@@gsopoagmle Lol! You surely have never seen Rome traffic! Tel Aviv rush hour is for 98 year old Grandmas. No 2 line parking, no thousands of scooters everywhere. you even have street markings in Tel Aviv! No 5 cars squeezed parallel through an unmarked intersection that has 3 different takings. Rome has streets designed 2000 years ago
at minute 21:45 the system of the av didn't recognize the 60 kmph sign and drove 70 kmph the next few kilometers. Then there was another 60 kmph sign at minute 23:07, but this time the av system recognize it right. Also it turned to the right, but the blinkers weren't on. At minute 23:15 the av system should recognize the inner city sign (left side the yellow one), but unfortenatly it doesnt. When you see a inner city sign without a specific speed limit, then always 50 kmph is allowed. Instead it kept still going to 60 kmph. At minute 24:13 there is the next sign to that 60 kmph is allowed.
Can you imagine the great benefit for older folk who need to get around, and can't drive? or taxi availability during the holidays, or allowing people to go out late, drink and get home safely?
@@peypey_it_is: True. But no reason SERIOUS encryption, etc. can't be used -- just need SERIOUS regulations to strictly enforce that. Perhaps some sort of parity checking or other encoding to quickly detect code/instructions that aren't genuine. Again, not that most companies will go to the expense unless they are forced -- so lawmakers should get a clue and do that.
Overall I'm really impressed by the state of development of this. Other than misreading a lot of speed signs and illegal overtaking on the Autobahn (yes, it's not allowed at this position), it also caught my eye that it misses a lot - I mean, really A LOT - of correct indicating. E.g. I didn't once see it indicating when exiting a roundabout, which is mandatory, at least in Germany.
This is probably their first test in Germany. Test means it is not perfect yet and needs reworks. That is the exact purpose of testing. I mean they are not going to release this technology next month or something. It Wil be released probably 3 years from now. Maybe longer because the legal aspect of full self driving isn't clear yet.
At 08:02 you drove 60 where 30 is allowed... Earlier: You also entered a crossing earlier while cross traffic was still in the crossing, which is not allowed. You did not reduce speed as there where two people clearly trying to cross the road.
@@bencohen6407 Everything in the video is in kmh. Source: I lived in Munich for 30 years and we use kmh (as the rest of Germany... and Europe... and 95% of the world)
@@Fjoergyn1199 Of cause not. It wouldn't be legal as of today. So, the responsible person behind the wheel should have noticed the error of the AI and slowed down the car. But he/she didn't.
@18:52 there is a speed limit sign of 60 kmph, but the car was still driving 80 kmph over that small junktion. So it doesnt recognized the speed limit sign properly.
Also at about minute 17:15 the car didnt turn on the left blink to signalize other cars to change the lane. You couldn't see, if a car was behind the acceleration lane.
53:45 driver almost has a heart attack... :) Awesome job from mobileeye. It will give people even more confidence if in the 3d model the stationary cars parked by the opposition lane are not rendered weirdly with abrupt movement. Buses are detected as trucks in this video too.
This car sees more than my eyes, of course it cannot see everything and must make choices and predictions. In the end, these cars will only need 3-4 sensors I think.
Before the Truck appears it ist pretty strange that the Car accelerate over 50 kmph, which is the max speed limit at that street, because it is in urban. The Car was 58 kmph fast.
Recently on a climbing trip with the Swiss Alpine Club I took the post bus into the Safiental and the roadway is nothing like a Californian Inter-State or a town like Munich. In many places the surface is only one car wide and has no barrier, a mistake leading to a fall of a 300 metres into the river below. Drivers must judge where to pull in to allow oncoming vehicles to pass or whether an approaching driver intends to wait for them. As the post bus has precedence, drivers may be obliged to reverse into whatever ad hoc space one has previously passed. Many of the bends are blind, so one must be a able to stop suddenly on meeting an oncoming vehicle and be attentive for the sound of a car horn or the familiar tune of a post bus. Moreover, as it snows in the winter at 1700 m, drivers must be able to judge the depth of snow, the ability to stop on an icy surface, and in general to deal with a whole raft of unexpected situations. Failure to deal correctly with any of these may have fatal consequences. The problem for self-driving cars is how to deal with special cases. And the argument that the driver must be ready to take over is very misleading. Either you drive or the computer drives, other wise the human loses concentration. It isn't possible to blame the human when the computer fails.
