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

[rnortman]

Thank you for admitting that Tesla has real FSD progress.

I'm not sure he admitted that... I'll admit that they almost certainly have made progress on EAP features, and the current "FSD" option is more or less rebranded EAP with a couple of extra gimmicks. So they have made progress on "FSD" in quotes.

The question is, have they made any progress at all toward unsupervised self-driving? There is absolutely no evidence that they have and I doubt any such evidence will be presented on April 19. If April 19 actually happens.

 

[diplomat33]

The question is, have they made any progress at all toward unsupervised self-driving? There is absolutely no evidence that they have and I doubt any such evidence will be presented on April 19. If April 19 actually happens.

We will see on April 19.

 

[Bladerskb]

@diplomat33 do you actually expect real details to be given at this "Investor event" besides the usual 101, we have the best chip, best NN and all the data? Will Elon actually allow details to be given that every other sdc is giving including even APPLE?

The problem here is Tesla is exclusively leaves off of hype. Tesla is literally the breading ground of lies and myth. Just take a look at the stuff that ArkInvest puts out. Its so blatantly wrong you start to question their intelligence or if they are knowingly dispersing false info.

Will Elon actually allows Bannon and Katharpy to diverge actual real info? To me that would definitely dissolve the mysticism that surrounds Tesla AP and FSD.

 

[diplomat33]

@diplomat33 do you actually expect real details to be given at this "Investor event" besides the usual 101, we have the best chip, best NN and all the data? Will Elon actually allow details to be given that every other sdc is giving including even APPLE?

No, they won't divulge company secrets of course. But I do expect some information about AP3's general specs. And I think we can expect some information about future features and FSD roll out timeline. And of course, we will get some FSD demos that should show us a bit of what FSD is capable of right now.

 

[rnortman]

We will see on April 19.

I will see one thing, and you will see another. I will see no fundamental change that makes me think they may one day be able to operate unsupervised. You will see incremental progress in "features" in a controlled environment and believe that this brings them closer to unsupervised autonomy. You will post as such, and I will post in disagreement, again. Or maybe I won't bother. I'm getting rather worn out by this.

Maybe they'll even show the vehicle operating without a driver on a closed test track, or a parking lot! And you'll say that is proof that they are progressing toward unsupervised autonomy. And I'll disagree because you can go back more than 20 years and find reasearch groups demonstrating autonomous vehicles in controlled environments -- even some really neat completely off-road autonomous vehicles for the military and industrial/argicultural applications. But demonstrating it in a controlled environment -- that is, without other actors in the same environment that aren't part of the controlled test -- really tells you little about how ready it is for uncontrolled public roads.

 

[BigD0g]

Or maybe I won't bother. I'm getting rather worn out by this.

This!

 

[Bladerskb]

No, they won't divulge company secrets of course. But I do expect some information about AP3's general specs. And I think we can expect some information about future features and FSD roll out timeline. And of course, we will get some FSD demos that should show us a bit of what FSD is capable of right now.

Do you consider tesla's use of inception GoogleNet to be company secrets? Are you telling me all the self driving companies left and right have all been divulging company secrets while actually talking about simple details and giving presentations on their system?

You just admitted that Elon won't allow anything else besides the usual "We Da Bess". So what is there to look forward to from a guy who said you could summon your car from across the country Jan 2018 or you would be able to sleep in your car this month?

 

[banned-66611]

4) Tesla will provide FSD test drives that includes new features still under development.

That just means some more level 2 driving aids.

 

[Bladerskb]

@strangecosmos talks about how he's only writing articles to learn. But actually he presents his articles in SA as facts and have been wrong for going on almost 3 years straight and thousands of TSLA shareholders use his articles to trade daily. Literally ArkInvest gets their talking points from his "learning" articles

Tesla bull and bear debate the stock as Elon Musk faces off with SEC

Diplomat, you talk about Elon releasing AP3 specs but mind you that ArkInvest is the most ignorant Tesla shareholder i have ever seen and are already spreading lies about Tesla's AP3.

