Autonomy Investor Day - April 22 at 2pm ET

[diplomat33]

When and then after it being feature complete, how long until the march of 9s become reliable enough for usage?

According to Tesla, they will be feature complete later this year. Based on everything we know, I definitely think that will happen. In terms of how long until the march of 9's, that is harder to know. Personally, I think Musk is too optimistic about that part. However, I do think that the growth of the Tesla fleet and the increasing amount of data that Tesla can collect as a result, will help Tesla achieve that march of 9's faster. The big problem with the march of 9's is that there is just so many edge cases out there, probably millions and millions. If your fleet is small, you are just not going to do enough driving miles in enough different locations to see all those edge cases in a timely manner. That is the advantage that Tesla has. They have an every growing huge fleet of cars driving on lots of different roads in different areas. When you have hundreds of thousands of cars around the world all collecting data and sending it back to the machine for learning, like Tesla does, you are much more likely to encounter all those edge cases quicker. So I don't think it will take years and years for Tesla to complete the march of 9's.

 

[Lasairfion]

The importance of 'Dojo' really cannot be underestimated. I understand that Elon didn't want to talk about it further, but it's a game changer in hitting all the needs of AI learning. There are two types of training for AI. The labelling and learning approach mentioned extensively in the presentations for Autonomy Day (and in a great video by CGP Grey), and the second type where the machine takes the best of its own decisions, and pulls them together into one and then trains against that a bajillion times at super-speed before folding in those results for the next time.

OpenAI's defeat of the human team in the DOTA Competition is proof that Elon's companies are way, way ahead at this game.

 

[mongo]

This one is definitely not being reported. It happens for so many reasons that it would be impossible to filter out "bad choice being made by autopilot" versus "driver choice."

On an individual event basis, sure. However, wouldn't it work from a macro heat map POV? If many Teslas report disengagements in the same location, that is something to check into. Just need to upload GPs coordinates and timestamp.

 

[CarlK]

The importance of 'Dojo' really cannot be underestimated. I understand that Elon didn't want to talk about it further, but it's a game changer in hitting all the needs of AI learning. There are two types of training for AI. The labelling and learning approach mentioned extensively in the presentations for Autonomy Day (and in a great video by CGP Grey), and the second type where the machine takes the best of its own decisions, and pulls them together into one and then trains against that a bajillion times at super-speed before folding in those results for the next time.

OpenAI's defeat of the human team in the DOTA Competition is proof that Elon's companies are way, way ahead at this game.

I'm sure they did not tell use everything but what have showed in the presentations are truely amazing already. Those shadow mode NN learning to make machines to predict intentions of cars and pedestrians is mind boggling. That is the remaining missing link and the game changer imo. Tesla is the only company that has the infracstructure, camera equipped large machine learning fleet, to do this. As it turned out Lidar's main problem is it does not have the resolution to do this (that's from the original Google Lidar guru Levandowski's recent comment).

Well, Tesla is an auto maker. Their business is to sell cars. So yeah, they are going to try to sell cars. But I don't believe "FSD" is a ploy or a scam. Tesla's self-driving is very legit. The cars have the hardware for self-driving and the dev team has made great progress towards self-driving.

And as I mentioned before, if people think Tesla cars are full self-driving now where they can sleep in them, I blame the media for the most part. They are the ones who put out clickbait articles about how Tesla cars are Full Self-Driving now. Musk talks about the cars having the hardware NOW but he's been consistent that being able to sleep in your Tesla while it drives you is still in the future.

It's absurd that some detractors are trying to downplay this great progress as a ploy to fool people to overlook other things. So Tesla started autopilot five years ago, hired people to design HW and SW for FSD at least three and half years ago and put it on the master plan part duex three years ago and put sensors and computers in all cars it makes all because it could be used in Q1 2019 when they know company will have problems? That would be the best long term planning I've ever seen. Even that why spend all those precious money on the project not to mention cost of fitting every car with sensors and computers if you are not serious about the project? Wouldn't it be much better just to keep the money to add to the bottom line as profit if the intention was just for having something to fool people? Those FUDster will do anything to serve their agenda and they even have guts to accuse others are doing it. I think they are really scared from what they heard.

[

 

[ChrML]

I respectfully disagree. I feel Elon goes into his snake-oil salesman role every time he pitches that ”every mile driven the fleet learns” mantra. The contrast couldn’t have been clearer on the 22nd when he was next to Andrej Karpathy who was giving clear, plausible answers on what the ”shadow mode” triggers and data gathering actually did — and Elon tried to interject with this every mile thing. Just like most of that day, Elon looked much less believable than his fellow presentation givers. And indeed not technically accurate as we’ve known about this.

He’s an engineer, we always say. On this, he is not showing engineer credentials, but something else.

