How Mobileye is getting CRUSHED by Tesla
Vložit
- čas přidán 4. 09. 2021
- After Tesla's AI Day, where Elon Musk and the Artificial Intelligence and Autopilot team unveiled breakthroughs in software and hardware, the case for Mobileye's automated self driving system is looking bleak. Mobileye is perceived as the leader in the industry, however their HD mapping approach has now put them far behind in AI, which they believe is insufficient for Robotaxis. Tesla however has gone all in on machine learning and AI, and is preparing for a Robotaxi future with or without Mobileye.
#Tesla #ElonMusk #Mobileye
Website: themarketisopen.com
Patreon: / themarketisopen
Twitter: / themarketisopen
E ln(musk) T-Shirt: www.amazon.com/gp/product/B07...
Disclaimer: We are not financial advisers, and nothing on this channel is meant to be financial advice. The ideas expressed on this channel are purely opinions and should not be regarded as objective information. Nothing on this channel is a recommendation to buy or sell securities. Do not assume that facts and numbers in any video are accurate. Always do your own due diligence.
Do you think Mobileye's mapping strategy is still sufficient for autonomous driving, especially after Tesla's AI Day reveals in artificial intelligence and machine learning? Be sure to watch:
Tesla’s BREAKTHROUGH DOJO Supercomputer Hardware Explained: czcams.com/video/pPHX7e1BxSM/video.html
What Cathie Wood JUST Said About Tesla: czcams.com/video/b7MM8nd2tcs/video.html
Visit our website themarketisopen.com for instant stock quotes and financial data
No, no way José!
I just realized that Tesla is actually an AI company!
I’ve never thought mapping and lidar was an acceptable solution
So Tesla motivation to sell FSD to legacy OEMs is to allow ICE vehicles to be fully autonomous? Or other automakers could use it to create their own robotaxi fleet and undercut Tesla. Given many legacy automakers are going out of business is it worth Tesla's time to adapt FSD for all the different models? Believe answer is no on all counts. FSD is the ideal tech advantage for Tesla, puts them in a different league. Their speed of development is not so much a commercial moat as an unbridgeable chasm that is forever widening.
@@chrisprophet8696 Agreed! Before watching this, I didn't know that the legacy automakers were using essentially a 3rd party technology. They are totally screwed...
After seeing Sandy drive the MachE blues cruise, I do not believe any other company will solve autonomous driving before Tesla does..
I thought Mobile eye was a serious competitor, but now I know the Mach E was using it, thanks to this video.. not anymore.
You are right, very shady Mobil eye hides the flaws. It raises the expectations Mobil eye will never be able to fulfill.
I'm with you full stop
Though i have 1 question for any early model s owners did the system have the same problems back then when tesla was useing mobileye ?
@@boatthere867 - That is a Logical Question! But, ad I understand it, Tesla took what they provided, and "Added" some of their own "Very Early" AI to the Mix! Mobileye was not "Happy" about that, as they complained that "Tesla was Pushing it beyond its design Limits", even then!
Anecdotally, my Coworker, with a Model S, with that AP1 System, told me, his wife said that "the car drives better than he did!" And that was on a Trip from Toronto to Key West & back! So, lots of opportunities to use it, and he did!
I believe he has since said, that the new Model S he traded it in for, initially with AP2, was worse, by a Lot, over his AP1 Experience! But that it has since Exceeded the Capabilities of the older system, at the time he last drove it,
It would be nice to hear if any AP1 Drivers, still using it, today, have seen any additional improvements! (Remember, it just had the Single Forward Facing Camera, as I recall! And the Less Capable Ultrasonic Perimeter Sensors (but even they helped avoid some "Side Swipe" Collision events!), And, it had the Forward Radar, too, I think!
Your summaries are always more insightful than most of the other Tesla channels. You're not just repeating tweets and others' statements. Well done.
+1
What is insiteful???
This incites much hatred as usual from Tesla haters.
"Insightful" he's not 'inciting' anything!
@@gregbailey45 thanks, corrected. Covid brain is my excuse.
Tonight I summoned my Tesla from its parking spot to the front of the building so I didn’t have to walk.