Tesla is currently using an Nvidia chip with 12 teraflops of performance. A tenfold improvement would suggest Tesla’s chip has roughly 120 teraflops of performance. Nvidia’s Xavier chip, which is currently sampling and likely won’t be installed into vehicles until late 2019, has roughly 30 teraflops of performance. In other words, traditional automakers have committed to a chip with inferior specs, and will likely have to wait for Nvidia’s next autonomous product before they can have a processing system comparable to the system Tesla is testing today.

Tesla’s Autonomous Opportunity is Severely Underappreciated


This is why @diplomat33 that i'm so vocal. You have a company going around spreading balant lies and no one says anything. Even Elon in their podcast doesn't rebuke them.

"Our analyst on AI, deep learning spent 9 years at Nvidia and concluded just looking initially at the specs that Tesla is atleast 3 years ahead of every one out there"

Why give AP3 specs when you have non-sense stupidity like this being spread for you on CNBC and all over the media for free?

 

[Bladerskb]

On the other hand, if the solution requires massive amounts of unlabelled driving data, then for now Tesla is the best poised to solve autonomy first. For instance, if the solution requires billions of miles’ worth of state-action pairs for imitation learning (or a sampling of rare state-action pairs that statistically only appear often enough in billions of miles of driving), then right now as far as I know, no other company is set up to collect that data on that kind of scale.

This is true for Tesla whether a) large-scale imitation learning is a sufficient solution by itself, or b) large-scale imitation learning is insufficient by itself but necessary to create smart enough agents to enable RL in sim — which is a sufficient solution. Whatever the exact solution is, if it requires massive amounts of unlabelled driving data, then Tesla is the best bet.

I continue to be right...

"You don't need billions of hours of driving data, we can learn to lane follow from scratch using about 10 mins of data."

With only 20 Hours

This confirms my statement was right. (to no surprise)

What Nvidia did was prove that you only need 72 hours (3k miles) of data to create a NN agent that can drive like a human in alot of cases.

 

[Tjhappel]

Your super smart... why do you argue so much when you could just agree and short the *sugar* out of the stock since if what you say is true ultimately the whole thing will blow up and people will see your truths?

Do you not want to make money or just arguing for fun? Your obviously super smart but your arguing with god knows who man, getting old.

 

[diplomat33]

@strangecosmos talks about how he's only writing articles to learn. But actually he presents his articles in SA as facts and have been wrong for going on almost 3 years straight and thousands of TSLA shareholders use his articles to trade daily. Literally ArkInvest gets their talking points from his "learning" articles

Tesla bull and bear debate the stock as Elon Musk faces off with SEC

Diplomat, you talk about Elon releasing AP3 specs but mind you that ArkInvest is the most ignorant Tesla shareholder i have ever seen and are already spreading lies about Tesla's AP3.

Tesla’s Autonomous Opportunity is Severely Underappreciated

This is why @diplomat33 that i'm so vocal. You have a company going around spreading balant lies and no one says anything. Even Elon in their podcast doesn't rebuke them.

Why give AP3 specs when you have non-sense stupidity like this being spread for you on CNBC and all over the media for free?

Clearly, you are pretty mad about Arkinvest. Why? What specific lies is Arkinvest spreading? I read the article multiple times and I did not see anything blatantly wrong in it. It's just a speculative piece about how FSD will impact Tesla. Maybe they are right, or maybe they are wrong but I did not see any blatant lies.

You quote the passage about the AP3 specs. What's wrong about that quote?

Also, Arkinvest has nothing to do with Tesla and nothing to do with the FSD event on April 19. They are just an investment firm speculating about TSLA and FSD. The topic here is the FSD event in April, not what some investment firm is saying.

 

[NetworkNerd]

I continue to be right...
...
This confirms my statement was right. (to no surprise)
...
Like how many times do i have to be right (about tesla) before people start noticing?

