-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Mobileye - Most Impressive Self Driving Demo Yet (CES 2019)
- Thread starter Bladerskb
- Start date
Glamisduner
Active Member
Right but does it require some extra tech to do this, or are the cameras enough? I mean I think the volvo xc60 just uses the cameras. Wife said the pro pilot works good too, but I haven't tried it yet.
Cameras are enough for sensors, but it'll probably need HW3 for compute power.
Glamisduner
Active Member
I don’t agree MobilEye is late to the NN game, they simply use a more diverse and efficient set of technologies that allows them to do much more than Tesla does currently with much less computing power. And MobilEye is significantly more accurate and reliable in my view.
We have previously agreed to disagree I think on our belief on how machine learning plays a role in this so I’ll leave it at that.Suffice to say in my view MobilEye is ahead of Tesla in that part. I acknowledge you disagree.
I do agree in-house integration can provide efficiences. And like @Bladerskb I have always thought car makers often are slow on the uptake of new technologies like those in MobilEye’s chips too. Tesla was quicker and that made them a great MobilEye integration partner for consumers.
But so far in this particular space (driving aids and autonomous systems) Tesla got better results when they partnered up. I sincerely believe AP today would be multiple times more impressive than it is had Tesla upgraded it together with MobilEye components instead of going it alone.
Tesla and EyeQ4 together would have been a formidable partnership. Too bad it did not happen.
Why not reserve your judgement until AP3?
As to the partnership I think Tesla never wanted one preferring to do it on their own. Breakup slowed them down and it was the right move by Mobileye. However, I does not seem that Mobileye took advantage of that, probably because they have to move with the speed of ICE manufacturers.
Tesla's current NN does not support traffic signs detection and alot of other advanced NN features. Volvo on the other hand uses eyeq3.
CarlK
Active Member
On the contrary it was Tesla that took the initiative to break up with Mobileye, first signs appeared in late 2015 when Tesla hired Jim Keller and tried to recruit George Hotz to "get rid of Mobileye". Mobileye wanted to stay with ADAS where it was the leader and did not want itself or others to move into the autonomous driving areas. The explanation it gave at the separation was just a spin to save its stock price but it somehow worked. Why would a sane company want to dump your most important customer, if that customer still is interested in the relationship, instead of to work with it even when you have different ideas of how to approach it?
In general having "partnership" is a sign of weakness by admitting that you don't have the technology and not thinking you could develop one. Why would one wants to share the know how with partners if you already have the leading technology?
Last edited:
@Vitold @CarlK
I am very interested in HW3 as my car should be slated to receive the upgrade. I will gladly comment when it does.
That said what I referred to was my belief of where things stand as they are shipping in cars. I believe the chips MobilEye ships in production cars contain superior autonomous/driver’s aid technology compared to the technology shipped by Tesla today.
If that changes with HW3 and software for it, great. For example the traffic sign recognition is desperately needed. It is such old hat on other premium cars.
As said I fully acknowledge from consumer perspective Tesla provides faster access to their technology than MobilEye and their slow as molasses car maker partners. That’s why I drive a Tesla.
But I do believe MobilEye itself is superior in its technology as shipped today. Vastly so.
I am very interested in HW3 as my car should be slated to receive the upgrade. I will gladly comment when it does.
That said what I referred to was my belief of where things stand as they are shipping in cars. I believe the chips MobilEye ships in production cars contain superior autonomous/driver’s aid technology compared to the technology shipped by Tesla today.
If that changes with HW3 and software for it, great. For example the traffic sign recognition is desperately needed. It is such old hat on other premium cars.
As said I fully acknowledge from consumer perspective Tesla provides faster access to their technology than MobilEye and their slow as molasses car maker partners. That’s why I drive a Tesla.
But I do believe MobilEye itself is superior in its technology as shipped today. Vastly so.
CarlK
Active Member
Name one area in a real car that uses Mobileye chip that is superior. Where it is at lane change? How does it do on not specially mapped or geofenced roads? What about suggesting and excuting lane changes or highway exists? Which Mobileye equipped cars have these most advanced features? OK reading speed limit which was done by human annotated data but Tesla is using NN which do need enough amount of data to tune it. It will get there.
You seem to want to just ignore what I've been showing you before that as late as 2016 Mobileye was still pooh-pooh'ing NN deep learing Waymo (and Tesla) were doing. The Tesla/Mobileye fallen out began when Mobileye learned Tesla was developing its own NN algorithm. Tesla did not want to share the development with Mobileye and Mobileye in turn was not allowing Tesla to access the raw cameras data for that use. Of course Mobileye changed its tune and practice of NN deep learning more recently when it became the thing. Although it's still more talk, which I found this company is extremely willing to do, than delivery.
Mobileye Bullish on Full Automation, but Pooh-Poohs Deep-Learning AI for Robocars
Last edited:
@CarlK
I base my opinion on what I have learned (as an amateur of course) about the MobilEye chips themselves, irrespective of how they are used by the OEMs. My view is the capabilities in the chips are vastly superior to what is being used by OEMs. I have acknowledged the latter negative many times. Car companies are slow to use the features. Tesla was quite quick to use them, they would be doing wonders with EyeQ4 I think... too bad the partnership ended at AP1...
As noted by both of us, one significant feature that is being used by OEMs and has been for a decade (Tesla’s AP1 included) is the speed limit sign recognition. Not only that but other signs are read as well and are used by OEMs like roadworks. Another is obstacle recognition used by adaptive suspensions by several premium manufacturers ie potholes and the like, this Tesla does not yet have support in their shipping NNs as far as we can tell — Tesla’s non-vehicle obstacle recognition is quite limited...