Regarding "Unedited 1-Hour Mobileye in Munich": at 8 min 37 sec, the speedlimit is 60, the car shows 70, seconds later 80 at 9 min 6 sec, end of speed limit 60 km/h, the car shows 80 before this sign Is that the common speed in Munich? Or is the card material to old?
Probably a bug. At 08:36 the car enters the main road with a 60 km/h sign on the right side. After joining the main road the car accelerates till 90 although at 09:07 the car comes across a sign with a crossed 60 km/h. From the roundabout up until the sign at 09:07 the car went to fast.
Probably a bug. Not the only mistake the car made. At 08:36 the car enters the main road with a 60 km/h sign on the right side. After joining the main road the car accelerates till 90 although at 09:07 the car comes across a sign with a crossed 60 km/h. From the roundabout up until the sign at 09:07 the car went too fast.
I only watched until 9:07, but until then there were already some speed violations, he just didn't pay attention to the signs. I am surprised that something like this is put on the net. If German authorities see this, that's not good advertising.
yes it also have cameras and sensors/radar technology which covers the rear parts of the car but of course also the front and side angles. A good example is the situation with the emergency car 41:44 .
6:10 Note 1: The Car didn´t, technically, completely stop before crossing the line, it´s unlikely you´d get into trouble with Police but it could very much happen. Violation 1: The Car entered the junction as another car was approaching. 7:10 Note 2: The Limit in the Display went to 60 while it still was 50?? Driving up to 10 over is mostly accepted but defenitly noteworthy, seems to be happening quite a few times too. 8:02 Violation 2: Driving 54 where only 30 is allowed, wow. 8:40 Violation 3: Limit of 60, in the Display set as 70, going up to 80 and the Car accelerating to 75 11:08 Violation 4: Crossing the Line with Traffic approaching. 12:24 Violation 5: Passing on the right. 15:10 Violation 6: Not moving over to the right Lane, even though there was no traffic and more than enough time to do so. 17:15 Violation 7: Limit of 60, driving 78 and display showing 80. Pretty sure i´ve also seen some missing turn signals, stopped watching at 20 Minutes. Considering that these are basic mistakes to make and that it´s been almost two years i hope it significantly improved.
42:00 Stopped behind an ambulance being on duty, waiting that it starts. After two cars had overtaken the MobileEye car and the ambulance it seems convinced to start it own pass by task.
Most interesting part of the video. Goes to show how many edge cases there are in real life. Object detection/segmentation is the "easy" part. Modeling human behavior, a bit harder.
@@stri8ted the shown trip was only at the outer quarters of Munich (I‘m living here) and these main roads are mainly main road for flowing traffic in and out and around city center. Interesting would be a trip innert the city, starting in -1 floor of a parking house, driving at 7:30am to an elementary school, then to office on the other side of city center (what I would do by underground) - best tangenting the pedestrian areas with a lot of walking workers (having their mind elsewhere or on smartphones ) heading for their work .... and all reverse at 4:00 pm.
Yes that wasn't perfect but hardly something anyone would throw out self-driving for. Also often in those situations the back cars pull ahead because they cut out quickly making it dangerous to switch lanes until they pass.
@@whateverrevetahw6846 Only time will tell... I'm not tesla fan boy enough to think others might do a better job. Looks like tesla is making great progress.
@@monkeysezbegood Tesla's fsd blindly follows their flawed gps map through fences, it's a complete scam. When mobileye's supervision L2+ adas system is released on consumer cars it will vastly outperform 'fsd beta' and be a wake up call for the blind tesla fans.
@@whateverrevetahw6846 you obviously know nothing... this Mobileye video is impressive but they were using a Lidar based system at first. Tesla was the first to go all vision
@@henryabanto2155 Tesla's first autopilot system that musk started claiming would be fully autonomous soon was from mobileye, nearly everything musk says is copied straight from mobileye's plans minus the lidar redundancy.
Pretty good. Interesting to see. I gather the HD mapping is generated from the cars since it was sparse and imperfect. It sees some phantom road work cones but I'm guessing that's from memory of prior passes. Close call at 53:40. Bad truck driver. It saw an opening car door. It did pretty good. Vehicle recognition to the side wasn't entirely stable.
funny tesla FSD would go full on evading here, and in this particular case it would be right thing to do, but in other cases Tesla evasion without slowdown is dangerous, it seems like Tesla FSD always prefer to evade, and mobileye autonomous driving always prefer to stop.