What a dickwad

 

[ZeApelido]

Blader, you are such a hack.

Learning lane lines in 10 minutes?

Okay buddy boy. Sure.

 

[diplomat33]

"You don't need billions of hours of driving data, we can learn to lane follow from scratch using about 10 mins of data."

What Nvidia did was prove that you only need 72 hours (3k miles) of data to create a NN agent that can drive like a human in a lot of cases.

Your post is misleading. Yes, you only need a small amount of data to create a good enough NN to handle the cases in that data. The 10 mn of data worked to teach lane follow for those cases and the 3k miles of data created a NN good enough to handle "a lot of cases" in that 3k miles of data. So no you don't need billions of miles of data just to learn a few small cases. But that is a no-brainer. The method will fail if the car encounters a case that it has not learned yet. So the 10 minutes of data does not teach the car how to lane follow in all cases and the 3k miles of data is not good enough to handle all driving cases. If you want to use this method to learn more driving cases, you need more data. Ultimately, if you wanted to use this method for a general solution of self-driving that can handle ALL driving cases, yes, you would still need billions of miles of data.

 

[diplomat33]

Tesla is rolling out of NOA without confirmation:

Since we first introduced Navigate on Autopilot last year, Tesla drivers have traveled more than 66 million miles using the feature, and more than 9 million suggested lane changes have been successfully executed with the feature in use. We’ve heard from our customers that it makes road trips and highway driving more relaxing, enjoyable and fun, and gives them an easy way to follow their car’s navigation guidance when traveling on an unfamiliar route.

Today, we’re beginning to roll out our latest version of Navigate on Autopilot for a more seamless active guidance experience. In this new version, drivers will now have the option to use Navigate on Autopilot without having to confirm lane changes via the turn stalk. Here’s how it works:

In the Autopilot settings menu, a driver can press the Customize Navigate on Autopilot button which will now display three additional settings – Enable at Start of Every Trip, Require Lane Change Confirmation, and Lane Change Notification. Through the Enable at Start of Every Trip setting, Navigate on Autopilot can be set to automatically turn on each time a driver enters a navigation route. Once enabled, anytime a driver is on a highway and uses Autopilot with a location plugged into the navigation bar, the feature will be on by default. If a driver selects ‘No’ to Require Lane Change Confirmation, lane changes will happen automatically, without requiring a driver to confirm them first. Drivers can elect to get notified about an upcoming lane change by receiving an audible chime as well as a default visual prompt. Additionally, all cars made after August 2017 will also have the option to have their steering wheel vibrate for the alert as well.

Each of these notifications are meant to provide drivers with the opportunity to check their surroundings and determine whether they want to cancel the lane change before it’s made. Cancellations can be made by moving the car’s turn signal or by pressing the lane change cancellation pop-up notification on the car’s touchscreen. This feature does not make a car autonomous, and lane changes will only be made when a driver’s hands are detected on the wheel. As has always been the case, until truly driverless cars are validated and approved by regulators, drivers are responsible for and must remain in control of their car at all times.

Through our internal testing and Early Access Program, more than half a million miles have already been driven with the lane change confirmation turned off. Our team consistently reviews data from instances when drivers took over while the feature has been in use, and has found that when used properly both versions of Navigate on Autopilot offer comparable levels of safety. We’ve also heard overwhelmingly from drivers in our Early Access Program that they like using the feature for road trips and during their daily commutes, and we’re excited to release the option to the rest of the Tesla family.

These new settings will be available to customers who have purchased Enhanced Autopilot or Full Self-Driving Capability. They will begin to roll out today via an over-the-air software update to customers in the U.S., and will be introduced in other markets in the future pending validation and regulatory approval.

Introducing a More Seamless Navigate on Autopilot

Note it took Tesla 66 million miles of driving data and 9 million miles of data of suggested lane changes to make NOA without confirmation possible!

