Mobileye rips Tesla Autopilot, Chairman says it dumped Tesla

[rog]

I think what people seem to overlook/forget: Tesla won't remove the camera sensor and solely rely on radar in 8.0+ (how else would AP center the car in a lane?). So, the interesting questions are:

1. How will Tesla manage the sensor fusion without MBLY's input/contribution until AP2? Are they stuck with whatever the chip is capable of now?
2. What did Tesla come up with in terms of advanced signal processing of the new radar solution? I believe they rely on the MBLY chip to do that - does Tesla have enough know how in-house, or will they have to continue relying on MBLY?

 

[Chopr147]

I don't think they will be relying on MBLY for much at all. That was the reason for the break-up (In Tesla's view)
Elon made it clear MBLY was moving too slow. I try to read this article with an open mind but keep coming back to MBLY feeling left out of the future. So they blame Tesla for damaging autonomy in the long run. Not buying the sour grapes

 

[KZKZ]

Your characterization does not match what you quote. It doesn't say anything about the Florida accident. Also Tesla saying Mobileye currently being the best in the industry doesn't make any comment on whether Tesla can do better when they develop their in-house solution.

I'd highly recommend you go back and read what Musk said about Mobileye after the Florida fatality.

 

[Stoneymonster]

I think what people seem to overlook/forget: Tesla won't remove the camera sensor and solely rely on radar in 8.0+ (how else would AP center the car in a lane?). So, the interesting questions are:

1. How will Tesla manage the sensor fusion without MBLY's input/contribution until AP2? Are they stuck with whatever the chip is capable of now?
2. What did Tesla come up with in terms of advanced signal processing of the new radar solution? I believe they rely on the MBLY chip to do that - does Tesla have enough know how in-house, or will they have to continue relying on MBLY?

Tesla has insinuated from the very beginning of AP that it's their software for the most part, on MobileEye's platform. I suspect they are using MobileEye as a image processing and deep learning accelerator, and not using any Mobileeye secret sauce. They very definitely have the expertise in house now as they've been on a machine learning and AP hiring tear for a couple of years.

 

[KZKZ]

Actually, practically every public statement Tesla has made since launch has emphasized that you must be paying attention (it was even in the press launch). They also emphasize keeping your hands on the wheel. Most of the hype is by journalists. I made a post a while back that quoted some of these (Tesla telling journalist to keep hands on wheel and paying attention, but journalists ignoring and pulling shenanigans because it gives them clicks).

I agree with the others. Mobileye was silent when Tesla was getting good press and didn't make much comment even after the Florida incident. Only after the break with Tesla (and massive drop in Mobileye stock) did they make such a statement. Sounds like sour grapes.

Mobileye did comment immediately after the disclosure of the Florida Autopilot fatality. It was then that Musk stated it was Tesla's proprietary software, not Mobileye that should have seen the truck.

 

[KZKZ]

Actually, practically every public statement Tesla has made since launch has emphasized that you must be paying attention (it was even in the press launch). They also emphasize keeping your hands on the wheel.

More importantly, Tesla's implementation of AP does not back up the supposed emphasis to keep hands on the wheel. Even when 8.0 launches, a driver can keep their hands completely off the wheel for 3 minutes before a nag warning.

That kind of implementation does not match the warning for the driver to "always" keep their hands on the wheel.

 

[KZKZ]

Tesla has insinuated from the very beginning of AP that it's their software for the most part, on MobileEye's platform. I suspect they are using MobileEye as a image processing and deep learning accelerator, and not using any Mobileeye secret sauce. They very definitely have the expertise in house now as they've been on a machine learning and AP hiring tear for a couple of years.

That's what Tesla has said. And that's why Mobileye finally dumped Tesla because they felt Tesla was "pushing the safety envelope" too far, trying to do image processing of things the base system with only a single camera was ever intended to do.

With a string of AP accidents and at least one, and probably two AP fatalities, who can argue with Mobileye's assessment.

 

[msnow]

That's what Tesla has said. And that's why Mobileye finally dumped Tesla because they felt Tesla was "pushing the safety envelope" too far, trying to do image processing of things the base system with only a single camera was ever intended to do.

With a string of AP accidents and at least one, and probably two AP fatalities, who can argue with Mobileye's assessment.

It makes zero sense that a young struggling company would "dump" any customer much less their largest and highest profile customer.

 

[Cosmacelf]

What exactly do you think Tesla could have done a better job at? Every time you engage AP you get a dashboard alert telling you to keep your hands on the wheel at all times and an audible chime telling you the system was engaged. Even still AP isn't enabled upon delivery and you have to enable it yourself and accept the T&Cs when you do...

I just don't quite understand why you think the messages have been mixed or that Tesla hasn't done enough...

Jeff

I told you what Tesla could have done, I guess you didn't read it (ironic). I said they could have written a blog post detailing AP limitations and emphasizing that people must always be looking ahead. They still could write that blog post.

Interestingly, today, a take over scenario occurred to me on a 1 hour AP trip. Was cruising around 72 mph when the freeway lane dramatically slowed down to a stop, very quickly. The car in front of me swerved into the shoulder. The Tesla collision warning alarm sounded even as the AP was slowing the car down. Since I was watching, I didn't wait to see if AP would work, and just slammed on the brakes.