Nobody else can come close to doing this reliably right now. Can’t wait for more
This is a strong episode, with content that is fresh. It hasn’t been rehashed 1,000 times. Kudos.
Finally an in depth comparison. I've been wondering about mobile eye's advancement/strategy on fsd for a while. Thank TMIO
The strategy you have when you don't have a strategy!
Tesla uses alien technology. Mobil Eye human technology.
LOL people that don't understand AI should not try to make "in-depth" videos, the result is always as laughable as this :D
Ford‘s Blues Cruise is highly dangerous for the User. Safety doesnt seem to have priority at Ford.
If it's released in the state that Sandy tried out, we will see fatal accidents within weeks
"Perception of traffic is easy, understanding traffic is hard". Tesla is solving the hard part first.
There's no way that mapping is ever going to work for FSD. AI is absolutely necessary for handling the prediction of what other road users are doing.
I think "Maps for Navigation" are important, most of all, but, as to the actual Road Space, the AI Approach, will win.
I know that LORAN (LOng RAnge Navigation) came out, first for Naval Navigation (Ships & Boats), then some guys stuck a Small Boat Version of a Early, & Bulky LORAN Unit on the Front Passenger Seat, of a Light Plane, like a Cessna 172, and began to use that as an Extra Help for Radio Navigation! That was in the Late 1970's or Early 1980's!
Then UPS, and Aviation Avionics (Aviation Electronics, for Communication & Navigation) companies started to make LORAN Units, to fit Standard Avionics Stacks, in standard widths, first with simple LED Digital Alphanumeric Displays, in singular, then Dual Outputs, with selectable choices! Also, they were, what we called, a "Full Height" Box, like 2.5" to 3" Tall! Later, they Shrunk down to "1/2 Height" Boxes!
By this time, Military GPS (Global Positioning System) was getting started! Soon Commercial Avionics Manufacturers started making Early GPS Aviation Units, that frequently sat in the Stack, along with another unit, for LORAN!
Then, a few Avionics Makers, released a fully Functional, GPS & LORAN Dual Capability System! I think, at the high End, at least one added "INS" or "Inertial Navigation System" capabilities!
LORAN had Excellent 2D Navigation Capabilities, but, Zero 3D Ability! GPS, added that 3rd Dimension, but was a Weak, Space Based Signal. LORAN could do things GPS couldn't, and GPS could do things, LORAN couldn't! Together, they were an Excellent Aviation Pair! Of course, adding a "Radar Altimeter" into the LORAN mix, would have helped it, for Approach Work, but not so much with Cruise Condition, as to Altitude!
Today, GPS, for Aviation use, has added "WAAS" or "Wide Area Augmentation System" to the Mix, using Fixed Ground Based Systems, that sent out a Correction Signal, to bring GPS Position Accuracy to about a 10 Fold Improvement, or something like the 1 Foot position location, in "X, Y, & Z" space vs about a 10 Foot Possible Range of Positions! But, the Avionics in the Aircraft had to be able to receive and integrate the WAAS Signal! But, with this fully operational, they could use it for the Hardest Instrument Approaches, called "Category III" (3), that Places like London, Heathrow, already had, without GPS, but now, airports besides those with Ground Based Category III Approach Equipment, could also allow such an approach! (So long as they were in the Range of the WAAS Signal!)
However, as this is for Aviation use, where there are very few obstacles to block signals, I doubt it could be taken advantage of, for Road Traffic, especially in areas with Tall Buildings, Large Billboards (that today are more frequently High Resolution Pixel be Boards, with TV like Clarity, that probably offer a Bigger Signal Block), so that is part of my understanding, as to one of the things, Cellular 5G is Targeting: A GPS-Like Function! Maybe Intel / Mobileye, have some interest in that idea, I don't know, but, I have heard of recent developments, that even use WiFi for Navigation In Building Robotic Navigation Systems Function!
And, we already know, that a Cell Phone Can be Located, by "Pinging" it, from a selection of Cell Towers! So, it is possible that some Players might be seeking to Augment GPS, with additional Sensors and integration of Location Techniques, via some of these other choices!