I don’t share your view on the machine learning differences or its significance in this case as I believe MobilEye’s approach to machine learning is more efficient and does not have downsides compared to Tesla. I know you feel different, we just disagree there.
I base my opinion on what I have learned (as an amateur of course) about the MobilEye chips themselves, irrespective of how they are used by the OEMs. My view is the capabilities in the chips are vastly superior to what is being used by OEMs. I have acknowledged the latter negative many times. Car companies are slow to use the features. Tesla was quite quick to use them, they would be doing wonders with EyeQ4 I think... too bad the partnership ended at AP1...
As noted by both of us, one significant feature that is being used by OEMs and has been for a decade (Tesla’s AP1 included) is the speed limit sign recognition. Not only that but other signs are read as well and are used by OEMs like roadworks. Another is obstacle recognition used by adaptive suspensions by several premium manufacturers ie potholes and the like, this Tesla does not yet have support in their shipping NNs as far as we can tell — Tesla’s non-vehicle obstacle recognition is quite limited...
I don’t share your view on the machine learning differences or its significance in this case as I believe MobilEye’s approach to machine learning is more efficient and does not have downsides compared to Tesla. I know you feel different, we just disagree there.
I do not see how Waymo's and Mobileye's opinions are different.
Amnon from the article you linked:
"But Shashua has poured a little cold water on the idea of cars being self-taught..."
“What makes both driving assist and autonomous driving real is the ability to find a needle in a haystack,” he said. “There are many rare events that need to be covered to reach 99.999% capability. Building for demonstration is manageable, but building something that will reach production-worthiness requires this remaining 10 percent, and that makes all the difference.”
Waymo guys about ChauffeurNet:
“The model learns to respond properly to traffic controls such as stop signs and traffic lights. However, deviations such as introducing perturbations to the trajectory or putting it in near-collision situations cause it to behave poorly, because even when trained with large amounts of data, it may have never seen these exact situations during training,” explained Mayank Bansal and Abhijit Ogale, both Waymo researchers.
Also some details:
"Waymo’s current system, known as the “Waymo planner,” uses some machine learning but is mostly rule-based. "
about ChauffeurNet:
“In order to drive by imitating an expert, we created a deep recurrent neural network (RNN) named ChauffeurNet that is trained to emit a driving trajectory by observing a mid-level representation of the scene as an input. A mid-level representation does not directly use raw sensor data, thereby factoring out the perception task, and allows us to combine real and simulated data for easier transfer learning.”
Sources:
Waymo presents ChauffeurNet, a neural net designed to copy human driving
Waymo tests imitation learning for path planning
strider
Active Member
This thread is hilarious. Everyone yelling about who's baby is the least ugly. I've had EAP for 2 months now and it is complete and utter garbage. It gives everyone in the car nausea (with it's herky jerky braking, accelerating, and swerving all over the lane) and me high blood pressure (by cruising full speed into slowing traffic before slamming on the brakes, when approaching a slower car it slows down to match the speed of the car, then changes lanes, then speeds up - this is wildly unsafe not to mention causes the aforementioned nausea). One night a couple weeks ago we were driving home on the Interstate in freezing fog and the system was paralyzed. There's no fall back to regular old hold the speed cruise control. I could go on and on.
The industry has a VERY, VERY long way to go before anyone can use the term "self-driving".
The industry has a VERY, VERY long way to go before anyone can use the term "self-driving".
Spot on post, I was expecting so much more from EAP. Where I drive, it's a jerky mess, but does not induce nausea because disengagement is just a moment away.
I see you are in OK, would expect it would work better away from the chaos and traffic congestion of Chicago?
I blame the false perception partially due to \ lack of accountability from Tesla owners who are afraid to criticize to instead tout Tesla's non-existent capabilities. Or does EAP really work for other Tesla owners and I am just an outlier ?
Still, I do love my S 100D.
CarlK
Active Member
Not necessarily an outlier but user is still an important part of it and it's not for everyone. That's why some even says it's better to be either full autonomous or nothing. Although I love what I have. I'm sure there are a lot of people like me too,
Last edited:
strider
Active Member
I agree. I think so many people on this board "talk their book" instead of reality. We love our car too.
I've only been in 1 traffic jam in this car and I found EAP to work pretty well in that scenario. It allowed me to relax and not stress about trying being stuck. I set the following distance to 1 and let the car do it. Of course here we only have 2 lanes in each direction so you pretty much stick to your lane and slog it out. Where you live there are tons of lanes so there's a lot of jockeying back and forth. I can see EAP being dreadful there.
My main use for it would be on Interstates where being smooth is the name of the game and EAP is not smooth. As a nerd (and extensive reader of this board) I'm pretty versed in its limitations so if I drove by myself a lot I would use it more. But with passengers, ugh.
I'm actually very impressed I didn't get a ton of dislikes from my post. Well done TMC!
Soda Popinski
Member
Are you sure Tesla was compared in that one? Reading the blog, it looks like they just compared new features in new 2019 cars, based on this quote: "Every fall, we line up a range of new models with the latest and most compelling automotive technology from the past year."
When will you post the list of 10 comparisons? Off the top of my head, the Consumer Reports one comes to mind, where Tesla lost on driver attentiveness.
CarlK
Active Member
It's obvious Tesla was not compared in that review. The poster certainly knew it. Not to mention driver attention monitoring, along with LIDAR road mapping, are GM additions that has nothing to do with Mobileye capability (other than its incapability). The poster knew that too.
Soda Popinski
Member
I was under the impression it's due to patent protection. MobilEye holds a 2006 patent concerning this. I wouldn't be surprised if they refused to license it to Tesla. Bad blood and all.
Glamisduner
Active Member
Similar threads