Anyways, the point is that that Tesla could do an even better job at reminding people to always be looking ahead.

 

[987S4]

What Tesla needs to do is create a series of videos that detail when AP will not work. They already have a series of videos that owners can watch before taking delivery, they need to do the same to illustrate the limitations of AP.

I know the limitations of AEB and TACC are in the manual, but to see offset crashes on video really illustrates the current AP limitations.

In the latest China fatality, many Tesla fanatics could see the similarities between that and other AP crashes. Most people don't frequent Tesla forums and don't have the same level of knowledge of the system. Instead they see TACC work 99% of the time, and the don't understand fully that it recognizes an already stopped vehicle differently than a vehicle that is in stop and go traffic.

AP educational videos need to be watched before delivery, and updated with each iteration of AP.

 

[Wuggado?]

And the average owner ain't reading Elon Musk blog posts buried in the "Updates" section of teslamotors.com.

True that. The only time I see an Elon blog post is when it gets repeated here.

 

[Patrick W]

To be blunt, the guy in Florida might be alive today if Tesla had done a better job at emphasizing that you MUST be looking forward AT ALL TIMES.

Media reports I heard noted the guy had several (8, IIRC) speeding and reckless driving tickets. I don't know that any auto-drive system can help with people who drive recklessly.

 

[Cosmacelf]

True that. The only time I see an Elon blog post is when it gets repeated here.

Which means you do see it. Elon's blog posts get reported on in thousands of press and blog publications.

 

[xav-]

I think Mobileye is a good short. It's so freaking expensive.

 

[trils0n]

Tesla said they broke up with Mobileye because Tesla was developing their own "Tesla Vision" solution, and Mobileye was upset at that and threatened to cut off the EyeQ3 supply if Tesla didn't cease development.

Tesla responds to Mobileye’s comments on Autopilot, confirms new in-house ‘Tesla Vision’ product

 

[Lex]

• Car & Driver: ~2x better than the next closest competitor (29 interventions vs. 56 for the BMW 7-Series)
• Autofill: ~10x better than the next closest comparison (3 interventions on the Model S vs. 31 for the E-Class)
• Motor Trend: 6-9x better than the next best system (12 interventions for the Model S vs. 113 for the S-Class)
• The Drive: “Without a doubt, [Autopilot] is the best ADAS system on the market.”

Boom goes the dynamite.

 

[Cosmacelf]

Wow. I wonder how Tesla is disseminating that response that Electrek has? Surely they didn't just send it to Electrek?

Anyways, pass the popcorn, this isn't over yet

 

[stopcrazypp]

I'd highly recommend you go back and read what Musk said about Mobileye after the Florida fatality.

My point was what you linked didn't even mention the Florida incident, so doesn't match what you are characterizing. If you are going to make a claim then your evidence will have to back it up. You should be the one digging for the reference. I'm not going to bother digging through Elon's tweets (esp. given I don't even have a twitter account, so I don't follow Elon's account in the first place).

 

[stopcrazypp]

Mobileye did comment immediately after the disclosure of the Florida Autopilot fatality. It was then that Musk stated it was Tesla's proprietary software, not Mobileye that should have seen the truck.

Yes they made one single paragraph statement and said their system didn't handle cross traffic detection, but didn't comment any further about Tesla's implementation.
Tesla Autopilot partner Mobileye comments on fatal crash, says tech isn’t meant to avoid this type of accident [Updated]

It's only now after the backlash in their stock from the news of Tesla breaking up with them that they are making such accusations that Tesla's implementation is unsafe.

The latest articles recently posted makes the motivations more clear for the break up. Tesla was already working on an in-house solution for a while now and Mobileye was not happy about that.
This was confirmed by Galves, Mobileye's spokeperson (although he spun it as safety being the "primary "reason):

“The reasons [Shashua] stated for the breakup are absolutely primarily the reason the relationship ended,” he said. “Were there commercial issues going on at the time? Yes, but those probably shouldn’t be discussed publicly.”

Tesla says jealousy, not Autopilot safety concerns, caused breakup with Mobileye

 

[xav-]

So who's right who's wrong? I think both tesla and Mobileye are right.

I think that Mobileye was legitimately concerned about the negative publicity that the autopilot accidents would bring. I think it's fair to think that they viewed their system as not perfectly suited for that application. If you remember correctly Elon Musk was very prompt to state that Mobileye was not to blame in the May fatal accident.

Does that mean that they ditched tesla for that sole purpose? I don't think so, but knowing that their relationship with Tesla was going to end anyway, they probably did not find worthwhile to do their best to stay in business with them (so likely did not offer good renewal deals etc).

Was Mobileye so jealous of Tesla developing their own system that they ditched them? I doubt it honestly. I don't think they would do that unless tesla had the intent to redistribute their in house system which we know isn't he case here.

What does that mean with AP20? In my case it means that AP20 is not going to be in scope for a while. It seems judging both from Mobileye and Tesla's communication that the deal with Mobileye ended before Tesla wanted it to end. We better hope that Teslas in house system is not too much behind schedule (and of course that it really works!)