Firmware 9 in August will start rolling out full self-driving features!!!

[emmz0r]

Starting to wonder if the AP 2.0 dev team is one guy that works part time.
The Tesla Motors & NVIDIA Automotive Partnership

NVidia has demonstrated FSD > a year ago. Should be much more refined by now. As I mentioned earlier, Tesla probably has access to NVidia's source for self driving. Their demo looked pretty good. Even OpenPilot from comma.ai with ONE camera seems to be handling this fine.

I expect nothing less than to be blown away in August.

 

[conman]

I expect nothing less than to be blown away in August.

I really want to believe this, truly, however I expect nothing less than to be disappointed.

 

[Joe F]

But don't take anything i say for granted. I don't know for sure, its just what I heard and what I've experienced myself.

However, the geolocation about automatic lane change is very true! tested that with AP1 and AP2 just last week.

So you're suggesting someone geofenced, road segment by road segment, an entire length of a road or highway, as opposed to indicating a particular class of a road has been selected as available to auto lane change? Perhaps you might want to rethink that idea. That would be a lot of work.

There have been reports though where map data has been wrong for road segments, especially speed limits, and we've al experienced that PITA problem. There's also reports that suggest over/under pass or even adjacent roads have caused incorrect speeds, and probably would affect ALC at that position, but that's not the same as an entire road being geofenced.

 

[S100D]

I'm already still on 21.9



Not really a divided highway is this? Asking because I believe AP is for divided highways.

Not true, autopilot works on any roads that have lane markings.

AP1 autopilot will do auto lane changes on secondary roads which is something that Tesla could give us AP2 owners but they refuse to do so.

 

[S100D]

One class action suit down (don’t spend that $20-$280 all in one place) and more suits to come at this rate, yeah.

Has anyone received the $20 to $280? I am still waiting...

 

[ricebucket]

Starting to wonder if the AP 2.0 dev team is one guy that works part time.
The Tesla Motors & NVIDIA Automotive Partnership

NVidia has demonstrated FSD > a year ago. Should be much more refined by now. As I mentioned earlier, Tesla probably has access to NVidia's source for self driving. Their demo looked pretty good. Even OpenPilot from comma.ai with ONE camera seems to be handling this fine.

I expect nothing less than to be blown away in August.

You're really underestimating the difficulty of this.

FSD, in general, has *never* been successfully accomplished in the history of the world.

There are lots of companies racing to it, and some are making good progress, like Waymo. But it is still a very hard problem.

A demo on select roads is infinitely easier than the real world. Maybe even 90% success is possible at this stage. But the remaining 10% will probably be harder than the first 90%. This is true for most things in tech and in the world (e.g. getting to 0% crime from 10% crime).

And many things in life do not require the high accuracy required for FSD. For example, 99% reliability for FSD (I'll simplify this to mean num-accidents / num-times-driven) is clearly insufficient unless you want an accident in your car every 3 months. Yet, 99% is probably higher accuracy than most medical diagnoses.

(In fact, I'm not sure if I would be comfortable with even 99.9% reliability... expected one accident every 2.7 years...)

Therefore, and based on the progress I've seen so far, I do not expect Tesla to be anywhere close to FSD by version 9.

 

[diplomat33]

Therefore, and based on the progress I've seen so far, I do not expect Tesla to be anywhere close to FSD by version 9.

I don't think Tesla has ever promised that. Musk simply tweeted that V9 would be the beginning of self-driving features being released. So we might see initial self-driving features in version 9 or version 9.1 or version 9.2. And even some features won't be the entire FSD. We are not going to see the entire FSD released in August. Frankly, I am a little worried that 2 things are going to happen. 1) Some optimistic folks are going to assume that the entire FSD is coming out in August and are going to get super disappointed and mad when it does not. 2) The naysayers are going to claim that FSD was supposed to come out in August and when it does not, they are going to claim that Tesla failed at FSD and take it as more "proof" that FSD is never going to happen and it's all a big scam. But in both cases, Tesla never promised FSD in August.

 

[ricebucket]

I don't think Tesla has ever promised that. Musk simply tweeted that V9 would be the beginning of self-driving features being released. So we might see initial self-driving features in version 9 or version 9.1 or version 9.2. And even some features won't be the entire FSD. We are not going to see the entire FSD released in August. Frankly, I am a little worried that 2 things are going to happen. 1) Some optimistic folks are going to assume that the entire FSD is coming out in August and are going to get super disappointed and mad when it does not. 2) The naysayers are going to claim that FSD was supposed to come out in August and when it does not, they are going to claim that Tesla failed at FSD and take it as more "proof" that FSD is never going to happen and it's all a big scam. But in both cases, Tesla never promised FSD in August.

