Elon: "Feature complete for full self driving this year"

[CarlK]

Yeah. That's why the Lidar cars are out driving themselves around today while Teslas still are trying to figure out whether the tree limb is in the middle of the road or fifteen feet above it. Put down your Kool-Aid cups, Tesla fans.

Lidar cars are running on designated areas in certain cities and will continue to only able to do that for a long time if not forever.

Tesla cars can be used just like any regular cars anywhere in the world and do things it can do. It will have challeges and accidents just like we all do but the machine is set to improve over time and we are not.

 

[Bladerskb]

Lidar cars are running on designated areas in certain cities and will continue to only able to do that for a long time if not forever.

Tesla cars can be used just like any regular cars anywhere in the world and do things it can do. It will have challeges and accidents just like we all do but the machine is set to improve over time and we are not.

Are other ADAS cars like Tesla, for example supercruise, also better than these lidar cars because they are not running on small designated areas in certain cities?

 

[CarlK]

Are other ADAS cars like Tesla, for example supercruise, also better than these lidar cars because they are not running on small designated areas in certain cities?

ADAS and FSD are two very different things and have little relationships to each other. GM's supercruise is independent of its autonomous effort. Tesla sort of integrated the two in the beginning since it fits its stratage the best but the separation is starting to show now. Even when Tesla was running AP1 it was planning the FSD strategy like NN machine learning from everyone of its cars (that Lidar cars can not do) and designing the AI chip that is necessary for FSD but not ADAS.

 

[rnortman]

Can we at least agree that the AP3 computer is an improvement over the current AP2 computer and that it will give Tesla more computing power to work with?

Yeah no doubt, but temper your enthusiasm with the reality that HW3 is probably going to have a lot of bugs and instability for many months, like maybe until the end of the year before it becomes stable. HW2.5 had its own special bugs for ~6 months and that was a small change. HW2, well, it took like 2 years just to catch up to AP1.

But sure, eventually it will get better. What I'm really interested to see is exactly who gets the upgrade. Part of me is hoping that EAP purchasers will get the upgrade, because frankly it looks like they're having trouble delivering EAP on HW2.x, and the new FSD is pretty much the old EAP. But probably they just won't ever really deliver solid EAP, and we will have to either live with more or less what we currently have or fork over the money for "FSD" aka "FSD is the new EAP" if we want it to actually work well.

 

[electronblue]

But probably they just won't ever really deliver solid EAP, and we will have to either live with more or less what we currently have or fork over the money for "FSD" aka "FSD is the new EAP" if we want it to actually work well.

Ah! Good point. FSD is the new EAP. That limits retrofit requirements to FSD buyers only while still allowing to sort of retire EAP at whatever level they want to leave it at eventually and call everything else FSD...

 

[rnortman]

Ah! Good point. FSD is the new EAP. That limits retrofit requirements to FSD buyers only while still allowing to sort of retire EAP at whatever level they want to leave it at eventually and call everything else FSD...

Yeah, I think that's exactly what they're thinking.

 

[Bladerskb]

ADAS and FSD are two very different things and have little relationships to each other.

Not according to Tesla. They define their FSD as Level 2 ADAS.

GM's supercruise is independent of its autonomous effort.

But according to you a non geofenced ADAS like current AP is better than all those cars running around with Lidar

Tesla sort of integrated the two in the beginning since it fits its stratage the best but the separation is starting to show now.

There is no separation

Even when Tesla was running AP1 it was planning the FSD strategy like NN machine learning from everyone of its cars (that Lidar cars can not do)

Tesla isn't machine learning from everyone of its cars though.

and designing the AI chip that is necessary for FSD but not ADAS.

Them designing their own chip is meanless as others have options for chips like Xavier, EyeQ4 and EyeQ5.

 

[rnortman]

Frankly I really have no interest in hearing those amateurish opinions. You certainly have the right to say Tesla is not moving fast enough but we should put things in their right perspectives to not look stupid.

Yeah, nobody wants to hear stupid, amateurish opinions.

it was planning the FSD strategy like NN machine learning from everyone of its cars (that Lidar cars can not do)

Ahem.

 

[Engr]

Yeah. That's why the Lidar cars are out driving themselves around today while Teslas still are trying to figure out whether the tree limb is in the middle of the road or fifteen feet above it. Put down your Kool-Aid cups, Tesla fans.

If you believe there are lidar cars out there driving themselves without similar problems, then I've got a strong Brooklyn bridge to sell you.

 

[CarlK]

Not according to Tesla. They define their FSD as Level 2 ADAS.

I believe you are confusing FSD to AP.

But according to you a non geofenced ADAS like current AP is better than all those cars running around with Lidar

There can't be such comparison. There is no Lidar ADAS car. All Lidar cars are designed to be autonomous.

There is no separation

Not true. The most significant one is even that Tesla will put the new HW3/AI chip in every new car it will not retrofit any existing cars with it if it did not purchase the FSD option. AP does not require the new HW.

Tesla isn't machine learning from everyone of its cars though

.

No one knows how Tesla's system works but every car is designed in the way that it can contribute to the community machine learning.

Them designing their own chip is meanless as others have options for chips like Xavier, EyeQ4 and EyeQ5.