What really happens is for some miles driven, in some cars, a trigger is launched that collects some data that is sent to Tesla. Tesla then uses this small amout of data for validation and improvement that eventually leads to new software updates and NNs being sent to the fleet. ”Every mile driven” the fleet out there does not learn a thing, nor does it even collect much data every mile driven other than some general statistics.

Why the need to tell a lie? The deployment and validation, and data gathering, benefit is impressive as it is. But the fleet does not ”learn” anything ”every mile driven”. You could see how this misunderstanding crept even into the audience when they were asking about NNs being trained in the cars...

I say this with great appreciation for the improving data gathering triggers Karpathy provided new insights into. As I’ve said the trigger part of the presentation was the highlight of the day for me. I think they are making strides in this area. Just give us Karpathy giving it like it is and I’m a happy camper on this one.

What would they need to do to make the "billion miles gathered" statement true?

I don't see any reason to upload terrabytes of data saying "now it's ok too".

As long as the hardware continously monitor performance locally, report back the number of miles driven on autopilot, and send based on the neccessary triggers and disengagement reports to report every mishap. Then I believe this counts as "gathering data from billion miles driven".

Because it's all the data they need to add 9's to the reliability. And using shadow mode to measure whether a change was positively or negatively impacting the reliability before deploying next revision.

 

[CarlK]

What would they need to do to make the "billion miles gathered" statement true?

I don't see any reason to upload terrabytes of data saying "now it's ok too".

As long as the hardware continously monitor performance locally, report back the number of miles driven on autopilot, and send based on the neccessary triggers and disengagement reports to report every mishap. Then I believe this counts as "gathering data from billion miles driven".

Because it's all the data they need to add 9's to the reliability. And using shadow mode to measure whether a change was positively or negatively impacting the reliability before deploying next revision.

Billion miles driven and monitored is not the same as billion miles of data recorded. Only data of needed edge cases (requests on those are sent to cars) and special cases determined by the car (like outside cars changing lanes or driver intervention) will be recorded and used. Karpathy has described these very clearly in the presentation. Certain people seem to refuse to understand the whole thing but could only take a sentence or two out of context to argue. It does nothing but to reenforce the appearance of their ignorance and lack of objectivity.

 

[Xminus6]

Interesting that this study by Cornell came up in my news feed. It confirms, through their independent research, that cameras, mounted high on the car CAN have accuracy that rivals LIDAR for a fraction of the cost.

https://arxiv.org/pdf/1812.07179.pdf

 

[diplomat33]

Here is a testimony from Morgan Stanley:

Overall, pretty positive. The reference to "no LIDAR" is key. Tesla is achieving this without LIDAR. But we know FSD is far from perfect since it is not feature complete yet. So the one disengagement is not surprising. FSD will get even better. FSD is making progress and seems very promising to me.

 

[CarlK]

Interesting that this study by Cornell came up in my news feed. It confirms, through their independent research, that cameras, mounted high on the car CAN have accuracy that rivals LIDAR for a fraction of the cost.

https://arxiv.org/pdf/1812.07179.pdf

After Monday's Karpathy presentation and Levandowsk interview the issue is pretty much settled and Lidar is dead. I'm only waiting for people like Lex Fridman to come out for the final confimation. He already seemed to be tilting toward camera + NN over Lidar in his latest talk and put Tesla and Waymo as leader of each. Those Lidar people, which is all of them, will have to switch at one point when it makes neither economical nor technical sense. It will take them at least two or three years even if they switch today.

 

[Guy V]

I do have some abilities that a stationary camera does not. For one, I can move my head. Second, I can clean my eyes if something gets in them. Third, I have redundant vision (two eyes) in all directions through moving my head. I can even leave the car if need be.

For example: Tesla’s entire coverage on a 90 degree intersection leftwards or rightwards is the respective B pillar camera. No other sensor can see the intersection towards the left and right. There is no radar pointing there, the ultrasonics do not see far enough nor are fast enough and the side marker cameras have no coverage there. The fisheye sees forwards but not wide enough.

You are making life and death decisions to merge based on a single, small, stationary camera input without a wiper.

But your processing speed and reaction time are beyond pathetic.

Of course I am nearly 70 and mine are likely much worse. That is why I am looking hopefully forward to FSD for safe, convenient and enjoyable independent mobility.

 

[Bladerskb]

Well, Tesla is an auto maker. Their business is to sell cars. So yeah, they are going to try to sell cars. But I don't believe "FSD" is a ploy or a scam. Tesla's self-driving is very legit. The cars have the hardware for self-driving and the dev team has made great progress towards self-driving.

No one is saying that FSD is a scam. We are saying the Level 5 in 2 years which elon has been saying since 2015 IS a complete scam and a dangerous one at that.