Yes, I know he didn't promise all of FSD by version 9. I'm just saying don't expect progress to be that smooth or easy. And definitely not "blown away" level expected by the poster that I was replying to.

 

[rnortman]

You're really underestimating the difficulty of this.

FSD, in general, has *never* been successfully accomplished in the history of the world.

This is really the core issue here. Musk fans love to point out that he has a history of doing things that people said were not possible, and he does it by being bold, taking big risks, and pushing his engineers hard. But his previous successes -- Paypal; sexy, high-performance, mass-manufacturable EVs; rockets -- are all basically taking something that has been done before in some form and making them better -- sometimes substantially, leaps-and-bounds better, but still basically applying principles and technologies that already been proven in the lab combined with good, creative engineering and design.

Autonomous vehicles are fundamentally different. Nobody has ever done it. We're not refining an existing, proven technology. We're not taking something that works in the lab and scaling it up. We're still at a point where breakthroughs are needed -- particularly if you're talking about a camera-based (no lidar) system like what Tesla bet on -- it's still a research project, not an engineering and product development project.

Musk has no track record with things at the basic science stage of development. He has a track record of product development based on technologies and principles that are basically already reasonably well understood and proven at least in the research labs. But he thought (no doubt influenced by what that equally foolish Comma.ai guy was telling him) that self-driving cars were pretty much the same thing -- that we already basically knew how to do it and you just needed to be bold, take a risk, and push hard on your engineers and it would happen on a schedule that people said was impossible.

He was wrong. Self-driving isn't like his other bold projects. It's very different, and as evidence of that, we're coming up on the two-year anniversary of the AP2 announcement and we barely have AP1 parity (still lacking in some ways). Tesla has been slowly watering down the EAP and FSD option descriptions on their website over the past year or so -- compare what they're selling now to what they were selling with the original EAP and FSD orders. If they were ready to blow everybody away with how much progress they made, do you think they'd be watering down their promises? Only if they needed to move the bar lower so that they could jump over it more easily would they do this...

 

[diplomat33]

I still think that it is important to clearly define what the goal is for what Tesla nicknames "FSD". If the goal is full L5 autonomy then yeah, I think we will be waiting a very very long time for that. But if the goal is more L3 autonomy, especially on the highway, where the car can do autopilot truly hands free, that is a "FSD" goal that I think Tesla can realistically achieve with AP2 in the near future.

 

[croman]

I still think that it is important to clearly define what the goal is for what Tesla nicknames "FSD". If the goal is full L5 autonomy then yeah, I think we will be waiting a very very long time for that. But if the goal is more L3 autonomy, especially on the highway, where the car can do autopilot truly hands free, that is a "FSD" goal that I think Tesla can realistically achieve with AP2 in the near future.

That is changing the goal posts. Musk is very clear. Tesla FSD is L5. Nothing less. It might not start out that way, but that's what they intend with FSD (how else is it fully self driving if its not at least L4?). I'd be thrilled with L4 in a year but my experience teaches me this won't happen. Tesla is mostly smoke and sometimes mirrors. Rarely do they deliver what they say they will. Sometimes they deliver more but mostly they don't deliver it or deliver it poorly (e.g. everything to do with AP2 to date).

 

[diplomat33]

That is changing the goal posts. Musk is very clear. Tesla FSD is L5. Nothing less. It might not start out that way, but that's what they intend with FSD (how else is it fully self driving if its not at least L4?). I'd be thrilled with L4 in a year but my experience teaches me this won't happen. Tesla is mostly smoke and sometimes mirrors. Rarely do they deliver what they say they will. Sometimes they deliver more but mostly they don't deliver it or deliver it poorly (e.g. everything to do with AP2 to date).

Well, I am moving the goal posts but only for the initial roll out because I am not expecting L4/5 right away in the first update. But I do expect that Tesla is aiming for L4/5 as the ultimate goal. Personally, I think the Tesla page is describing L4. But I realize that people can have different interpretations of the FSD page and see it as L4 or L5.

 

[RDoc]

If summon is going to work if you are in NYC and the car starts out in LA, it's going to have to be pretty close to L5. That was promised years ago.

 

[rnortman]

Well, I am moving the goal posts but only for the initial roll out because I am not expecting L4/5 right away in the first update. But I do expect that Tesla is aiming for L4/5 as the ultimate goal. Personally, I think the Tesla page is describing L4. But I realize that people can have different interpretations of the FSD page and see it as L4 or L5.