@@sddndsiduae4b-688 I need to watch this more closely, but to me a cursory glance suggests that this Mobileye tech is ahead of Tesla. Am I wrong? The only caveat is that Tesla is literally letting relatively unsophisticated users/owners of Tesla vehicles post videos without controlling them at all, there is not only no editing, there is no withholding of videos. Mobileye obviously has the ability to just choose the best out of many videos and then post it while claiming it is "unedited".
@@paulhamrick3943 they on equal mobileye better drive because of mapping but if some changes on the road ... i.e. in unmapped area Tesla tech is extremely ahead of others.
This is amazing! The software seems to have a preference for the left lane though. And the speed limit recognition seems to be a bit weird. There was no speed limit of 35 or 40. I would like to see the system at night while it is heavily snowing.
i think lane preference is written on HD map, i.e. easy to change globally or for particular street, Tesla FSD for other hand always prefer right lane, especially on one way streets, because it may misunderstand line separator (two way or one way case).
Bei der Verkehrszeichenerkennung kann Mobileye noch nicht sehr weit sein. Ab Minute 8:00 fährt das Auto trotz Beschilderung einer Geschwindigkeitsbegrenzung auf 30 km/h in einem Baustellenbereich mit 60 km/h. Bei Minute 47:20 dann über eine längere Strecke mit 80 statt erlaubter 70. Da will ich kein Passagier sein wenn das nicht funktioniert!
Does the system also detect when a truck is turning left while the car is making a lane change? Sometimes you have to get back in the lane because some idiot drives right in front of the car to change lanes.
Poor editor, hope you muted the program monitor. I mean hey _I like_ this song, but after hearing it so many times... :P Great video you guys, looking forward to more!
@@JJRicks Sadly I don't think so, many of the trackers for the blurred field were doing weird stuff which definitely happens when humans edit... I would love to have a simple "blur license plates" button in my video editor, but I don't think such a thing exists, yet. Should be not very difficult to implement, though, as there are libraries out that for license plate detection.
@@DerKatzeSonne ahh well shoot, guess that's a serious bottleneck. Yea if there was a "license plate blocker" effect in premiere or something that would be useful. I think youtube can do faces automatically, thinking back
Quite capable the system, but too much driving on straight roads and avoiding tricky and complicated situations for example in residential areas, where it would need to make more complex decisions on its own to succeed.
53:40 Slowing down towards an oncoming truck that is protruding into our lane --- wait ... I would want it to move over/to_the_right and not just slow down ... what good is just slowing down?!?
It probably recognized that the car would'nt be hit. But it's these "wierd" scenarios that makes it hard for autonomous vehicles. Picture a scenario where a truck is going to smash into you and the only way to steer out of it is by steering right into a pedestrian. Or a scenario where a motorcycle is going to hit you and you can only steer into a brick wall.
The in car screen with the animated top view and the zoomed view of it on the left hand side are different. Examples: Number of visible cars at 02:40, current speed (permanent). Why? Overall it looks like smooth driving to me.
I like that animated display. I couldn't quite tell, but if not....: a good idea would be to identify possible "threats" with the flashing colors; yellow for caution, red for warning. A "threat" could be -objects (people, animals, etc) moving randomly, such as kids on a playground or crowded sidewalks -a bicyclist or runner on the road -any "aggressive" vector (tail-gaters, lane-cutters, racers) I'd love this tech to grow quickly, my eyesight is awful and I have ZERO desire to be uber-ed around by other people. Like anyone else, I value my independence of movement.
Different approach than Tesla, that's clear and known by all. Maybe there is a market for both of them. If I envision autonomous container trolleys on the Highway that drop their containers from e.e. Rotterdam to a Munich terminal at city boarders, this might be a working solution as well. Nevertheless, HD maps are cumbersome for 100% road coverage. The first car meeting any change (i.e. not on the map yet) must be able to handle it. Once handled and shared, the next one will have an advantage.
I haven't had time to read up on the differences between Mobileye's approach and Tesla's approach. What are those differences exactly? It looks like they are both vision based. I must say, this video suggests to me that Mobileye is very far along, a cursory glance at this video suggests that they are as far along as Tesla is.
