Waymo’s “commercial” ride-hailing service is... not yet what I hoped

[CarlK]

It is ready for all roads in the geofenced area. Ofcourse what gonna be publicly displayed is the road that is most tested on until they are ready for production.

Who told you that they are "ready" for all roads in the geofenced area? Everyone has a plan to run in some small geofenced areas but all the "demos" we've seen were run on a much smaller stretch of roads. No one has shown that you can take its car and let it to run wild in their designated geofenced areas. Not to mention many of those what we've seen were edited videos with no third party observers on board. That's not gaming the system. That's border line cheating if you're trying to use those to prove capabilities of the system. Maybe too harsh words but these are no different than TV commercials.

Again there's no gaming its simply logistics. Elon gaming is simply wrong. Their system is simply incapable of going 3,000 miles (cross country) without requiring intervention. While you can define a static route to take. You can't game road users and objects. Your system still have to handle it.

True capability is when you can buy a car off the lot and do what it's supposed to be able to do. Only Tesla is offering you that now and likely in the foreseeable future. Elon could easily put a fleet of cars to run only that route for a few hundreds times to get the car perfectly trained on that route. That's not in the spirit of how this thing should be done, that is to prove (to the public and to themselves) the capability of car autonomously driven across the country. And that is exactly gaming it.

 

[Bladerskb]

True capability is when you can buy a car off the lot and do what it's supposed to be able to do. Only Tesla is offering you that now and likely in the foreseeable future.

Where can I buy this car so I can have it self drive to me?

Elon could easily put a fleet of cars to run only that route for a few hundreds times to get the car perfectly trained on that route.

Your car DOESNT learn. Your car isn't trained by having it drive a particular route.
You still hold onto myths that have been debunked years ago. Where have you been? Are you trolling?

 

[CarlK]

Where can I buy this car so I can have it self drive to me?

Everything Tesla says its car could do you can do it with yours (that is if you do own a Tesla). Eveything Mobileye says its car does you can, err, watch that presentation or (edited) video again.

Your car DOESNT learn. Your car isn't trained by having it drive a particular route.
You still hold onto myths that have been debunked years ago. Where have you been? Are you trolling?

Such an ignorant comment. That's exactly how machine learning works. Do the same thing over and over again and improve the statistics with each trial. It's much easier to learn one route with a thousand trials than to spread that to thousands of miles everywhere.

 

[acoste]

... Also there were never intentions to make it a joint venture or even let anyone to know what's going on there.

I'm pretty confident about the GM story. And better believe my sources than your statements since you seem to be heavily influenced by Elon.

Only Tesla is working on "general autonomous driving" that it will eventually let you to buy a car that you can have it to drive itself anywhere in the country....

I find it very naive on your side that you think only Tesla is shooting for a general autonomous driving. Everyone targets that, but step by step. I already showed Audi. BMW does the same. My bet is that the Daimler-Audi-BMW-Mobileye-Bosch-Delphi team will bring up a level 4 system covering the entire Germany then expand in Europe sooner than Tesla will cover US. I know it's harder to cover US. Thinking about "general" Tesla doesn't shoot for all weather for example. Once the front radar is covered by snow, the system stops. Audi has a heater there at least. But all of the solutions are very far from solving the weather problem. I don't think there will be a general solution by 2030 that can cover Russia for example.

Given Tesla's twisted marketing phrases it is possible that they are going to say first that they solved the problem. However they are not going to be the first who has a !reliable and working system!.

 

[Bladerskb]

Everything Tesla says its car could do you can do it with yours (that is if you do own a Tesla). Eveything Mobileye says its car does you can, err, watch that presentation or (edited) video again.

Really... is that why the media is calling him out?

Tesla and Uber have faced more visible setbacks. While Tesla CEO Elon Musk has said the automaker will likely be able to launch an autonomous ride-hailing service by the end of 2019, pending regulatory approval, he also said in 2015 that Tesla would have fully-autonomous vehicle technology ready in about two years. The automaker has not yet said that its vehicles are capable of full autonomy. Tesla has also missed multiple deadlines set by Musk to send a self-driving vehicle across the US.

Tesla did not respond to a request for comment.