No it's extremely meaningful even that you can't see it (what else is new?). Aside from the fact that Tesla chip is the most advanced one (real not announced) on the market the chip is mission specific tailored specifically for Tesla SW requirements. Others have to be Jack of all trades and master of none because they have to sell to different customers to run very different SW. That's the power of vertical integration and why Elon wanted to do everything this way.

 

[diplomat33]

Yeah no doubt, but temper your enthusiasm with the reality that HW3 is probably going to have a lot of bugs and instability for many months, like maybe until the end of the year before it becomes stable. HW2.5 had its own special bugs for ~6 months and that was a small change. HW2, well, it took like 2 years just to catch up to AP1.

Fair point. Yes, I am sure AP3 will have bugs at first. Anytime, you write new software for new hardware, or even port existing software to new hardware, there will be bugs and problems. I am not expecting AP3 to be plugged in and instantly be bug free right out of the gate.

But sure, eventually it will get better. What I'm really interested to see is exactly who gets the upgrade. Part of me is hoping that EAP purchasers will get the upgrade, because frankly it looks like they're having trouble delivering EAP on HW2.x, and the new FSD is pretty much the old EAP. But probably they just won't ever really deliver solid EAP, and we will have to either live with more or less what we currently have or fork over the money for "FSD" aka "FSD is the new EAP" if we want it to actually work well.

I am afraid this is where I will have to disagree.

First, sure, it would be great if EAP owners also got the AP3 upgrade but please don't count on it. Tesla has not made that promise. In fact, Tesla has given several indications that they will not do that. They are calling AP3, the "full self-driving computer" hinting that it is reserved for FSD. The promise explicitly states that FSD owners will get the upgrade with no mention of EAP owners. And Musk has repeatedly stated that EAP does not need AP3. Those are 3 reasons why EAP owners will most likely never get AP3. So I would caution you not to get your hopes up for something that Tesla has not promised and in fact has strongly hinted that they won't do. That's a sure fire way to be disappointed.

To your second point, my personal experience with EAP is very solid now. I am very happy with EAP now. And once we get Enhanced Summon and the removal of the "hold the wheel" nag during auto lane change, EAP will be feature complete. Even without those additional features, Tesla has given us a really good EAP now. I am sure EAP will get additional improvements but we are getting close to the end for EAP. But it sounds like EAP has not met your expectations which is why you are hoping for AP3 and hoping that FSD gives you the EAP that you want.

To your third point that "FSD is the new EAP. Yes, right now, FSD might look like the old EAP but Tesla is promising future FSD features like "automatic city driving" that were never EAP features. So, clearly, Tesla intends FSD to eventually become more than EAP. Tesla intends EAP to basically be what it is now (plus Enhanced Summon) and then FSD to take things to the next level.

 

[OPRCE]

And once we get Enhanced Summon and the removal of the "hold the wheel" nag during auto lane change, EAP will be feature complete.

One major feature still missing there is the ability to *not* collide at high speed with parked fire-trucks, etc., which may be one reason EAP cars will ultimately need HW3 in order to achieve acceptable safety, but then the same argument would apply to basic AP also, so all may need HW3.

 

[rnortman]

To your second point, my personal experience with EAP is very solid now. I am very happy with EAP now.

And as I believe you know, my experience with NOA is quite the opposite. I think TACC is (mostly) great, Autosteer is OK, and NOA in my experience is a gimmick at the best of times, occasionally outright dangerous, and always 100% useless. It is very interesting that people have such vastly differing experiences with this feature.

 

[verygreen]

But the software that @verygreen did his analysis on could have been made to run on devkits.

no. it's production setup.

 

[CarlK]

it was planning the FSD strategy like NN machine learning from everyone of its cars (that Lidar cars can not do) .

Ahem.

Looks you took my comment out of context. You can't equip all cars, or even some of cars, you sell with Lidar. Basically I believe the reason Tesla did not choose the Lidar route.

 

[rnortman]

Looks you took my comment out of context. You can't equip all cars, or even some of cars, you sell with Lidar. Basically I believe the reason Tesla did not choose the Lidar route.

That is absolutely why Tesla chose not to use lidar, but what does this have to do with machine learning? You think the guys doing lidar aren't doing deep learning? Using lidar doesn't prevent you from also doing everything Tesla is doing... and more.

 

[CarlK]

That is absolutely why Tesla chose not to use lidar, but what does this have to do with machine learning? You think the guys doing lidar aren't doing deep learning? Using lidar doesn't prevent you from also doing everything Tesla is doing... and more.

Tesla's is crowd sourced machine learning. Lidar can only be put in a few test cars. Machines will learn much faster with Tesla's approach because much more and faster machine learning miles can be acquired. That's what every expert in the field, even the one who was in charge of ML and was let go by Tesla, says.

 

[electronblue]

Tesla's is crowd sourced machine learning.

No, it is not. Tesla consumer cars do not do any learning. They receive NNs trained on a few test cars and simulators just like everyone else.

 

[ZeApelido]

I think he means they are crowdsourcing the data for machine learning.

 

[electronblue]

I think he means they are crowdsourcing the data for machine learning.

There really isn’t any proof of that either beyond some limited trigger collection.