@@paulhamrick3943 Quite simple: Testa approaches each road as never seen that particular road before. No maps, no object DB etc. Like a human driving somewhere he has never been before. The computer AI should figure out how to drive based on his generic experience and "lessons learned". Other insert mapping data into the system so it knows what to expect. Without the (HD) map data the system is less to not operating well. Like a really driver without his (sorry, her :-) navigator. No chance to win then.
@@oscarjager OK when Mobileye describes their tech as "camera-only"... they are omitting that they actually use HD maps, although maybe not the same type of HD maps that Waymo uses? Obviously Waymo uses HD maps that are meticulously and intentionally compiled by driving their lidar scanners over every street (and ever parking lot?) in the area that is served (geo fenced). But it sounds like Mobileye is gathering HD maps in a more crowd-sourced way?
That is an astonishing rule that I have never heard of anywhere else. Why is that? I can understand shifting to provide space for vehicles as needed, but not constantly driving off-center.
@@TROPtastic It creates more distance between vehicles. It helps to estimate when it's safe to overtake (yes, you are allowed to shift towards the left of the lane in order to check traffic, and then move back to the right). And before anyone says 'the distance between the 2 centres is the same as between the right edges': theoretically, yes, but in reality people seem unable to drive in the middle of the lane whereas following a white line and driving right next to it turns out to be easier. Anyway: the point is that the manufacturers of self-driving cars don't seem to care about our traffic laws. Perhaps they are counting on it that the laws will be changed in order to align with their software, rather than the other way round.
The big challenge with this tech is that they need to scan ALL the roads before it can work on ALL the roads. If something changes before it is scanned the vehicle will not work on that path. Excellent for often-used roads (highways, and major arteries), but Its unlikely to work anywhere else.
I don't think it followed the other cars. There must be some timer waiting for the stopped vehicle to move and after the timer times out without the other vehicle moving, it overtakes. It'd be interesting to know, though, if the camera could see far enough on the other lane.
It waited to see if the SUV would back up or move forward. Once satisfied the SUV was moving forward at speed, only then did it continue to drive forward. I watched that several times as it is bookmarked at the beginning in the description of the drive (up top).
What can I say ? It is fascinatingly boring to watch it under "fair weather situations"... I was wondering about many potential obstacles that are likely to arise under foul weather like gusty cross winds, fallen branches, homeless people by roadside, black ice , heavy rains,, heavy low fogs (my hometown every Januay) mud slides, any "nothing" obstacles that will make you stop needlessly and risk being whiplashed by cars hitting you from behind, you name the long list.. this wiii make it really fun watching.. Let me know when you made more unedited hours of your Mobileye autonomous vehicle rides. I will keep my eyes peeled..
??? You mean, just driving straight on the motorway without any manoeuvre is awesome? Or the part where it wasn’t sure where that truck was on the right?
Excuse me, but what happens at 30:32 when the car decides to unnecessarily swivel to the other lane? I dont think a car could fit into the remaining space..
8:12 If not already done, maybe pay some attention to "bus stop with bus"-recognition. As far as I see that bus driveby with changing passengers was bit to fast and very close for the Strassenverkehrsordnung: Straßenverkehrs-Ordnung (StVO) § 20 Öffentliche Verkehrsmittel und Schulbusse "(1) An Omnibussen des Linienverkehrs, an Straßenbahnen und an gekennzeichneten Schulbussen, die an Haltestellen (Zeichen 224) halten, darf, auch im Gegenverkehr, nur vorsichtig vorbeigefahren werden." Even more important if hazard warning is enabled or schoolbus sign on it, than even for oncoming traffic: die an Haltestellen (Zeichen 224) halten und Warnblinklicht eingeschaltet haben, darf nur mit Schrittgeschwindigkeit und nur in einem solchen Abstand vorbeigefahren werden, dass eine Gefährdung von Fahrgästen ausgeschlossen ist. Die Schrittgeschwindigkeit gilt auch für den Gegenverkehr auf derselben Fahrbahn At least to give the priority if the bus continiues (5) Omnibussen des Linienverkehrs und Schulbussen ist das Abfahren von gekennzeichneten Haltestellen zu ermöglichen. Wenn nötig, müssen andere Fahrzeuge warten. Quite interesting, not only to read the STVO :)) but what it's about for human and machine to collect and process complex information.