Abuelsamid and Kelly said the autonomous driving industry has done a poor job of communicating the progress of its technology in a realistic manner. Abuelsamid highlighted Tesla as the worst offender.

"I would put Tesla at the worst end of the scale in terms of managing expectations," he said.

Musk has developed a reputation for setting aggressive goals and making bold proclamations about Tesla's products. He told Recode in early November that Tesla was far ahead of its competitors in developing autonomous driving technology that can be used in all circumstances, and Tesla has since 2016 sold a hardware upgrade it says will give vehicles full self-driving capabilities when the necessary software is ready. The upgrade was removed from the automaker's website in October, but Musk has said it would remain available to those who ask for it.

Abuelsamid said Tesla's autonomous driving technology lags behind its competitors'.
"They have the least capable technology of any company," he said.

Unlike many of its competitors, Tesla's autonomous driving hardware includes only cameras and radar, forgoing the lidar sensors (which emit pulses of light that bounce off objects to determine where they're located) that some companies and experts say are necessary for vehicles to drive without human assistance. A self-driving Tesla vehicle would be vulnerable to weather or road conditions that could cover its cameras or radar and render them ineffective, Abuelsamid said.

"The hardware that they have on the vehicles today will never, ever be Level-5 capable," he said, referring to the designation for vehicles that can drive without human assistance under any conditions.

Outline - Read & annotate without distractions

Such an ignorant comment. That's exactly how machine learning works. Do the same thing over and over again and improve the statistics with each trial. It's much easier to learn one route with a thousand trials than to spread that to thousands of miles everywhere.

Your car doesn't do that, it doesn't get better each time it drives a route. Your car simply does inference on the NN it already has and uses traditional programming for control. It doesn't learn, it doesn't get better. @verygreen can you set this guy straight, i'm tired of talking to him about the same thing for the past week. he won't listen.

 

[CarlK]

I'm pretty confident about the GM story. And better believe my sources than your statements since you seem to be heavily influenced by Elon.



I find it very naive on your side that you think only Tesla is shooting for a general autonomous driving. Everyone targets that, but step by step. I already showed Audi. BMW does the same. My bet is that the Daimler-Audi-BMW-Mobileye-Bosch-Delphi team will bring up a level 4 system covering the entire Germany then expand in Europe sooner than Tesla will cover US. I know it's harder to cover US. Thinking about "general" Tesla doesn't shoot for all weather for example. Once the front radar is covered by snow, the system stops. Audi has a heater there at least. But all of the solutions are very far from solving the weather problem. I don't think there will be a general solution by 2030 that can cover Russia for example.

Given Tesla's twisted marketing phrases it is possible that they are going to say first that they solved the problem. However they are not going to be the first who has a !reliable and working system!.

I'm really puzzled by your comments especially if it's from someone who says he's familiar with the industry. My guess is you're familiar with the auto industry but not the tech industry. Your definition of "general" autonomous driving is very different from that of people use in this industry. No one else has hundreds thousands of cars on the road, now or in the foreseeable future, to acquire machine learning data covering every road on the map whether it's Europe or the US. Tell me how are they going to make it (general autonomous driving) to happen regardless of what hardware they put in the car.

Even Waymo (and probably few others that are copying Waymo) which has been using simulation to supplement real world data is changing direction and buying more test cars now. You can never get the same result from those few simple simulation or some practice runs on limited roads. Machine learning basics that many (you and many old time auto companies) still have no clue off. When any of those companies put tens or thousands of cars on the road then we can argue about who's ahead again. At this moment, no matter what Tesla could or could not do yet, it has no competition is this area.

Daimler, Audi, BMW or any of those Detroit companies all can say what they want to do but those companies have never been very good at computers and technologies. None of them could even get OTA out years after Tesla showed them how great that is. Hard to see how could they catch up those Silicon Valley companies Waymo and Tesla/SpaceX that have all those top tech talents in the world. Not to mention company run by technology oriented CEO instead of bean counters with good marketing departments.

 

[CarlK]

Your car doesn't do that, it doesn't get better each time it drives a route. Your car simply does inference on the NN it already has and uses traditional programming for control. It doesn't learn, it doesn't get better. @verygreen can you set this guy straight, i'm tired of talking to him about the same thing for the past week. he won't listen.

