Perhaps Mobileye should have been given more credit , and less focus on autopilot vs EV

[eye.surgeon]

With the less than stellar performance thus far of Tesla's own AP2, one has to wonder if perhaps Mobileye was not credited sufficiently for the success of AP1. Certainly in hindsight Mobileye's product seems much better than what Tesla has developed thus far on their own.

I also wonder if perhaps autopilot is a distraction that will serve to delay the acceptance of EVs. After all, Tesla mission statement is to advance EVs, not to advance self driving cars. It seems like AP is consuming a tremendous amount of engineering and resources that could perhaps be better served developing the drivetrain and battery.

Thoughts?

 

[FlatSix911]

Tesla should have hired this guy ... Tesla made it difficult for hacker geohot to buy a Model S and use his own software to power Autopilot 2.0

If you think that the rivalry between George ‘geohot’ Hotz’s small autonomous driving startup comma.ai and Tesla is only in the hacker’s head, you might be wrong. When his self-driving car/machine learning startup ‘comma.ai’ came out of stealth mode, Hotz released an email conversation he had with Tesla CEO Elon Musk in which he was offered a contract with a “multimillion-dollar bonus” for him to build a new Autopilot system in order for Tesla to discontinue Mobileye’s part in the program.

It prompted a response from Tesla in which I’m just now realizing the company ended up lying about misrepresenting its future plans for Autopilot. Tesla defended Mobileye since Musk’s conversation with Hotz made it sound like Tesla wanted to get rid of Mobileye.

 

[stopcrazypp]

With the less than stellar performance thus far of Tesla's own AP2, one has to wonder if perhaps Mobileye was not credited sufficiently for the success of AP1. Certainly in hindsight Mobileye's product seems much better than what Tesla has developed thus far on their own.

I also wonder if perhaps autopilot is a distraction that will serve to delay the acceptance of EVs. After all, Tesla mission statement is to advance EVs, not to advance self driving cars. It seems like AP is consuming a tremendous amount of engineering and resources that could perhaps be better served developing the drivetrain and battery.

Thoughts?

Putting aside all the arguments about specific promised dates (which is a whole other subject), HW2 is about 4-5 months in after release. I'm pretty sure HW1 was not anywhere near the stage AP2 is now 4-5 months in (March 2015). Of course you have to factor in a lot of the things Tesla learned in developing AP1, but I don't think Tesla is as far behind as people put it.

As for Mobileye not getting enough credit, I'm not sure I necessarily agree. Can you name another Mobileye vehicle that has anywhere near the capabilities that AP1 has? I can't think of an example. Tesla has pushed the Mobileye chip far beyond what anyone else would have thought possible.

As for autopilot being a distraction, I'm not sure I agree at this point anymore. Given how well known it is now, it seems like it is a major defining feature for Tesla that is just as important as being an EV.

 

[stopcrazypp]

Tesla should have hired this guy ... Tesla made it difficult for hacker geohot to buy a Model S and use his own software to power Autopilot 2.0

If you think that the rivalry between George ‘geohot’ Hotz’s small autonomous driving startup comma.ai and Tesla is only in the hacker’s head, you might be wrong. When his self-driving car/machine learning startup ‘comma.ai’ came out of stealth mode, Hotz released an email conversation he had with Tesla CEO Elon Musk in which he was offered a contract with a “multimillion-dollar bonus” for him to build a new Autopilot system in order for Tesla to discontinue Mobileye’s part in the program.

It prompted a response from Tesla in which I’m just now realizing the company ended up lying about misrepresenting its future plans for Autopilot. Tesla defended Mobileye since Musk’s conversation with Hotz made it sound like Tesla wanted to get rid of Mobileye.

After seeing his handling of his comma.ai system, I think it was definitely for the better that Tesla did not hire the guy. He has a very careless view for safety: dropping his system as soon as NHTSA asked him to submit information demonstrating his system is safe, an examination of his code showing poor choice in picking Python for a realtime system, plus poor error handling.
After mothballing Comma One, George Hotz releases free autonomous car software

 

[Tam]

...less than stellar performance thus far of Tesla's own AP2...

MobilEye system was a distraction with very slow moving progress and dead end technology for your hardware.

It wanted you to buy an intermediary hardware if you want to prevent the Florida crash--Lateral Turn Across Path (LTAP) detection capabilities, not in 2016 but 2 years later in 2018.

After that how many more that you have to trade in your cars to get upgraded MobilEye hardware before you can get to Driverless function?

I wouldn't worry of subpar AP2 in the beginning because that is what you expect an AI artificial intelligent system to do.

For example an AI system can lose many many chess games in the beginning but it will eventually learn to win on its own.

...autopilot is a distraction...

A goal for safety is not a distraction, it is a necessity.

 

[malcolm]

The basic false assumption which many people (including Mobileye with its 5 Series, 6 Series and Shield+ systems) seem to be making is that because humans accomplish so many tasks using the high resolution of visible wavelengths, then that must be the minimum requirement for all tasks.