45:34 lost human tracking on the right (at least 0.3 seconds before she was partially occluded). p.s. in average it is on equal with Tesla FSD now (at FSD beta start Tesla was worse on pedestrian detection), need much more hours of ride with surround detection to decide which camera only system is better now. 55:30 positions of temporary road fences is far from perfect.
@@MsAjax409 we can only compare if tesla does this route 1000 times, and mobileye too, and then we can count absolutely necessary intervention (or crashes), for now it is very difficult to compare (not enough info from mobileye and no repeated runs from Tesla on same route), but i did see Tesla crashes on FSD in static obstacles (which would be mapped in HD maps).
This is why I own more Lucid (CCIV) stock than TSLA. Mobileye will get there first with detailed maps and lidar. The safer option will get approved before we can rely only on AI.
You're right, but I think in this case it's not illegal, because the lanes were already split up. Technically and legally this wasn't passing on the right.
@@Schreibtisch1 I think that if the lanes were already splitted up they would have to be marked with a continuous line. From my personal german driver perspective I would agree that this is a fault.
@@toniengel6908 Die StVo sagt das Gegenteil, laut § 7a Absatz 1 ist es erlaubt in diesem Fall erlaubt. Also das gilt technisch nicht als Rechtsüberholen.
@@Schreibtisch1 Ok, nicht nur bei durchgezogener, sondern auch "von Beginn der breiten Leitlinie rechts von dieser". Der Überholvorgang beginnt aber schon vorher bei exakt 12:26 und findet dann später links von der breiten Linie statt und auch ohne Abbiegen. § 7a Absatz 1 belegt eigentlich den Regelverstoß - das ist ja keine Riesensache und passiert öfters, sollte halt angepasst werden... Insgesamt ist das Fahrverhalten natürlich sehr beeindruckend!
@@joeyng1282 Edge cases. Ghost drivers. Demonstrations/jaywalking. Weird road construction with bad signs. Different kinds of debris on the road. Avoiding a redlight driver heading for collision. Car heading for rear ending. Heavy fog, rain and snowstorm. Dirt roads. Going through gates. Boarding a ferry. Underground parking lots. Bad roads with very low vision parts/merges. Not driving like a grandma that will cause traffic jams with its slow reaction times if everyone drived like it. The list goes on. Last but not least: video made by independent party also showcasing its biggest flaws rather than what we must assume is its near-peak performance.
@@gsopoagmle You indirectly accuse me of confirmation bias, but I'm just being is skeptical. And trying to voice that this clearly isn't enough information to conclude much. Being skeptical of extraordinary claims is by no means unreasonable. And I'm sure their test in India will go great. But we've had the technology to build localized autonomous driving for more than a decade, so I'm more interested in understanding its limits than seeing it work in single scenarios.
@@Muskar2 Mobile eye founder is an expert in artificial brain and cognitive intelligence. He earned his PhD on this from MIT in 1993. Six years later he founded his company. I think he knows what the challenges are in this area, certainly because he lives in Israel. Their highways are excellent but the inner cities road and traffic are completely different compared to for example the Netherlands. I know that because I know both countries quite well. Yes you can expect that he knows what bad road conditions are. The fact that car drives like a grandma is completely understandable. The most stupid thing you can do during the test is getting your company and technology canceled because your car caused a deadly accident. The car looks also slow because it is driving according to the speed limit while many other drivers always ignore the limit.
Way too much positional pop-in. They should do a better image recognition and classification, and then tie that in and correlate with actual lidar sensor. Use AI.
HD maps are dumb... You just had exactly this one route mapped. All side-streets are un-mapped, so you can't drive on them. Elon is right when he says that HD-maps are cheating.
@@acemav Well, it's actually true. Refusing to use LIDAR and only relying on cameras + AI does make the computational part much more complicated, but the hardware is cheaper and less complex and some LIDAR systems do look very ugly. The question is whether that is a good trade-off. Elon Musk seems to think so, I am less convinced ...
@@Everest314 I believe there is no lidar in this mobileye car. This is impressive what they do. However you have to keep 2 things in mind: 1) tesla does not share info between cars to have updated HD map just yet. They have all they need to implement that in the future if needed, but the fact Tesla's doesn't know the road ahead except for the google maps low detailed features is way more impressive. 2) you can't compare a commercial footage made by the company that makes the system to the youtube FSD footage where youtubers tries to see the limitation of the system. Actually, Tesla have shown multiple long footage where FSD behave properly, as good as mobileye shows here if not better.
