AP2 - Definitely heading in the right direction...

[NerdUno]

Personal attacks aside, the comments from actual AP2 users have been extremely helpful to many of us in identifying limitations and avoiding potential accidents. Just because we all don't sing the praises of Tesla 24/7 doesn't mean we want to or should take our marbles and go home. A more honest presentation of the existing Tesla feature set from Tesla would go a long way towards eliminating future complaints from new owners. And it would make the driving experience considerably safer.

 

[Driver Dave]

Personal attacks aside, the comments from actual AP2 users have been extremely helpful to many of us in identifying limitations and avoiding potential accidents. Just because we all don't sing the praises of Tesla 24/7 doesn't mean we want to or should take our marbles and go home. A more honest presentation of the existing Tesla feature set from Tesla would go a long way towards eliminating future complaints from new owners. And it would make the driving experience considerably safer.

I read the clear descriptions for AP2. Understood it would not be turned on when I received the car.

I read each update. Understood what each one was adding.

I tried carefully each new features. Understood pretty quickly what it did and did not do.

I tried the last update. Was pretty surprised to see how well it worked but understood I needed to pay attention and not use when the road went beyond its current capabilities.

I read comments here how tesla hid information buried in tiny fine print and that AP2 is dangerous.

At that point, I did not understand.

 

[NerdUno]

You forgot to mention that Tesla also changed the fine print.

 

[Taylor Sherman]

Well, I bought the car and the delivery person described TACC + autosteer like they worked. And so I turned it on - it does say "Beta" in the control panel, but it's not like it says "hey, this can't really see cars to the side of you yet and may not brake if someone pulls in front of you etc".

I'm not saying Tesla is underhanded, but giving people easy access to features that don't really work yet is a real gamble business-wise.

 

[Taylor Sherman]

Addendum: then, talking over things with a friend that owns an AP1 car (again, everything the salespeople said about AP2 was about its superiority to AP1), he said "oh yeah, those things just don't work with AP2, they work great for me on my older car haha but they should work for you soon". He then sent me a link to the great spreadsheet of all the AP1 stuff that doesn't work in AP2 yet. That was pretty annoying and enlightening to see.

 

[essmd]

But AP1 v AP2 is irrelevant in so far as they both require you to be alert and ready to intervene. The only difference is AP1 requires less intervention, but it still requires it and always will

Agreed, but compared to AP1, AP2 requires a higher vigilance given it is less accurate.

AP2 is very close to AP1, and the latest update strating to roll out as I write this essentially makes them equal, if not AP2 better. From now on AP2 will exceed AP1 capabilities. Oh, and 6-8 months is incredably fast for Tesla to duplicate Mobileye-functionalty that had taken Mobileye years to develop.

I promise I am a more qualified evaluator of AP1 vs AP2 having AP1 for 8 month and AP2 for almost a week, and I say you are not correct. Are you sharing a dream you had, seen the future update that no one else has, or just another Tesla Fan Boy? Duplicate??? the deficits of AP2 are just frustrating.

That was the PR stunt that MobileEye chief pulled, to tarnish Tesla.

It was obvious that Tesla wanted to iterate and go at a faster pace than what MobileEye could possibly do, and when Tesla started plans of using nVidia hardware, MobileEye learnt they are being shunted out anyway and then decided the pull the plug earlier than Tesla would have done. So the gradual phase out that Tesla had planned did not materialize.

Besides there was also a rumor that Mobile wanted IP rights to all the data Tesla is collecting through their hardware - which is laughable and outrageous. This is like Intel or Oracle stating that all your data belongs to us !!

It seems like all of the stories about what Tesla and Mobile Eye thought are just that...stories, more like fiction. Nobody but those engaged with the meetings, correspondence, or phone calls really know what each party thought. All that we do know is that the split has not done well for the owners...so far.

Not so fast . AP1 reads speed limit signs and displays vehicles in adjacent lanes on the dash. Still some more catching up to do.

Isn't AP2 supposed to already read speed limit signs? I though it does, yet mine does not seem to do so.

 

[Shock-On-T]

Yep, they don't get it... over, and over, and over, and over... thank god for the ignore function.

The ignore function: it's like shaking a rock out of your shoe. Bliss!

 

[Shock-On-T]

"Auto" pilot in an airplane (this is the origination of the term) will surely fly you into a mountain if the pilot does not intervene. In an airplane it is meant to Help and assist the pilot in the duties of flying the plane. It is never meant to take over as the operator responsible for the safe operation of the plane. It is the same as the autopilot in the Tesla.