So robot vision must make use of the resolution which 400nm to 700nm provides.

Granted, if you want precise hand/eye coordination or recognise alphanumeric characters on a page or on a street sign at the side of the road, then visible wavelengths are the way to go, but owners have been shocked to discover that at the present state of the software, they can tape up almost all of the car's cameras and many of the car's functions are not impaired.

But is high resolution vision needed to drive a car?

If you suffer from myopia, try this experiment: slide your glasses down your nose a little and drive a few hundred yards on the freeway looking over the top of them instead of through them.

Are you still alive? Of course you are. Other cars and trucks are big blocky things. You may not be able to read numberplates or the make and model, but you can still pilot your vehicle safely. You only need fine detail for things like lane markings (where one good eye/camera will do).

Most of the time freeway driving can be accomplished just fine with blurry (i.e. lower resolution) human vision. In robot terms this is equivalent to radar and ultrasonics (if you want to block those, you'd have to tape metal plates or special acoustic tiles over them)

So why do Teslas spend ages "calibrating" the cameras? What is there to calibrate? I imagine that it's matching the hi-res camera image to the low-res radar "image", which can be stopped if the camera is installed tilting too high or too low.

So why do the cars have eight cameras? Because they will be needed/used in the future for true autonomous driving - where no one is in the vehicle*

*legislation required

 

[JoaoD]

With the less than stellar performance thus far of Tesla's own AP2, one has to wonder if perhaps Mobileye was not credited sufficiently for the success of AP1.

Then can you tell us one system that uses Mobileye that is as good or better than AP1?

 

[virtualsmack]

tesla needs something like AP to differentiate from other manufacturers. Good foresight on their part to realize that someone else will eventually come up w an EV that meets and even surpases their initial proposition of range + luxury but Tesla will be differentiated with their level of AP2.

It's incredibly hard for companies to reinvent themselves while they are experiencing growth and success. Usually they double down on whatever it is that makes them successful so it's rare and smart for Tesla to do this, IMO.

As for mobileye and AP2, people will pay for efficiency vs convenience. There is huge/more $ in AP for mass transportation and shipping/deliveries so if Mobileye's technology isnt capable of fully autonomous driving and delivering on that efficiency down the road, then ditching them wasn't a bad idea. Personally i'm not judging AP based on its performance in cars today but based on eventual commercial applications.

 

[googlepeakoil]

Tesla executives probably want to push AP as it provides the largest gain of revenue for next to zero incremental cost. All cars will have the hardware. Even the ones that don't pay for the self-driving features. Tesla need to massively proove that it provides a benefit, it is safe to get buyers to pay for it. Versus all other "upgrades" it's the least incremental cost (it is effectively zero to Tesla to deploy to 1million people vs 100 people). So that's 1million cars x circa $8k. = $8 billion of possible revenue if they could "convince" everyone to buy all the AP upgrades. Personally i'm erring on the side of 90% sure I won't buy any self-driving feature. I'm one of the skeptics that needs convincing - and can't see the benefit of AP to me at anything like the price Tesla want for it.
To cut to the chase... there's several fatal crashes caused by people's over-confidence in the self-driving abilities. People that might be alive if the car was just an electric car. They wouldn't have hit a lorry, driven into a street sweeper on a smoggy Chinese motorway.
I don't see - while there's other road-users a car ever being self-driving, or 100% safe, or close to an attentive human.

 

[Gouldness]

Agreeing more and more on the Mobileye credit front. The more i look into the Mobileye 560 the more I see how incredibly overpriced AP is for current CPO prices (a $20k delta). I know it doesn't have TACC but dang, that's a pricey upsell.

Personally I'd be willing to pay a few grand to have a steering column motor installed along with a delphi radar unit and use Comma.ai's Openpilot machine learning.

 

[AnxietyRanger]

It prompted a response from Tesla in which I’m just now realizing the company ended up lying about misrepresenting its future plans for Autopilot. Tesla defended Mobileye since Musk’s conversation with Hotz made it sound like Tesla wanted to get rid of Mobileye.

Tesla certainly admirably used smoke and mirrors to hide their AutoPilot direction. A lot of people were confident they were firmly in bed with Mobileye. I remember the ridicule I got for suggesting already back in 2015 that Tesla might go with Drive PX (AP2 is based on PX2):
Model X mule(s) show signs of nVidia Tegra X1 Drive PX platform - no rear mirror!

Well, the future turned out different than Tesla was saying, but in every other way it was not really a surprise at all. I presented a lot of points in that thread supporting the speculation that Drive PX might be in the plans (e.g. the talks with Nvidia in public).

That said, in fairness to Mobileye, they do have a good image recognition piece of hardware and software. Much better than most manufacturers make use of at this time. Their roadmap of course is more conservative than Tesla's aggressive method. Mobileye is proceeding more slowly and carefully and in iterations. We shall see which method ends up being better, Tesla's aggressive one or Mobileye's conservative step-by-step...