Outside of Cities, I imagine a Commercial Freight Traffic "WAAS-like System" might someday, be Proposed, for More Precisely Positioning Some (Autonomous) Vehicles!
Basically, we already use GPS for Surveying, for Construction Grade Leveling, and many other "Tricks", today! And, placing a fixed Location, Precision Surveyed, GPS Receiver / Transmitter, is one of the Tricks, for creating Increased Position Accuracy, from the right GPS Systems! Such systems send out a "Correction Signal" that removes a lot of Variance, from the Space Based Satellite Signal, caused by various Attenuation Interference factors, like Volcanic Ash, Smoke, Lighting Storms, or other elements! (Basically, that's what WAAS is, and does!)
Yet, I still wish LORAN was operational, to provide additional Redundancy! (it was Decommissioned, years back!) It offers a signal that travels better over land, and less attenuation, from Terrain and other "Obstacles!"
@@ezicarus8216 You give no justification for your statement. Humans are terrible at driving. 1.3 million people are killed on the World's roads each year. Every time you drive or even walk beside a road you put your life in the hands of drugged, inattentive, sleepy, irresponsible or inexperienced drivers.
Nothing is idiot proof because idiots are so damn creative. Never discount the intense danger idiots present.
Agreed. Solving computer vision is the only way to go.
@@ezicarus8216 That's why we have 40,000+ fatal accidents per year in the US, humans can't predict what other humans will do very well; but AI can.
Mobileye's approach is like the fable of the man who stepped on a sharp rock in his bare feet, so he decides to cover the whole world in leather to protect his feet.
perfect analogy
I love how Tesla can see a problem and then go out to their over one million "data collectors" and ask the fleet to send back similar situations and corresponding videos. Everyone else has maybe one or two examples of that specific problem whereas Tesla probably gets back hundreds of thousands of this one type of problem/example to analyze and write a future OTA to fix.
Statistical analysis, 10,000 Frenchmen can't be wrong. (Or one is an outlier).
That was a great analysis explaining the difference between the two systems
You are one of the few youtubers that does comprehensive analysis and narratives vs just reading a news release and commenting. This synthesis is really insightful. Keep up the great work. Much appreciated. Weekly releases or such seem like the right cadence to do the analysis and synthesis.
Tesla will win the FSD battle and OEM's will be forced to license FSD from Tesla. A future revenue stream!
Do you think that the point that Tesla licenses FSD to other manufacturers, FSD might be more profitable than selling the cars?
@@jamesaspinwall - Possible! But, not to the Point of Ending Tesla's Vehicle Production Plans, to have at least 1 Vehicle in each Category or Class!
If anything, it would likely Accelerate Vehicle and Energy System Sales, Production, and Gigafactory Build Out Rate!
Already, I am Anticipating an Announcement, on the Northeast USA Site, location, within 12-18 Months, of Giga Austin's first Deliveries!
Also, Tesla has Said, that another Asian Factory is in there mind, and not in China, so many people are thinking Malaysia! I expect that, too, to be announced, within 12-18 Months of Giga Berlin's First Deliveries!
A man d, within 15-24 Months, U expect, yet another Location, to be Announced! Then, there is the Indian Question: How will that "Play Out?" Will they open direct sales there, for Imported Vehicles, and wait 3-6 Month's, before announcing a Factory? Or will it not be announced for 6-24 Months?
@@ezicarus8216 - And Who is Clamoring to License those Patients, Today, Pray Tell?
An up and coming Mobileye powered _last-mile robo fleet ( est 2022 )_ will require occasional remote control _(AKA "tethering" )_ assistance to *provide operational inputs* whenever these vehicles can't figure out a situation. They also admit, that while carrying passengers...when that time comes.... the passengers might, from time to time, be needed to make diving inputs. What if those passengers have never driven a vehicle? or are impaired? Lol.
Ouch!
They will still supply inputs, just look at all the crash videos.
The remote operator is required by law for ALL level 4(+) systems at the moment where these systems go public (Germany + Israel) if no safety driver is on board.
