

I've got to agree with emmz0r, to me EAP is the stepping stone that FSD features are going to be built on. In order to become that great soccer player you need to develop your motor skills and endurance before learning more advanced skills like ball control, defense and kicking in goals. Why work on an EAP system that does lane keeping, TACC and a few other features and then have a separate team not using any of that and doing their own thing to have a vehicle that does lane keeping, TACC AND stopping at traffic lights, stop signs, completing turns through unmarked intersections, etc.?

Tesla may have different teams working on individual situations (making a left turn on a divided road with no lane guides, slowing for a yellow light and stopping at a red, waiting for a green light before proceeding, stopping at a stop sign, etc.), but I've got to think it's all being built upon what they've got so far.

Again, maybe I'm wrong and before you know it the folks who paid for FSD will get an update that makes EAP look like my 2 year old trying to drive the Model S for kids (I know it says ages 3 to 8, but he's a big kid and can reach the go pedal... just need to work on his "lane keeping" skills)... we'll just have to wait and see.

EAP and FSD appear to have very different design goals. At least, the way I read the descriptions on the page, EAP is a driver assist. It will keep you in the lane and keep you at the right speed, and do some other cool things like off ramp exits in the future, but basically still a driver assist. On the other hand, FSD is designed to be full self-driving where you can get in and the car can take you from A to B without any driver intervention. So very different goals. That is why I see the two as separate. Now, maybe I am wrong and EAP will be "FSD light" where it will be the same as FSD, just stripped down of certain features. But since we don't have FSD yet, it is hard to make that determination right now.

 

[chillaban]

I strongly doubt it's like that. To follow your analogy I think EAP is a soccer player that only knows how to run and avoid other players.
EAP is a subset of code in FSD, not an entirely separate thing. That would be incredibly ineffective and useless from a engineering / coding perspective (yes I am one, and yes I have worked on computer vision)

I think EAP is more like putting a runner on a soccer field and discussing how well he is playing soccer.

FSD is supposed to understand the big picture and the rules of the road while EAP is more super-glorified adaptive cruise control + lane keeping.

 

[Kant.Ing]

FSD if Level 5 should make its human passengers (including the human behind any wheel) like a sea captain. However, current (E)AP uses the human driver like a lowest ranking seaman on deck.

 

[RDoc]

I'm a full self driving skeptic but totally disagree with you. AGI is a quantum leap past full self driving. Don't confuse the two. FSD requires ANI, not AGI.

That's a pretty bald statement, I'd like to see your definitive proof of why that's such an obvious fact.

Depending on your definition of ANI and AGI, I'll stick to my conclusion. If FSD were in the ANI category, we could train animals like pigeons to at least drive passably, but we can't. I'm very doubtful it's possible to teach a primate monkey to drive in all but the simplest situations.

For real FSD in a complex situation like a city with bicycles, cars, lights, pedestrians, running children, etc. the driver has to rely on a very wide knowledge of expected behaviors, potential scenarios, etc. as well as straightforward object recognition. I'm unaware of any AI system that can do anything remotely like that, or even has the potential to.

AGI doesn't have to be Skynet to be very different from ANI.

 

[d21mike]

EAP and FSD appear to have very different design goals. At least, the way I read the descriptions on the page, EAP is a driver assist. It will keep you in the lane and keep you at the right speed, and do some other cool things like off ramp exits in the future, but basically still a driver assist. On the other hand, FSD is designed to be full self-driving where you can get in and the car can take you from A to B without any driver intervention. So very different goals. That is why I see the two as separate. Now, maybe I am wrong and EAP will be "FSD light" where it will be the same as FSD, just stripped down of certain features. But since we don't have FSD yet, it is hard to make that determination right now.

Not sure you read all of the original pages about EAP. The part in BLUE is what I expect both EAP and FSD to do (the same) while on Freeways. Surface streets is much different and EAP will not do much more then it does now. FSD will do so much more on surface streets then EAP but I assume that will be simply a switch in the code. Seems like what you are saying is the code base for freeway driving doing all that is in blue below will be completely separate for EAP and FSD. To me that makes zero sense. To do all below in blue will require a lot more then EAP is capable of doing now.