71% of people around the world believe they can buy self driving cars today

And as I mentioned before, if people think Tesla cars are full self-driving now where they can sleep in them, I blame the media for the most part. They are the ones who put out clickbait articles about how Tesla cars are Full Self-Driving now. Musk talks about the cars having the hardware NOW but he's been consistent that being able to sleep in your Tesla while it drives you is still in the future.

Now you are arguing in bad faith.

We have a history of Elon's statement. When someone keeps telling you, you will be able to sleep in your car in the next two years, every year since 2015. Don't you think the person who is being told that will believe that is already happening? and as he keeps saying it every year he gets new people who hasn't heard it before as well. You watched yesterday when he says he will have cars that are Level 5 can drive anywhere, any location, any weather conditions, without any restrictions in 2020 AND even said that people will not need to pay attention Q2 2020. Literally, and your response is to blame the media? seriously?

December 2015: "We're going to end up with complete autonomy, and I think we will have complete autonomy in approximately two years."

Elon Musk Says Tesla Vehicles Will Drive Themselves in Two Years

January 2016: "In ~2 years, summon should work anywhere connected by land & not blocked by borders, eg you're in LA and the car is in NY"

Elon Musk on Twitter

June 2016: "I think we are less than two years away from complete autonomy, safer than humans, but regulations should take at least another year," Musk said.

Two years until self-driving cars are on the road – is Elon Musk right?

March 2017: "I think that [you will be able to fall asleep in a tesla] is about two years" -

Transcript of "The future we're building -- and boring"

Nov 15, 2018: "Probably technically be able to [self deliver Teslas to customers doors] in about a year then its up to the regulators"

Elon Musk on Twitter

Jan 30 2019 "We need to be at 99.9999..% We need to be extremely reliable. When do we think it is safe for FSD, probably towards the end of this year then its up to the regulators when they will decide to approve that."

Tesla Q4 Earnings Call

Feb 19 2019 : "I think we will be feature complete full self driving this year. The car will be able to find you in a parking lot, pick you up, take you all the way to your destination without an intervention this year. I'm certain of that. It will be essentially safe to fall asleep and wake up at their destination towards the end of next year"

On the Road to Full Autonomy With Elon Musk — FYI Podcast

April 12th 2019 : "I think it will require detecting hands on wheel for at least six months.... I think this was all really going to be swept, I mean, the system is improving so much, so fast, that this is going to be a moot point very soon. No, in fact, I think it will become very, very quickly, maybe and towards the end this year, but I say, I'd be shocked if not next year, at the latest that having the person, having human intervene will decrease safety. DECREASE! (in response to human supervision and adding driver monitoring system)"

 

[Bladerskb]

According to Tesla, they will be feature complete later this year. Based on everything we know, I definitely think that will happen. In terms of how long until the march of 9's, that is harder to know. Personally, I think Musk is too optimistic about that part. However, I do think that the growth of the Tesla fleet and the increasing amount of data that Tesla can collect as a result, will help Tesla achieve that march of 9's faster. The big problem with the march of 9's is that there is just so many edge cases out there, probably millions and millions. If your fleet is small, you are just not going to do enough driving miles in enough different locations to see all those edge cases in a timely manner. That is the advantage that Tesla has. They have an every growing huge fleet of cars driving on lots of different roads in different areas. When you have hundreds of thousands of cars around the world all collecting data and sending it back to the machine for learning, like Tesla does, you are much more likely to encounter all those edge cases quicker. So I don't think it will take years and years for Tesla to complete the march of 9's.

Musk talks about the cars having the hardware NOW but he's been consistent that being able to sleep in your Tesla while it drives you is still in the future.

If we discount everything he has said till acouple days ago.

if we only take what he said in his autonomy day, Elon reiterated AGAIN that you won't need to pay attention and drivers can look out the window, Q2 2020. This is his very words and what he has been saying for 5 years. Yet you blame the media. You are literally arguing in bad faith.

 

[Guy V]

Comparison between human eyes and the car's cameras are not that meaningful. It's the processing behind them that's truly important. Human's possess a higher ordering thinking compared to a NN. NN is based on probability and approximation. At some level I'd say it's equivalent to human's first order processing. Human intuition is a lot less straight forward to model. Basically, when you're driving you 'know' not to drive into the car in front of you or drive off the road, through a stop sign, into a deer, or into a barrier. The NN 'learns' that humans don't do this, so it doesn't do it either. But if all of a sudden you hit a puddle and your entire car gets covered in mud and you can't see anything, as a human you can use all the prior knowledge of driving, the immediate past, etc to decide of your best course of action. It's likely a situation you've never encountered but your brain's higher order thinking can figure out what to do. The NN will be completely lost and (at least in the current implementations) it will just force the driver to take over.

I'd like to see these 'unknown' situations applied to the FSD and see how it reacts.