Tesla would "need" this too if they would offer level 4(+) without a driver.
That was surprisingly good. Much deeper than I expected. Thanks.
Phillip.
great video. mobileye is stuck exactly where waymo is. Their robots can drive perfectly where humans tell them how, otherwise they are bricks.
One correction re: maintaining camera data is that it shouldn't much matter as long as the cameras are near their initial positions. Programming a simple offset from base is super easy. They can auto-calibrate like Tesla cams can, and as long as its not so far off point that they dont cover the initial set point, it drives fine. Also, maintaining db entries is trivial as they work with OEMs anyway well before the cars are outfitted with the tech.
"otherwise they are bricks" you meant "moving bricks". There is a difference.
@@jamesaspinwall I mean bricks in that they will not drive if their lidar localization is invalid. They will not move for safety's sake. Brick example here (this isn't lidar issues, its a novel static agent problem, but if you enable the system in a new location it will do the same thing.)
czcams.com/video/B4O9QfUE5uI/video.html
I enjoy Mobileye on my 2015 Subaru Forester. It works day and night, snow and rain. Does not work if looking into the sun. I love adaptive cruise control and look forward to my next car - a Tesla built in Texas.
It's very noble of you to wait for the us made model rather than taking the better quality Chinese made.
@@maxmeier532 All Tesla's sold in the US are built in the US. He is just waiting for the improved build of GigaTexas as opposed to Freemont.
@@maxmeier532 You think the reason that the Chinese made Tesla have better quality is because they are made in China? No lol. It's because they are made in a newer factory. Texas made model Y will be the same quality as China, if not better.
@@maxmeier532 The quality is reflected by older equipment in US. Austin will have state of the art equipment. Front and rear megacastings won't hurt either.
If anyone from Mobileye watched this video they would really question their jobs by now. I thought the same thing for AI Day: if any OEM vice-president for IT watched Elon and his team explain neural nets, exaflops and the D1 chip, they would be questioning their own futures at Ford, GM, Toyota and BMW.
Mobileye's engineers must believe that they are right. They wouldn't question their jobs. There are plenty of jobs for them anyway.
The other companies own the press. Tesla could 100% work and the others could drive off cliffs 100% and the media would only report on how reckless Tesla is to not use the industry standard system.
Vision-based (camera) autonomy, powered by AI and a fast computer, is several magnitudes more difficult that camera+radar+lidar systems. However, the latter will rely heavily on the mapping of very small geo-fenced precints (which can change, as roads are modified or created).
The status quo gives the impression that systems like Mobileye are ahead in the game, and in simple terms, this may be the case.
But vision+AI will rapidly leapfrog these mapping systems, rendering them not only redundant, but logistically inapplicable in the real world.
Tesla is not concerned with the opinions of observers or analysts... They know that once vision+AI is "solved", there is no threat from alternative technologies.
"several magnitudes more difficult" Once you transform it into a vector space, it becomes an easier problem to solve. Easier, not easy.
@@jamesaspinwall Actually a reprise of the issue that Elon's Zip2 solved by using Vector graphics for directional mapping compared with the existing at the time slow loading data heavy cumbersome mapping
Good video. Answers a lot of the questions I have. Thanks.
The Mach E and the 'sharp turn' on an interstate? What a joke. I've driven most of California's very twisty Route 1 using FSD. There are some very tight corners marked 15 mph, that I'll take over on because FSD either doesn't want to do it, or it's too slow, but all in all FSD is great and makes this trip a real joy.
The Los Angeles 710 Freeway, the one closest to my home, and which I've been using for many years was for about 25 years under continual construction. (Bit of a boondoggle. They'd do things like install Jersey Barriers, and soon as this was done they tore them up to lower the road bed a couple of feet for truck/bridge clearance. Then they installed new Jersey Barriers.) Still it meant that this road would change not just daily, but multiple times every day. Mobileye (or Waymo) would have to have a vehicle just dedicated to a road like this.