TGranted, if you want precise hand/eye coordination or recognise alphanumeric characters on a page or on a street sign at the side of the road, then visible wavelengths are the way to go, but owners have been shocked to discover that at the present state of the software, they can tape up almost all of the car's cameras and many of the car's functions are not impaired.

You are not serious, are you? In suggesting that the tape-up experiment tells us anything about non-visible wavelength? What it tells us is that the current EAP only uses a single camera as it is basically aiming for AP1 parity at this stage. Taping over those other cameras just means the system is not using them yet at all.

I agree computers can use a wider range of wavelengths (and Elon Musk has suggested as much in this postings), but taping over the cameras has got nothing to do with that. It was merely evidence of the fact that EAP currently only uses the front camera (the second taped camera causing problems apparenly was just an issue with its tape blocking partially the main camera)

 

[malcolm]

You are not serious, are you? In suggesting that the tape-up experiment tells us anything about non-visible wavelength?

Seems to, yeah. Tape the cameras up andstuff still works. Something's doing it and it doesn't appear to be light

 

[AnxietyRanger]

Seems to, yeah. Tape the cameras up andstuff still works. Something's doing it and it doesn't appear to be light

Yes. Radar and navigation data is doing it by the looks of it?

I agree the cameras can see more wavelengths than a human eye but the taped up example does not sound very relevant to me. Darkness, some inclement weather scenarios, sure...

Taped up I'd say it is more about non-camera redundancy than anything. That is also one thing why I worry about Tesla's limited radar coverage compared to competition.

We shall see, of course.

 

[Uncle Paul]

I do not go along with the theory that because some features still work with only 1 camera functional, that the other cameras are dormant.

I can tape up one of my eyes and still drive down the road, parallel park, keep distance from the cars around me and maintain a proper speed, however I can function much much better with both eyes functioning.

I have faith that Tesla is doing the right thing for us. It is in their best interests to provide a well functioning automatic driving function that interfaces well with an active and aware driver. Believe that they are working hard to release additional functions as soon as it is prudent and safe to do so.

Just as the well taught robots can help out a ton at the Tesla factory, they are always monitored and controlled by nearby people.

 

[Shock-On-T]

With the less than stellar performance thus far of Tesla's own AP2, one has to wonder if perhaps Mobileye was not credited sufficiently for the success of AP1. Certainly in hindsight Mobileye's product seems much better than what Tesla has developed thus far on their own.

I also wonder if perhaps autopilot is a distraction that will serve to delay the acceptance of EVs. After all, Tesla mission statement is to advance EVs, not to advance self driving cars. It seems like AP is consuming a tremendous amount of engineering and resources that could perhaps be better served developing the drivetrain and battery.

Thoughts?

Completely agree.

There are so many players in autonomous driving, it does not need to be done by Tesla. Just buy the best system out there (like whatever MobileEye is bringing out next).

Tesla cars were cool long before AP1.

 

[AnxietyRanger]

I do not go along with the theory that because some features still work with only 1 camera functional, that the other cameras are dormant.

In this case we also have sources saying Tesla is currently using only one camera for EAP. Both Electrek's source(s) and a knowledgeable poster on TMC that has been right so far, despite strong opposition (reminds me of how Eds was received).

Of course, eventually they will use four for EAP.

 

[paulch]

Autonomous driving is important for Tesla car sharing network, car sharing can greatly increase the utilization of a car, thus less number of cars need to be built to replace the ICE fleet. So this is aligned with Tesla's core mission.

Second, Tesla simply wants to the build the best car to make it impossible for ICE car makers to compete.

 

[xborg]

Tesla should have hired this guy ... Tesla made it difficult for hacker geohot to buy a Model S and use his own software to power Autopilot 2.0

If you think that the rivalry between George ‘geohot’ Hotz’s small autonomous driving startup comma.ai and Tesla is only in the hacker’s head, you might be wrong. When his self-driving car/machine learning startup ‘comma.ai’ came out of stealth mode, Hotz released an email conversation he had with Tesla CEO Elon Musk in which he was offered a contract with a “multimillion-dollar bonus” for him to build a new Autopilot system in order for Tesla to discontinue Mobileye’s part in the program.

It prompted a response from Tesla in which I’m just now realizing the company ended up lying about misrepresenting its future plans for Autopilot. Tesla defended Mobileye since Musk’s conversation with Hotz made it sound like Tesla wanted to get rid of Mobileye.

This guy is a con artist.

 

[eye.surgeon]

Autonomous driving is important for Tesla car sharing network, car sharing can greatly increase the utilization of a car, thus less number of cars need to be built to replace the ICE fleet. So this is aligned with Tesla's core mission.

Prediction: Tesla car sharing will equal battery swapping in it's success.

 

[AnxietyRanger]

Prediction: Tesla car sharing will equal battery swapping in it's success.

For AP2, I fear you are right. I just can't see that hardware suite being sufficient for widespread adoption of true driverless driving, let alone the sort of autonomous sharing service. Maybe California, maybe...

For AP3, things can be different. Just as battery swapping might get a new lease of life on the Tesla Semi (reading history on GeVeCo is interesting).