@@bboybasics2 So only radar and cameras? I don't know the specifics of this particular test, but afaik there is no plan for Mobileye to drop the lidar. 1) Well, in the end this is not about impressing someone. That Tesla makes it more difficult for themselves than they have to is imo more foolish than impressive. (Not saying their end product is bad, just questioning some decisions without which it should be even better.) 2) I am not reading anything into promotional material. In fact (having only skipped through this video), I haven't seen any situation here that should test the limits of a finished FSD system.
The number of robotic commenters here saying things like, "Awesome! I would like to purchase this immediately! We agree this is amazing technology!" Is a fucking embarrassment.
No bullshit, No PPT, No Exaggeration, No fancy event and No false promise.
Just a driving an hour as usual, COOL.
No views either, can you imagine how many views a video of tesla doing this would get?
cuz its Bullshit long term ? Lidar has no chance vs AI Vision camera in tesla.. get some education, thats why
@@Anonymous-vd5yd This is a 100% AI Vision Cameras-only AV test vehicle from Mobileye, by the way.
It’s just that they map out the roads and sense they did that I don’t really think Tesla has any competitors
@@Anonymous-vd5yd: It doesn't try to crash frequently in complex situations, unlike Tesla. So there's that.
Take your own advice.
Excellent! I can't wait for this to get to market place. This is the technology people want!
If you want to drive in Munich, and on this street exactly. But you have to be patient, the police will stop you and ask for your speed limits!
Also watch what a human-like decision, great. watch at 42.15 - 42.35. Big thumbs up. way ahead to its competition.
Very impressive. Please try next time Rome or Milan in rush hour! That would be fun ;)
I doubt Rome or Milan is more chaotic than tel aviv. This full self driving drove in tel aviv successfully during the rush hour.
@@gsopoagmle Lol! You surely have never seen Rome traffic! Tel Aviv rush hour is for 98 year old Grandmas. No 2 line parking, no thousands of scooters everywhere. you even have street markings in Tel Aviv! No 5 cars squeezed parallel through an unmarked intersection that has 3 different takings. Rome has streets designed 2000 years ago
at minute 21:45 the system of the av didn't recognize the 60 kmph sign and drove 70 kmph the next few kilometers. Then there was another 60 kmph sign at minute 23:07, but this time the av system recognize it right. Also it turned to the right, but the blinkers weren't on. At minute 23:15 the av system should recognize the inner city sign (left side the yellow one), but unfortenatly it doesnt. When you see a inner city sign without a specific speed limit, then always 50 kmph is allowed. Instead it kept still going to 60 kmph. At minute 24:13 there is the next sign to that 60 kmph is allowed.
Can you imagine the great benefit for older folk who need to get around, and can't drive? or taxi availability during the holidays, or allowing people to go out late, drink and get home safely?
Don't forget the benefit of eliminating human error. Less bad drivers on the road and better traffic flow.
@@rp7296 beware the cyber
@@peypey_it_is beware the human lol
@@peypey_it_is: True. But no reason SERIOUS encryption, etc. can't be used -- just need SERIOUS regulations to strictly enforce that. Perhaps some sort of parity checking or other encoding to quickly detect code/instructions that aren't genuine.
Again, not that most companies will go to the expense unless they are forced -- so lawmakers should get a clue and do that.
Can we have the same demo in busy city centre, on narrow roads with unplanned roadworks? Would be more interesting than these boring straight lines...
Here is a demo from a busy city center, including narrow roads: ua-cam.com/video/kJD5R_yQ9aw/v-deo.html
Omg that's awesome! Thank you for improving the future
lol
Overall I'm really impressed by the state of development of this. Other than misreading a lot of speed signs and illegal overtaking on the Autobahn (yes, it's not allowed at this position), it also caught my eye that it misses a lot - I mean, really A LOT - of correct indicating. E.g. I didn't once see it indicating when exiting a roundabout, which is mandatory, at least in Germany.
This is probably their first test in Germany. Test means it is not perfect yet and needs reworks. That is the exact purpose of testing. I mean they are not going to release this technology next month or something. It Wil be released probably 3 years from now. Maybe longer because the legal aspect of full self driving isn't clear yet.