In fact it will fly you into the mountain automatically! No effort required!

 

[JonG]

Agreed, but compared to AP1, AP2 requires a higher vigilance given it is less accurate.

.

That's the mistake. They require the same vigilance and always will, forever, on AP and EAP and probably FSD, however good those system become. The difference is NOT the amount of vigilance, it's the amount of intervention.

It's ONLY when the cars have systems that are 100% self driving for large portions of the journey and require no unexpected intervention during those times, (and I mean no, none, ever) can your vigilance be dropped during those periods.

 

[dhanson865]

somebody remind me why Tesla changed vendors

There is no new vendor. AP2 is in house hardware replacing a prior vendor.

The old vendor tried to blackmail Tesla with an onerous contract or you would have the old vendors hardware in addition to what is currently known as AP2 and it would be purely superior to AP1 by now (and likely so 6 months ago even).

As is Tesla wouldn't sign the new contract and we have what we have now. AP2 without the old venders hardware instead of AP2 with the old hardware and new combined.

 

[JonG]

There is no new vendor. AP2 is in house hardware replacing a prior vendor.

The old vendor tried to blackmail Tesla with an onerous contract or you would have the old vendors hardware in addition to what is currently known as AP2 and it would be purely superior to AP1 by now (and likely so 6 months ago even).

As is Tesla wouldn't sign the new contract and we have what we have now. AP2 without the old venders hardware instead of AP2 with the old hardware and new combined.

Wrong.

Hardware in nvidia.

Which version of events do you want on the why? Tesla say because development wasnt moving as quick as they wanted, mobileye because tesla were exceeding what they considered safe especially after the Brown accident. Also speculation about volume buyers on mobileye kit wanted to slow Tesla down v Tesla wanting IP of all data.

 

[dhanson865]

Personal attacks aside, the comments from actual AP2 users have been extremely helpful to many of us in identifying limitations and avoiding potential accidents. Just because we all don't sing the praises of Tesla 24/7 doesn't mean we want to or should take our marbles and go home. A more honest presentation of the existing Tesla feature set from Tesla would go a long way towards eliminating future complaints from new owners. And it would make the driving experience considerably safer.

Yes, +1, like.

I'd love to see people stop using the phrase "tried to kill me" when discussing autopilot. But I very much want to know when it doesn't work to expectations.

It's a shame people gravitate to one extreme or the other. Or at least the ones at the extreme positions post more often than the ones in the middle.

 

[dhanson865]

Wrong.

Hardware in nvidia.

Which version of events do you want on the why? Tesla say because development wasnt moving as quick as they wanted, mobileye because tesla were exceeding what they considered safe especially after the Brown accident. Also speculation about volume buyers on mobileye kit wanted to slow Tesla down v Tesla wanting IP of all data.

Nvidia is involved in AP1 and AP2. The part that changed is removing Mobileye. That doesn't bring a new vendor in, it just removes one.

The old hardware was:

Main screen: Nvidia Tegra 3
Cluster: Nvidia Tegra 2
Autopilot: Mobileye EyeQ3
+ whatever Tesla added to the mix in hardware and software. Tesla's implementation wasn't stock off the shelf anything.

Previously Mobileeye pre processed video before handing over to the Tesla software (or you have to assume Tesla ran custom code on the mobileeye hardware). Now Tesla has to do that processing on the same system that is doing the rest of the work that Mobileeye didn't do before.

Yes, Telsa improved the remaining hardware, no they didn't add a vendor.

 

[lunitiks]

Define "involved"

 

[JonG]

Define "involved"

And while you're at it, define "in house"!

Maybe worth reading this first

Tesla Self-Driving Car Built on NVIDIA DRIVE PX 2 | NVIDIA Blog

 

[daktari]

Well, I kind of agree with @Driver Dave that AP2 (FW 17.17.17) is rock solid in a pretty straight autobahn/highway. On a european highway (speed limits 130 km/h) with bends, crests, undulations it does behave strange quite often, and is not always comfortable. These details need to be included in the discussion.

Yesterday I did a test together with my neighbour and his AP1 S85D on a local twisting roads. AP1 felt quite similar to AP2 in fact, speed 50 km/h. AP1 messes up some places AP2 doesn't and vice versa. See the video at bottom.

So, why am I impatient? Well I saw the FSD video from Tesla last year and ordered a new AP2 EAP+FSD car to get the feature they promoted instead of bying an inventory AP1 at a much lower price. I hope you have compassion for this (misleaded) choice, instead of pointing out how naive I was... Can't wait for the new firmware though!