Enhanced Autopilot
Enhanced Autopilot adds these new capabilities to the Tesla Autopilot driving experience. Your Tesla will match speed to traffic conditions, keep within a lane, automatically change lanes without requiring driver input, transition from one freeway to another, exit the freeway when your destination is near, self-park when near a parking spot and be summoned to and from your garage.

Tesla’s Enhanced Autopilot software has begun rolling out and features will continue to be introduced as validation is completed, subject to regulatory approval. Every driver is responsible for remaining alert and active when using Autopilot, and must be prepared to take action at any time.

 

[diplomat33]

Not sure you read all of the original pages about EAP. The part in BLUE is what I expect both EAP and FSD to do (the same) while on Freeways. Surface streets is much different and EAP will not do much more then it does now. FSD will do so much more on surface streets then EAP but I assume that will be simply a switch in the code. Seems like what you are saying is the code base for freeway driving doing all that is in blue below will be completely separate for EAP and FSD. To me that makes zero sense. To do all below in blue will require a lot more then EAP is capable of doing now.

Enhanced Autopilot
Enhanced Autopilot adds these new capabilities to the Tesla Autopilot driving experience. Your Tesla will match speed to traffic conditions, keep within a lane, automatically change lanes without requiring driver input, transition from one freeway to another, exit the freeway when your destination is near, self-park when near a parking spot and be summoned to and from your garage.

Tesla’s Enhanced Autopilot software has begun rolling out and features will continue to be introduced as validation is completed, subject to regulatory approval. Every driver is responsible for remaining alert and active when using Autopilot, and must be prepared to take action at any time.

I want to make sure I understand you. So, you think that EAP will be "FSD light", basically a software limited FSD. In other words, Tesla will use the same FSD code for both, just software limit some of the features for cars with EAP only? I do agree with you that could very well happen. Yes, it would definitely make sense for Tesla to do that. I just don't see it now with the current EAP. I know Karpathy rewrote the neural net that improved auto steer but other than that, I don't really see EAP as using the FSD software now.

 

[d21mike]

I want to make sure I understand you. So, you think that EAP will be "FSD light", basically a software limited FSD. In other words, Tesla will use the same FSD code for both, just software limit some of the features for cars with EAP only? I definitely think that could happen, yes. So I do agree with you that could very well happen. Yes, it would definitely make sense for Tesla to do that. I just don't see it now with the current EAP. I know Karpathy rewrote the neural net that improved auto steer but other than that, I don't really see EAP as using the FSD software now.

Yes. Exactly. I believe to fully do what EAP is documented to do it will be FSD on Freeways. On Freeways there will be no difference between EAP and FSD but on Surface Streets there will be a major difference. Assuming EAP continues to work on surface streets I am not seeing much change from today. However, for safety reasons it might be nice for it to at least STOP for stop signs and street lights. But it will not follow GPS/Navigation the way it will on Freeways. So, for those that do not need FSD on surface streets they could save some money and just get EAP.

 

[sillydriver]

Enhanced Autopilot
Enhanced Autopilot adds these new capabilities to the Tesla Autopilot driving experience. Your Tesla will match speed to traffic conditions, keep within a lane, automatically change lanes without requiring driver input, transition from one freeway to another, exit the freeway when your destination is near, self-park when near a parking spot and be summoned to and from your garage.

Yes. Exactly. I believe to fully do what EAP is documented to do it will be FSD on Freeways. On Freeways there will be no difference between EAP and FSD but on Surface Streets there will be a major difference. Assuming EAP continues to work on surface streets I am not seeing much change from today. However, for safety reasons it might be nice for it to at least STOP for stop signs and street lights. But it will not follow GPS/Navigation the way it will on Freeways. So, for those that do not need FSD on surface streets they could save some money and just get EAP.

I agree with @d21mike. It's one thing for Tesla to be very late delivering features, but imagine what would happen if they did begin to deliver the features in their own description of EAP, above, but labeled them as part of FSD and said that people with EAP had to pay the additional thousands of bucks for FSD to get them. The resulting legal dispute would be open-and-shut against the company, not to mention the customer uproar. No, FSD has to be something different than what Tesla has committed for EAP.