Not sure why this is so hard for you to understand. Anyway it has just proved (not that I did not know it already) that you are totally clueless of the subject you want to give tons of opinions of. Machine learning is nothing more than using statistics and big data through its algorithm to general actions. The more data it collects the better statistics and more accurate actions.

Maybe this simpler example could help you out to understand it more. The Google AlphaGo started as a pretty ordinary player and improved with each game it played. Eventually it was able to beat the best go player in the world with the ML algorithm written by programmers who were just mediocre players or may not even play the game. That's the amazing, and some say scary, part of machine learning.

BTW who is this @verygreen person and what he/she has anything to do with this?

 

[acoste]

Daimler, Audi, BMW or any of those Detroit companies all can say what they want to do but those companies have never been very good at computers and technologies. None of them could even get OTA out years after Tesla showed them how great that is. Hard to see how could they catch up those Silicon Valley companies Waymo and Tesla/SpaceX that have all those top tech talents in the world. Not to mention company run by technology oriented CEO instead of bean counters with good marketing departments.

Daimler and Audi are Silicon Valley companies and have autonomous vehicle testing permits in California since 2014.

 

[CarlK]

Daimler and Audi are Silicon Valley companies and have autonomous vehicle testing permits in California since 2014.

Yeah right. Almost every auto company from all continents has a small operation in the Silicon Valley. That does not make any of them a Silicon Valley company. You can't just hire a few engineers here and becoming a "Silicon Valley company". A Silicon Valley company needs to have that special culture, technology above everything else and risk taking, from the top guy down to every level in the company. True talents will not go work for anything but real Silicon Valley companies either. If you're a top talent in computer or AI would you want to take orders from executives back in Stuttgart or Detroit who have background in mechanical engineer or worse accounting or marketing with little idea of what you're doing (like a few who keep posting here)?

 

[electronblue]

Something @CarlK that occurred to me.

I feel the difference you and some others like me have is that you believe Tesla trains their NN with consumer cars. I do not believe that or expect that to happen.

I believe Tesla and incidentally MobilEye both use deep learning and training NNs to different extents but both train their NNs locally with engineering cars (and simulators) driven by test engineers. Not consumer cars.

Likely even Waymo keeps their NN training separate from their fleet though they could of course set it up differently as they have unlimited bandwith with cars that return to base every night.

Consumer car fleet can certainly be used for validation triggers and HD mapping — Tesla does the former and MobilEye with EyeQ4 the latter — but I do not believe anyone actually trains NNs using consumer car fleets and consumer driver inputs.

 

[Bladerskb]

I'm pretty confident about the GM story. And better believe my sources than your statements since you seem to be heavily influenced by Elon.



I find it very naive on your side that you think only Tesla is shooting for a general autonomous driving. Everyone targets that, but step by step. I already showed Audi. BMW does the same. My bet is that the Daimler-Audi-BMW-Mobileye-Bosch-Delphi team will bring up a level 4 system covering the entire Germany then expand in Europe sooner than Tesla will cover US. I know it's harder to cover US. Thinking about "general" Tesla doesn't shoot for all weather for example. Once the front radar is covered by snow, the system stops. Audi has a heater there at least. But all of the solutions are very far from solving the weather problem. I don't think there will be a general solution by 2030 that can cover Russia for example.

Given Tesla's twisted marketing phrases it is possible that they are going to say first that they solved the problem. However they are not going to be the first who has a !reliable and working system!.

Some of these ppl are fully entrenched in the koolaid they bleed green and yellow. They ignore everything tesla has right now and think of this dystopian future. While AP2 can't even beat Supercruise in lane keeping, and AP2 is barely better than AP1 (alot of people prefer AP1). Yet Elon is on the verge of allowing you to sleep in your car. Most Tesla Fans are not rational. Very few are, you will find them in this forum. Others don't apply logic to anything that deals with Tesla.

Elon literally said the code that was supposed to do the cross country IS the NOA. And we know how bad it is. but they will ignore it and talk about how many cars tesla has. Now there is this whole "THANOS... *cough* HW3 IS COMING!"

But the actual fact is that the AP engineers already have HW3 cars. If they had software to do cross-country, it would had been done now. Elon is the biggest bullsh*tter of the century.