 

[Driver Dave]

I hope you have compassion for this (misleaded) choice

Yep, I do, but I don't think you'll regret it. I got my AP2 thinking some of the features may have been here sooner, but definitely have been enjoying each update and following along as they develop it in realtime.

Also, I know that when software guys say a month think 2.

But in the end, this year has been fun watching it go from nothing to something, and I can't wait for the next 6 months, because unlike AP1, AP2 keeps moving along and they aren't even close to unlocking what all the hardware can do.

They just have to keep feeding it data... (which is what takes time)

And it will go from nothing to something to beyond everything.

 

[gearchruncher]

And it will go from nothing to something to beyond everything.

That is all what we hope for, but this optimism that Tesla will beat everyone is starting to see unfounded. It's taken them 4-30X as long as they thought to get to where AP1 was depending on what tweets you use. Even they say FSD isn't going to happen this year and it was announced last year. Assume it takes 4X as long as they are saying, which is at least 4 years. Do we really think no other car maker or AI system will be able to do what Tesla does by then? Waymo can drive 5,000+ miles right now with no disconnects while Tesla is at 3 miles. There isn't a lot of evidence that they will be the ultimate leader in this space. They're just first.

 

[RLC3]

"... Waymo can drive 5,000+ miles right now with no disconnects while Tesla is at 3 miles. ..."

We drove our AP2 Model S about 1300 miles on a road trip two weeks ago. I estimate that we used AP for about half the distance. We didn't use it on very curvy mountainous sections of the road where the 'jagged' steering in tight turns felt uncomfortable, although we did try it on those sections a few times.

We commonly drove for 45 minutes without dropping out of AP. When we did it was driver's choice, not because we would have hit anything. One reason we would drop out is that AP holds the center of the lane even when next to a semi in a curve, and we prefer to move to the far side of the lane. But in no case during our trip would the AP behavior have caused an accident. I can't speak to whether or not that would have been true in the sections of our trip where we chose not to use it.

So I am looking forward to smoother curves, and the many little touches like lane positioning that I hope will eventually come as the product matures, but we found using it to be a plus on our trip. And, BTW, my wife is a nervous driver, and even more nervous when I am driving. But during this trip she went from tightly wound whenever AP got engaged, to relaxed, and from seldom using it herself unless I suggested it to using it whenever conditions were good.

So yes, it is not yet a polished product. Yes, it's not a level 4 driving system that does not require you to pay attention all the time. And yes, Tesla promotes it. And yes, when I demoed it for a technically sophisticated friend without any hype on my part, he said "NO WAY!" when I turned it on and put my hands in my lap, and it followed the next curve perfectly, and slowed behind the car in front.

I'm looking forward to continued improvement.

 

[Driver Dave]

That is all what we hope for, but this optimism that Tesla will beat everyone is starting to see unfounded.

I don't see it that way.

Currently, Tesla is "live" while everyone else is in "demo" mode.

It is easy to have a car drive around itself. The tech is there. Nvidia has cars doing it. Waymo, others. A demo car driving around with AI, no biggie at this point. Demo land is very very easy.

Now the hard part is getting this tech "live", out in the wild, released, with actual customers, in real cars. That's a whole other thing.

Why is Tesla taking so long with some of these features? It's simple: when live, you can't just push things out. Tesla has a very careful path they need to follow to add these features into the car. It is not surprising they are taking it carefully, one step at a time. I can guarantee you there are self driving demo teslas right now, but no, they are not just going to release those demo, but will pick and choose features that are rock solid ready to go out into the wild.

So will Waymo, with tech, but no cars, be able to catch up? Will they do a deal with car companies and slap together something?
With other car companies figure out how to make electric cars? Will they be able to change their culture? Will they cling to the dealer model which just sucks for consumers? Will they have to 3rd party everything making it hard to make a profit and nothing really works that well together?

My gut take is, maybe, but don't bet on it.

There is just such a huge value to Tesla actually doing it now. Electric cars, supercharger network, self driving AI, no dealers, manufacturing, delivery, service... it is pretty amazing where they are at right now. And it's all live. No demo. Live.

And the other guys? It will be some time, a long time, before they are even close to what Tesla has put together.

And by the time they get to where Tesla is right now, Tesla will be way beyond it.

Look at it this way. Tesla was first in 2012. It was fair to say "They're just first" then.

It's 2017. They are no longer "just first" They are at another level, and accelerating.