As a Model Y FSD driver, and having driven now for almost 50 years - I sometimes disagree with Tesla's highway lane changes. I have so much experience that if I see a particular brand of car, an old car, a car with a personalized license plate, an elbow sticking out of the open window, a head resting on an arm - this means something to me and I'll change the lane I'm in, raise or lower my confidence in the safety (attention level) of that driver. As Tesla's 3D leveling expands its use of AI should arrive at some really interesting 'prejudices'. It might decide red Hyundais with personalized plates should be given a lot of distance -- meanwhile Waymo will still be orbiting a small part of Phoenix and Mobileye will be updating their basic map.
Absolutely right. You've reminded me that my mum used to say to be aware of "young men in old Fords" and I later learned "old men in hats" in the car. Both have worked well for over forty years. A commonly seen UK plate to watch for starts BO55 ###. Tells you all you need to know.
I am a senior software engineer working at one of the FAANGs. I closely follow AI progress. I must admit that your research on this area has humbled me. Great video. I have become a supporter of your channel!
Thanks for supporting my work!
Excellent topic. Was interested in a comparison of these two. Looks like it was very smart move that Elon broke off from Mobileye and became further vertically integrated with their own system which is the best assisted driving suite out there.
Excellent Breakdown of the differences between these approaches! Your content is a great example for other channels that actually want to CONTRIBUTE to the conversation, and not just post snippets of old information. Well Done!!
Elon seems to be asking the right questions and his team may deliver FSD or something useful and allowed under regulation first whenever that is!
Elon understands rapid iteration like no one else. Traditional engineering (design build test sell support improve replace…) is incompatible, and can't compete, with rapid iteration - not in DevOps, not in iterating the best space vehicle, not in the drive to true self driving. Evolution is so powerful it overcomes entropy; speed up evolution with rapid interation, and you're unbeatable. I'm heartened to hear that Tesla is applying this strategy to all aspects of this vertical, including the chips.
Very good and thought-provoking video - maybe just a bit of extra repetition in the narration.
Another great analysis of Tesla AI and the competition.
When I first heard about their shadow driving software, I did think what a great "little" idea. I still think it is a great way to test in the real world.
Great video!! Was waiting for this!
Great analysis, I feel more analysis of the competitors would be interesting!
I think you’re slightly off base. Unless something has changed MobilEye’s goal is the same as Tesla. They have said many times in the past they would migrate from being a ADAS company to an Autonomous driving company with an emphasis on vision. Their data collection partner is/was VW. Mapping was just a byproduct of collecting visual data and Tesla is doing the same. Having said that they do have a much more complicated environment compared to Tesla and therefore a much steeper hill to climb.
Oh yeah! Tesla will definitely license FSD to any OEM that wants it, and that will make them Thee standard barer of Autonomy for the entire industry! -- Keep up the Excellent analysis!
exceptional video. Thanks! I will start using your website, as you suggest.
Very good perspective and what most that aren't following the field closely don't realize.
Excellent information on the difference between Mobile and Tesla approach to self driving
I think mobileye is also working on visual only solutions, radar+lidar are just some extra layers of safety
Excellent detail descriptions comparing Tesla to mobile-eye.
Yes, and as someone else said we don’t hear the same information rehashed dozens of times in one video. Love your channel and I have for quite some time. JVV
Really great insight. Can you compare comma and tesla?
Nice job explaining the technical capabilities and the flaws and limitations of Mobile-eye. Job Well Done
Now Mobileye has seen Tesla drop radar and move forward with improved results, they must know who the winner is going to be.
You explained very well the shortcomings of the Mobileye approach and once a company goes down a certain route it’s very very difficult for them to change. They are sort of forced to keep convincing themselves they were always right.
So in a world without other vehicles, people, animals, weather, construction sites and change, Mobileye works perfectly 👍🏻
Such a great video! Added this to my Favorites!
I also saw a Model S down the drag track it won every race. The owner is waiting for the Plaid LOL
Thanks, for such a great and thorough description of the differences between the two approaches. I really really do appreciate this
Excellent comparison. Thank you👍🔋⚡️
Great understanding and content.
Thanks
You are my new favorite Tesla channel. Excellent!
Great explanation, thanks!!! 🙏
Great vid very informative and clear thank you!