Recent versions of EAP descriptions (once they took out the language about uninitiated lane changes) are solidly L2, whereas once upon a time it really read like a highway L3 system (particularly the uninitiated lane changes). Recent versions of the FSD description, which no longer include Park Seek Mode, could also be interpreted to mean L2, because nowhere -- nowhere -- do they say that you won't be responsible for operation of the vehicle at all times. Nowhere do they say you can read a book or take a nap. If you can't read a book while the car handles the driving, that's L2, not even L3, and certainly not L4/L5.

Now, they may, eventually, deliver more than they are currently promising. But you can't ignore the fact that they have watered down their official promises over time. If you buy the EAP/FSD as described right now, you are buying an L2 driver assistance package. Go read the descriptions on the order page right now. It's embarrassing compared to what they once promised.

 

[rnortman]

If summon is going to work if you are in NYC and the car starts out in LA, it's going to have to be pretty close to L5. That was promised years ago.

I think an L4 system could qualify for this, because it's possible to map and qualify every road on the route between LA and NYC, and stick to those roads, and still do this. That would be L4, even if there's no human in the car at all.

L5 means the car can do anything a human can do, including driving off-map, off-road, during a zombie apocalypse, etc.

 

[diplomat33]

Now, they may, eventually, deliver more than they are currently promising. But you can't ignore the fact that they have watered down their official promises over time. If you buy the EAP/FSD as described right now, you are buying an L2 driver assistance package. Go read the descriptions on the order page right now. It's embarrassing compared to what they once promised.

I admit that they have watered down their descriptions. It might be embarrassing to you but it is probably a smart thing to do. It brings the descriptions down to a more realistic level of what we might actually get.

 

[rnortman]

I admit that they have watered down their descriptions. It might be embarrassing to you but it is probably a smart thing to do. It brings the descriptions down to a more realistic level of what we might actually get.

Definitely a smart thing to do. Now if they just remove the FSD demo video from the site and quit pretending that this hardware might someday allow true, practical L4 autonomy I think they'll be in good shape with respect to their promises, at least for customers going forward. How they will deal with the people who bought based on the original descriptions will be interesting to watch.

 

[SlicedBr3ad]

Recent versions of EAP descriptions (once they took out the language about uninitiated lane changes) are solidly L2, whereas once upon a time it really read like a highway L3 system (particularly the uninitiated lane changes). Recent versions of the FSD description, which no longer include Park Seek Mode, could also be interpreted to mean L2, because nowhere -- nowhere -- do they say that you won't be responsible for operation of the vehicle at all times. Nowhere do they say you can read a book or take a nap. If you can't read a book while the car handles the driving, that's L2, not even L3, and certainly not L4/L5.

Now, they may, eventually, deliver more than they are currently promising. But you can't ignore the fact that they have watered down their official promises over time. If you buy the EAP/FSD as described right now, you are buying an L2 driver assistance package. Go read the descriptions on the order page right now. It's embarrassing compared to what they once promised.

You may have missed that they currently state the following for FSD on model 3 page.

"Full Self-Driving Capability
Select Option$3,000
In the future, Model 3 will be capable of conducting trips with no action required by the person in the driver’s seat."

you're not required to read a book but also not required to look at the road which is L3 at the minumum.

 

[rnortman]

You may have missed that they currently state the following for FSD on model 3 page.

"Full Self-Driving Capability
Select Option$3,000
In the future, Model 3 will be capable of conducting trips with no action required by the person in the driver’s seat."

you're not required to read a book but also not required to look at the road which is L3 at the minumum.

I disagree. The text implies that there is a a driver in the seat and that the car is "capable" of conducting trips with "no action required" by that persion. It doesn't necessarily mean the person doesn't have to pay attention and remain responsible for the safe operation of the vehicle at all times. This is a point that can be argued, obviously, because the language is pretty vague. The point is, they have language that they can argue does not promise you will not have to pay attention, instead of making any specific claim about this. And they did specifically remove the park seek mode from the description, and also smart summon, both of which implied unsupervised autonomy of some sort.

The key here is unsupervised autonomy. They have removed all promises related to that.

Edit: Also uninitiated lane changes have been removed from the EAP description; that is also a form of unsupervised (or perhaps weakly supervised) autonomy.

 

[diplomat33]

Definitely a smart thing to do. Now if they just remove the FSD demo video from the site and quit pretending that this hardware might someday allow true, practical L4 autonomy I think they'll be in good shape with respect to their promises, at least for customers going forward. How they will deal with the people who bought based on the original descriptions will be interesting to watch.

I've said it before that in retrospect, Tesla should probably not have done the FSD demo video and probably not even advertised a FSD package at all. Instead, they should have described the AP2 hardware as simply "adding new hardware to allow for future improvements to autopilot". That way, they are not promising anything specific at all and could announce new features when they actually become available.