First it was that the demo will happen by end of 2017. When end of 2017 came and nothing, he said it will happen very early of 2018. When that came and nothing, then he said in 3-6 months, when that came and nothing. Then he said in 3-6 more months. When that came and nothing, then he said its gonna happen on V10 Alpha guys, when HW3 launches.

Its complete bullsh*t, only a puppet will allow themselves to believe that horse crap.

Here's what Eletrek had to say about Elons BS last year.

The way I see it, it sounds like Tesla could enable level 4 autonomous driving over the next year under the “fully self-driving” option offered in the Model S, Model X, and Model 3.

Then, Tesla could gradually improve on the level 4 capabilities to eventually reach level 5 another year after level 4, which would be in line with Musk’s new timeline.

And he was right, based on elon's BS, that's how the timeline would go. But we are at the end of 2018 and where is L4 self driving? Lol anyone who actually believe that BS must also believe i have a miniaturized fully working fusion reactor in my house. Elon's predictions are not based on any kind of fact or evidence. You can ask elon to predict something morning, afternoon, night and he will give you three different predictions. AP engineers know how far they are, if you ask them on linkenedin, are you close to Level 5 next year, they would laugh. Which is probably how theinformation got their contacts. Therefore elon isn't getting his predictions from internal timeline but is purely bs pr to pump the stocks and sell cars.

Infact Mobileye's Amnon confronted Elon musk to his face while his AP engineers were scared to talk to him and this is how it went down.

"To make sure Tesla didn’t oversell the system’s capabilities, Shashua visited Musk at his factory in Fremont, Calif., and urged him to make sure the system would force customers to keep their hands on the wheel. “His whole engineering team was waiting for me because they couldn’t talk to him,” Shashua says, suggesting that Musk had been unpersuadable on the issue. “So I went there, I had the meeting, and I convinced him. He promised it wouldn’t be hands-free.” Two months later, in late 2015, Musk unveiled Tesla Autopilot."

 

[Bladerskb]

Waymo CEO Interview with WSJ
Listen to podcasts via our app. Get features like custom playlists, personalized recommendations & offline listening. | Stitcher

Look at the difference between tone versus the stuff that Elon puts out.

 

[CarlK]

@Bladerskb

Those are Mobileyes' spins to save the company and the stock price when it separated from Tesla's AP program. You just knowingly took the hook, line and sinker. Tesla has been long known to be developing it's own system and wanted to get rid of Mobileye. It was revealed in 2015 when George Hotz (Geohot with the reputation of being the first person who hacked iPhone when he was a teenager) in an interview that Elon was recruiting him and promised millions of dollars of bonuses when it gets rid of Mobileye. Mobileye sure have read that same news and knew Tesla's intention too. That accident just provided it an

Here is what George Hotz said “But to be honest, the Mobileye system is so easy to reproduce; you don’t need the best talent. I did it in a couple of months.”
In the wake of a fatal crash, Tesla will quit using Mobileye's chips for Autopilot vision

Another interesting comment: " — all the losers are trying to group up together,”
That's exactly what we're still seeing today. Why would you want to share with others if you are confident of your technology? Like if you want to believe those who are trying to sell you sure ways to make money on TV. Either you or them are the biggest fools.

 

[verygreen]

Everything Tesla says its car could do you can do it with yours

nope. Refer to the FSD video as at least one example, my car cannot do that. And the "Extensive validation" blah blah was a lie too. They changed codebase like at least 3x since then.

BTW who is this @verygreen person and what he/she has anything to do with this?

I am a _very_ curious person. So curious you won't believe it. So as a part of it, I take things apart to see how they work. This includes Tesla cars. So I happen to know a bunch of things about the inner workings that very few other people that are not under NDAs know. My knowledge is not perfect, but it's decent enough, I guess. And no, the Tesla cars (the kind that regular people own) don't "learn" anything as you drive them if you don't include installing firmware updates as "learning".

But the actual fact is that the AP engineers already have HW3 cars

Nah, from what I have seen, hw3 is in it's nascency, being barely brought up to actually boot ~ a month ago. In other words I don't think there are even mid-double digits of those hw3 cars that are actually fully functional yet. Ah, also you know what they call that hw3 thingie? "turbo"

 

[CarlK]

Something @CarlK that occurred to me.