I have 1 question for any early model s owners did the system have the same problems back then when tesla was useing mobileye ?
like the mack-E on highways not been able to do a simple bend in the road ?
Great explanation for those without a rocket science degree. Much appreciated.
Excellent content. Thank you for sharing it.
Some very good arguments to be used in discussions.
High quality content, keep it coming
Excellent overview. Very well written.
I enjoyed your in-depth presentation. 5 stars.
Excellent easy to understand video. Many thanks.
Great video, great explanations!
A Great video thank you so much !
5:05 It looks like by "not realistic for current AI" they mean they will do mapping until someone provides a good enough AI. Good luck with that, Mobileye. Tesla will license its AI in the future, so go back to mapping and you'll be fine.
Elon knows exactly what has to be done and goes after it. Because he is not beholden to anyone else he can follow his own path. I think the other competitors frustrated by how fast he is moving. They can’t keep up.
So is Blue Cruise just the branding of the Mobileye technology for use in Ford vehicles? Or does Ford (and everyone else for that matter) do something in addition, such as user interface? And what is GM using?
No one without an agenda can look at what Teslas driving is doing and think favorably of any of the competition.
I feel kind of bad for people so blinded by brand loyalty that they bought a VW or Ford EV. Wait, I don't feel bad for them at all! Let them figure it out.
Very well done!
Very good analysis - as usual - thank you
Great analysis. The big thing going for Tesla is they can keep rolling out FSD to a wider and wider number of "testers" regardless of gov't approval of level 4 or 5...just by saying the driver is always responsible for managing the system and taking control.
Many people will be familiar and comfortable with Tesla's system as it develops further and more is reported about it.
There are no laws regarding level 4 5 autonomous driving. Waymo already has limited numbers of robotaxis with no backup driver in San Francisco.
@@sava411 yes, I saw that. In one video (that I saw filmed by a passenger) the car got "stuck" by indecision for no obvious reason and after a while a man ran up to the car. He got in to help the car with the situation and left again as far as I can tell. That means interventions are more arduous and slower for the system to process. I don't know if the event I saw would teach Waymo anything.
They've way mo' to do I think.
@@andrewsaint6581 - "Lots 'Way Mo' to Do!" 😲😅🤭
@@andrewsaint6581 my point is not that waymo is good. My point is if tesla is using “we cant release because regulations” as an excuse for limiting their system to level 2 its bs.
Exactly. This tactic is critical to rapid innovation. The total number of testers is not overly critical. Just the fact that they can do this with a good sample set is fantastic.
Great work !
An awesome high-level comparison of the two systems! (Tesla and MobilEye) Tesla will be successful first because they have worked on the most difficult part first. (Perception of the world around the car). MobilEye has solved the easy part first (mapping the road in detail). MobilEye will then find that this is insufficient and have to start solving the hard problem. But of course, they will have lost precious time and spent countless $$'s. Lesson Learned? When tackling a difficult challenge, ALWAYS start with the hardest part of the problem.
This is the best comparison video yet. Thanks for making it.
If I may add... there are 5 points I'm thinking about that give Tesla an edge.
1. Tesla is at this for 4 years (or something like it), but mobileEye started 16 years ago. If parity with MobilEye's tech isn't already there, it's definitely not many years ago with Tesla's operation vacation. :)
2. I really wonder how good MobilEye is at handing true edge cases, since there are so few real autonomous miles being driven compared to Tesla's. Sure, they have many human miles with simple ADAS systems (mostly front camera's), but they don't really know how their system would have performed.
3. The way Tesla can simply upload algorithms to a portion of the HW3 fleet and see how it runs in shadow mode is a powerful way to iterate. Much more confidence going forward compared to MobilEye (either a small fleet of professionally baby-sit cars or a large number of cars that they can't simply push new algorithms to, because they're not the OEM of the car nor the driver consent).
4. The amount of TOPS (trillion operations per second) is also lower for the EyeQ5 compared to the 2019 FSD (HW3.0) of Tesla. This may see minor, but I think that with enough training and good enough datasets (DOJO + sourcing from the fleet) inference doesn't even need the highest amount of TOPS, because it can be more efficient and general at the same time compared to hardcoded algorithms.