That most often doesn't work out too well for us as we crash about 6 million times per year in the US alone. I am really expecting technology to be able to do a much much better job, starting soon and improving quickly.

 

[GeoX750]

Here is a testimony from Morgan Stanley:

Overall, pretty positive. The reference to "no LIDAR" is key. Tesla is achieving this without LIDAR. But we know FSD is far from perfect since it is not feature complete yet. So the one disengagement is not surprising. FSD will get even better. FSD is making progress and seems very promising to me.

We need more of the reports from the test drives. Where is the rest of this report and are there others?

 

[Bladerskb]

Here is a testimony from Morgan Stanley:

Overall, pretty positive. The reference to "no LIDAR" is key. Tesla is achieving this without LIDAR. But we know FSD is far from perfect since it is not feature complete yet. So the one disengagement is not surprising. FSD will get even better. FSD is making progress and seems very promising to me.

Here's a car driving without disengagement using late 2017 production NN, does that mean level 5 in acouple months which would be 2019 seeing the video is from 2018 according to Elon's logic?

 

[verygreen]

According to Tesla, they will be feature complete later this year

According to Tesla they were feature complete and just undergoing final validations in Oct 2016.
According to Tesla fully autonomous Cost to Coast run would happen by the end of 2017.
I am sure @Bladerskb has a bunch of additional examples.

The "according to Tesla" argument that implies their forward-looking statements mean anything other than "it's very unlikely, but there's a slight chance never the less" need to die.

 

[electronblue]

Here is a testimony from Morgan Stanley:

Overall, pretty positive. The reference to "no LIDAR" is key. Tesla is achieving this without LIDAR. But we know FSD is far from perfect since it is not feature complete yet. So the one disengagement is not surprising. FSD will get even better. FSD is making progress and seems very promising to me.

I disagree Lidar being key at all. Lidar is a redunancy element, it does not worry me in the least at this stage of Tesla’s development. Least of all in fair weather demo like this.

What Tesla needs is reliable vision. MobilEye and many others have done great demos with vision only. Tesla’s vision so far has not been so reliable and that is their first order of business. It is good to see progress as far as this demo goes.

The question would Lidar be needed comes much later when the redundancy questions rears its head. Before that reliable vision is needed. Once that is solved we can see where its limits, redundancy-wise- are.

 

[Atari2600]

Here's a car driving without disengagement using late 2017 production NN, does that mean level 5 in acouple months which would be 2019 seeing the video is from 2018 according to Elon's logic?

I guess if you let the Mobileye car blow through a red light it's not a disengagement is it?

 

[diplomat33]

Here's a car driving without disengagement using late 2017 production NN, does that mean level 5 in acouple months which would be 2019 seeing the video is from 2018 according to Elon's logic?

Why do you keep doing this? You keep posting demos of other companies like Mobileye to show that they have better FSD than Tesla. Yes, there are companies who have better FSD than Tesla. I don't care if other companies have something better than Tesla, I care about Tesla's progress and how they are doing. For one, I am a Tesla owner so I am interested in what features my car will have. I don't own a car with a Mobileye system so the fact that Mobileye is better, does not affect me. And frankly, a lot of these systems that are better than Tesla, won't ever make into a production car that I can actually buy so the fact that they are better is completely irrelevant.

 

[Bladerskb]

According to Tesla they were feature complete and just undergoing final validations in Oct 2016.
According to Tesla fully autonomous Cost to Coast run would happen by the end of 2017.
I am sure @Bladerskb has a bunch of additional examples.

The "according to Tesla" argument that implies their forward-looking statements mean anything other than "it's very unlikely, but there's a slight chance never the less" need to die.

Why do you keep doing this? You keep posting demos of other companies like Mobileye to show that they have better FSD than Tesla. Yes, there are companies who have better FSD than Tesla. I don't care if other companies have something better than Tesla, I care about Tesla's progress and how they are doing. For one, I am a Tesla owner so I am interested in what features my car will have. I don't own a car with a Mobileye system so the fact that Mobileye is better, does not affect me. And frankly, a lot of these systems that are better than Tesla, won't ever make into a production car that I can actually buy so the fact that they are better is completely irrelevant.

Its not irrelevant, its relevant to whatever Tesla is claiming which you keep avoiding. If other have better FSD than Elon and are not saying what Elon is saying. What does that say about what Elon is saying and have said for the last 5 years?

Elon is claiming Level 5 which i have pointed out multiple times over and over and over again and no one believed me.

Now we have heard it from the horses mouth. He is claiming level 5 in literally 6-12 months which he said will work anywhere, in any location, in any weather condition with absolutely no restriction, with no human driver.

Again you refuse to confront that, elon hasn't changed what he has said the last 5 years. He has double down and triple down. In response to it you blame the media. You have shown your true colors that you will never hold Elon on the irresponsible statements that he makes.