I feel the difference you and some others like me have is that you believe Tesla trains their NN with consumer cars. I do not believe that or expect that to happen.

I believe Tesla and incidentally MobilEye both use deep learning and training NNs to different extents but both train their NNs locally with engineering cars (and simulators) driven by test engineers. Not consumer cars.

Likely even Waymo keeps their NN training separate from their fleet though they could of course set it up differently as they have unlimited bandwith with cars that return to base every night.

Consumer car fleet can certainly be used for validation triggers and HD mapping — Tesla does the former and MobilEye with EyeQ4 the latter — but I do not believe anyone actually trains NNs using consumer car fleets and consumer driver inputs.

Hope this can give you a clearer view of how Tesla does this.

Tesla’s massive accumulation of Autopilot miles

New Tesla Autopilot To Operate In "Shadow Mode" Until Self-Driving System Is Ready To Activate

 

[electronblue]

Hope this can give you a clearer view of how Tesla does this.

Tesla’s massive accumulation of Autopilot miles

New Tesla Autopilot To Operate In "Shadow Mode" Until Self-Driving System Is Ready To Activate

I know these stories I just don’t think it means the same thing as you think it means.

I do not think Tesla trains NNs on consumer cars or has any plans to do so in the future. They do have validation and information gathering triggers and can certainly expand on that in consumer cars but that is not training NNs.

Consumer fleet in my view at best helps Tesla test their system not train it. Training NNs in my view is a very specific discipline and done on engineering cars and simulators. I understand you feel different.

 

[verygreen]

Hope this can give you a clearer view of how Tesla does this.

Tesla’s massive accumulation of Autopilot miles

This is a wishful thinking we've discussed at length in the MIT study thread. There's no "accumulation of miles" whatever people mean by that.

New Tesla Autopilot To Operate In "Shadow Mode" Until Self-Driving System Is Ready To Activate

The shadow mode is mostly a dream, product of Elon's vague wording combined with misunderstanding on the part of listeners. This was also discussed at length in various other topics.

 

[verygreen]

Training NNs in my view is a very specific discipline and done on engineering cars

I'd be surprised if they actually train NNs on engineering cars as well. It's not like they have powerhouse computing in those cars... They mostly gather data and debug on those. And even more often they just have "benchtop" units that they feed replayed data into to do some development/debugging and perhaps one of those days they'd have simulators that can feed data into the units too?

 

[electronblue]

I'd be surprised if they actually train NNs on engineering cars as well. It's not like they have powerhouse computing in those cars... They mostly gather data and debug on those. And even more often they just have "benchtop" units that they feed replayed data into to do some development/debugging and perhaps one of those days they'd have simulators that can feed data into the units too?

Useful correction. What I meant gathering data that trains the NNs is done on engineering cars not on consumer cars.

 

[CarlK]

nope. Refer to the FSD video as at least one example, my car cannot do that. And the "Extensive validation" blah blah was a lie too. They changed codebase like at least 3x since then.

You're talking about the demo video? No demo video has any value but that's all everyone else could have. I'm talk about what Tesla says the car it sold to customers could.

I am a _very_ curious person. So curious you won't believe it. So as a part of it, I take things apart to see how they work. This includes Tesla cars. So I happen to know a bunch of things about the inner workings that very few other people that are not under NDAs know. My knowledge is not perfect, but it's decent enough, I guess. And no, the Tesla cars (the kind that regular people own) don't "learn" anything as you drive them if you don't include installing firmware updates as "learning".

I bet you I know more than you if I had not signed any NDA's. Anyway just take a look at the two links I posted above. ML is such a simple concet I have no idea why it's so hard for some to fathom.

Nah, from what I have seen, hw3 is in it's nascency, being barely brought up to actually boot ~ a month ago. In other words I don't think there are even mid-double digits of those hw3 cars that are actually fully functional yet. Ah, also you know what they call that hw3 thingie? "turbo"

Tesla hired Jim Keller and the whole AMD team in 2015. What do you think Tesla would want people of such high caliber do? AI chips that will go into the HW3 board of course. The new specialized chip is said to be 10x faster than the Nvidia chip Tesla is using now for doing the task. Other than probably Waymo no one even has AI chip as good as the current Nvidia chip. Definitely not the Intel/Mobilieye consortium.