5. Risk-aversity: The OEMs will not want EyeQ chips if MobilEye has AV accidents. I think Tesla had proven that people continue to buy Tesla's even though there have been fatal accidents. I'm not saying that Tesla isn't carful, but they've already had their first fatality (sad to say) and MobilEye may be more risk averse because it's still ahead of the first fatal crash (something will fail eventually).
I found this video to be one of the best in depth video's about MobilEye, but it's already a bit old: czcams.com/video/9JWvzuOlAKs/video.html Still looking for a more recent one that covers everything...
You got the AI-reliance bit completely wrong. Mobileye system is as AI-driven as Tesla (the data is used for machine learning, not for creating a database of street signs :D). The REM mapping is not a substitute for AI, it is a supplement to it. They have LIDAR/RADAR not as a substitute for vision, but as a redundancy. Mobileye puts much higher premium on safety than Tesla. That's at the core of the difference in their technologies.
excellent content, thank you.
Well done!
One thing is for sure ... in ten years, we'll finally know who's right.
Picking who's right today, and investing in that company, is an opportunity that will no longer exist in ten years.
Such quality content. You have been delivering great content for a long time, but I feel like you have stepped it up lately.
A fine analysis. Thanks!
Very well done, thank you.
If you tried to make an animation about how Mobileye works, you might have to show cars crashing because of the lag in mapping info changes in roadway construction or previous accidents. I am still unsure if Tesla uses maps to facilitate self-driving. A little bit it seems.
Very nice summary, and you told me more about Mobileye's approach than I knew. Thanks!
AI Day flabbergasted me. My jaw hit the floor. I kinda-sorta knew what they were attempting to do, but the progress they can show on FSD and Dojo, their rising ability to use machine learning at large scales, is without precedent.
They haven't yet crossed the finish line with FSD. But we need to ask, where *is* the finish line?
Tesla can't decide that, because the answer isn't technological. It's political. The finish line is reached when jurisdictions will permit autonomous driving on their roads.
This is no simple step to take. Autonomous driving requires a legislative and regulatory foundation answering questions like, if the car turns out to be at fault for an accident, who pays? Where does the liability go? Or, how will a police vehicle direct an autonomous vehicle to pull over, and how will it comply? Will some jurisdictions apply restrictions on autonomous vehicles that don't apply to vehicles operated by humans, such as roads or lanes closed to autonomy, or hours of permitted operation, or rules for autonomous vehicles pertaining to weather or flooding or landslides or construction zones or emergency vehicle interactions, which could be different from rules applied to human drivers?
Nobody has a legislative template ready to go for any of this. The closest would be *pilot* robotaxi authorizations in a handful of limited areas (*not* Tesla pilots), but the lessons learned from those pilots haven't been ground into the legislative sausage needed for go-anywhere autonomy. It's not clear that those lessons from pilots will be enough for go-anywhere autonomy, anyway.
Insurers aren't on board, either.
There will be political opposition to Tesla receiving authorization for autonomous operation of its vehicles. Cab drivers. Petroleum companies, who are opposed in principle to *anything* Tesla wants to do. Truck drivers. Legacy automakers, who stand to lose a lot of new car sales. Legacy auto dealers, for the same reason. Used car sales operations - if some cars are autonomous and some are not, the value of those which are not will plummet. Technophobes, Luddites, TeslaQ, unions, anyone with an ax to grind against Tesla will appear in hearings to voice opposition. Some of them, stridently.
Many of those entities are major donors to political parties.
What will it take to get those authorizations? What will it take to get laws and regulations on the books governing autonomy? How will all of that unfold? Where is the damn finish line?
I'll tell you right now, nobody knows the answer to that question. Even Elon doesn't.
When true go-anywhere autonomy was technologically infeasible, it didn't matter. But now, Tesla is near to erasing that obstacle. The next obstacle lies ahead.
It's probably going to be very messy.
Superb content!
Great video
Very intelligent presentation.