 

[S4WRXTTCS]

Now that Blader has had his 15min to rant, can us FSD owners (or perspective owners) actually talk about this "Autonomy Investor day" as it impacts us?

To my knowledge Tesla has never set aside time like this to actually talk about a roadmap or to a deep dive into their self-driving technology.

In fact Tesla didn't have any kind of FSD roapmap until recently, and the recent roadmap leaves a lot of head scratching.

So obviously I'm encouraged by this event to get more details on exactly what they're planning on doing.

Can investors submit questions? I'm not an investor, but I imagine a few read these threads.

There is a hell of a lot of details that need to be spelled out.

How are AP2 cars going to be upgraded?
How will the upgrades impact Tesla financially? Do they have any tricks we're not aware of in upgrading them?
From a driving policy standpoint when will AP2 get fixed so it's usable in the right lane without "recentering" everytime a merging lane happens.
When will NoA be fixed so it's actually driving on Maps versus mostly acting as a driver torture device? Pressing "cancel this lane change" doesn't actually cancel it. It just delays it for a bit before it prompts again.

 

[S4WRXTTCS]

Tesla is rolling out of NOA without confirmation:

Introducing a More Seamless Navigate on Autopilot

Note it took Tesla 66 million miles of driving data and 9 million miles of data of suggested lane changes to make NOA without confirmation possible!

Man you're on the ball as that's the first I heard of that blog post.

I didn't even realize my car had a steering wheel capable of vibrating. That's good news for us, and sadly more bad news for AP2 people.

 

[diplomat33]

The fact that our cars can detect traffic lights AND we are getting NOA without confirmation almost back to back AND Tesla is doing a FSD demo on April 19 is very promising for "FSD". Things do seem to be moving more quickly now.

 

[mhan00]

Do you think they'll actually release NoA without confirmation in the next 6 months?

As it exists right now there are plenty of users like yourself or myself who feels like it has some serious issues.

You brought up a good example of how sometimes it seems to use proximity to an exit as to when to get over while failing to know which lane it's really supposed to be in for that exit. Other times the navigation is simply wrong, and has it taking an exit it shouldn't be taking.

Then we have the issues with the aborted lane changes that happens around 20% of the time (at least the last time I checked with the last 2018 release).

We've known for awhile that unconfirmed lane changes were being tested by a select group of people. So I fail to see what's different now. Aside from maybe a few more people on the EAP program. I haven't seen a whole lot leak out about the progress. So either people have gotten really good at not talking about fight club (which I doubt) or it's still a very tiny amount of select people. In that case it's likely months away.

The problem with releasing this before it's ready is the level of ballsiness that you brought up.

It's one thing to be ballsy, but it's another thing to be stupid.

Personally I think it would stupid to release it until the likelihood of it taking an exit it's not supposed to take is reduced to a very small number. If not it's going to result in drivers doing quick corrections from a lane that's ending to a lane left. Where they might do so in a rushed manner.

We can also pretty much guarantee that anything that lessens driver engagement, but maintains driver responsibility will result in a few crashes.

Even if the NoA was absolutely perfect in it's lane choices the human driver would intervene at times where the human was confused. Like "Oh, crap I think I wanted that exit..." followed by "Oh, sorry you were right".

Anyone watching the Tesla swerve around would be going "What the hell is that driver doing???"

I'd rather see this get put on the back burner, and instead for Tesla to focus on improving NoA along with navigation maps.

The answer is yes, since they released a blog post saying they’re releasing the NOA update with unconfirmed lane changes is releasing now. The caveat is that it requires your hand to be on the wheel (and providing a bit of torque to make sure your hand is there to intercede if necessary), but it’s another step. NOA has been great for me in my routes on SoCal, but others’ experiences has varied. I’ll be very interested to see how this performs. Hopefully I get the update before the end of the month since I’m planning a couple of trips out to some national parks for some sight seeing/hiking and the long, two way highways with limited cars should be a nice safe way to use it.