Thanks for this video. I learned a lot. Very well explained!
A comparison between Tesla FSD and Comma.ai would be nice.
I wanna know about the crashes into police and fire do the safety lights of emergence vehicles affect the cameras?
Maybe they do affect the cameras or software so the number one safety feature (the driver) should be able to recognise the flashing lights and take over as mandated by Tesla. I wonder why the number one safety feature fails. It seems that the best analogy is hitting your thumb while using a hammer and blaming the hammer for the injury.
Very good analysis.
Very nice. I didn't know Mobile Eye fell back on the crutch of mapping. Sad to see. Worse, seems like they are joining the crowd with LIDAR. LIDAR may be ready some day, but not today and not soon. The former is more concerning though.
Mapping is a poor substitue for AI aided by mapping.
Using Mapping for Dealing with the Road Space, is interesting, but would be Useless, in the case of a Sudden Flood, or a Road being "Washed out", where even Humans Fail at Managing, most of the time! Same thing with a Sudden "Sink Hole" in the Road or Pathway!
However, I suspect some aspect of Mapping, is important in general Navigation, from A to B, to C! At least for some form of Efficient Travel!
Don’t know much about LIDAR but doesn’t it have a lot of the same limitations as RADAR?
RADAR can’t see lane markings, can’t read road signs, can’t see what color a traffic light is or whether it’s blinking, can’t see vehicle brake lights, can’t see emergency vehicle lights, can’t see strobe lights on top of school buses, etc.
Thanks for this. Really an excellent update on Mobileye.
Tesla will have FSD before the end of 2021. In fact, I'm expecting the announcement in October at the annual shareholder meeting.
One of the things that has delayed Tesla is the revelatory process of original development. I think they're on their 3rd or 4th complete rewrite, which is a result of going as far as possible with a particular approach then discovering that further patching is pointless, and the only option is a complete rewrite.
The latest software integrates 3D images and time progression, which is fundamental for AI robotics, so it ain't just FSD anymore. Tesla is going to revolutionize manufacturing with smart robots that have a timesense, a memory of what it has already done and the next tasks that it has to do. Huge potential for future manufacturing environments.
I hope tesla can find away to offer just a little bit more from the mobileye chips that were installed in the early models S's and X's.
MobilEye doesn't use HD maps in the sense you're thinking. They crowd-source AV maps. Have you seen their REM explanation video?
Fresh meat someone's going to get more subscribers with content like that👍
excellent Sir
So what if road works change the road from one day to the next mobile eye would fail.
Keeping hands on the wheel reduces the reaction time substantially in the event of a needed intervention. With the camera looking at the driver’s eyes, it could happen that the driver has his hands busy and struggles to react in an emergency. The Tesla approach is so much more logical and safer.
Great analysis of how screwed to old OEM’s are.
Wow. No one can do it but Tesla. Great video.
Obviously Tesla has a great attitude/approach. I just wish Tesla would include infrared vision, the reason is that normal vision has its limit seeing in the infrared allows better night time vision, Tesla could easily combine infrared vision with normal vision, and get even more data.
No future in an approach filled with fatal flaws and opaque communication to the public. Wonder when Intel will wake up... Thanks for laying out the argument so beautifully. Look forward to your continuing coverage of Tesla AI and Tesla Vision -- opening up wonderful new avenues for the company to pursue. I along with all of you are thrilled to be along for the ride... hold on tight!!
And the OEM dependency seems to be a major handicap.
How is mobileye getting the data back? Are the cars / chips connected to home wifi or LTE?
Yes, one of the two, whichever the car supports and the user has in its settings
How would FSD licensing really work? Wouldn't VW for example first have to equip a large enough fleet for data collection to train/adjust the AI for all their models?
The FSD wouldn’t be model dependent as such but all models would have to be equipped with the full FSD sensor, computer and control set.
Impressive explanation of mobil eye compared to Tesla FSD. I didn’t realize how different the two driving system was till this podcast. Again impressive explanation. Thxs. I feel 300% better that I have invested 1/2 my retirement money in Tesla. Always a worrying thought you know. You want to invest in the best product