At 08:02 you drove 60 where 30 is allowed... Earlier: You also entered a crossing earlier while cross traffic was still in the crossing, which is not allowed. You did not reduce speed as there where two people clearly trying to cross the road.
'you drove' ? Are you talking to the AI driving the car?
30mph, not kmh.
@@bencohen6407 Everything in the video is in kmh. Source: I lived in Munich for 30 years and we use kmh (as the rest of Germany... and Europe... and 95% of the world)
Why mph? *confused*
@@Fjoergyn1199 Of cause not. It wouldn't be legal as of today. So, the responsible person behind the wheel should have noticed the error of the AI and slowed down the car. But he/she didn't.
This will be in all future Nio cars! Amazing
Will Nio install this technology on their FSD cars?
@18:52 there is a speed limit sign of 60 kmph, but the car was still driving 80 kmph over that small junktion. So it doesnt recognized the speed limit sign properly.
Also at about minute 17:15 the car didnt turn on the left blink to signalize other cars to change the lane. You couldn't see, if a car was behind the acceleration lane.
Awesome! Especially the traffic lights detection ;)
Is it detected or mapped data?
Highly interested, how the system will behave, when it's on "Autonomus Mode: Sport Mode" :D
Can't wait to watch this! Let's goooo Mobileye!
53:45 driver almost has a heart attack... :)
Awesome job from mobileeye. It will give people even more confidence if in the 3d model the stationary cars parked by the opposition lane are not rendered weirdly with abrupt movement. Buses are detected as trucks in this video too.
That was indeed too close for my comfort. 😱
This car sees more than my eyes, of course it cannot see everything and must make choices and predictions. In the end, these cars will only need 3-4 sensors I think.
I think the cameras that detect other cars are too low so they are blocked constantly by many obstacles.
Before the Truck appears it ist pretty strange that the Car accelerate over 50 kmph, which is the max speed limit at that street, because it is in urban. The Car was 58 kmph fast.
Recently on a climbing trip with the Swiss Alpine Club I took the post bus into the Safiental and the roadway is nothing like a Californian Inter-State or a town like Munich. In many places the surface is only one car wide and has no barrier, a mistake leading to a fall of a 300 metres into the river below. Drivers must judge where to pull in to allow oncoming vehicles to pass or whether an approaching driver intends to wait for them. As the post bus has precedence, drivers may be obliged to reverse into whatever ad hoc space one has previously passed. Many of the bends are blind, so one must be a able to stop suddenly on meeting an oncoming vehicle and be attentive for the sound of a car horn or the familiar tune of a post bus. Moreover, as it snows in the winter at 1700 m, drivers must be able to judge the depth of snow, the ability to stop on an icy surface, and in general to deal with a whole raft of unexpected situations. Failure to deal correctly with any of these may have fatal consequences.
The problem for self-driving cars is how to deal with special cases. And the argument that the driver must be ready to take over is very misleading. Either you drive or the computer drives, other wise the human loses concentration. It isn't possible to blame the human when the computer fails.
Good example was the ambulance. It was working in a defense way.
Regarding "Unedited 1-Hour Mobileye in Munich":
at 8 min 37 sec, the speedlimit is 60, the car shows 70, seconds later 80
at 9 min 6 sec, end of speed limit 60 km/h, the car shows 80 before this sign
Is that the common speed in Munich? Or is the card material to old?
Gute Augen Ralf!
Probably a bug. At 08:36 the car enters the main road with a 60 km/h sign on the right side. After joining the main road the car accelerates till 90 although at 09:07 the car comes across a sign with a crossed 60 km/h. From the roundabout up until the sign at 09:07 the car went to fast.
At 48:12 the speed limit sign on the street shows 70, but the car kept 80. Why?
Probably a bug. Not the only mistake the car made. At 08:36 the car enters the main road with a 60 km/h sign on the right side. After joining the main road the car accelerates till 90 although at 09:07 the car comes across a sign with a crossed 60 km/h. From the roundabout up until the sign at 09:07 the car went too fast.
I only watched until 9:07, but until then there were already some speed violations, he just didn't pay attention to the signs. I am surprised that something like this is put on the net. If German authorities see this, that's not good advertising.
Safe, confident drive. Well done Mobileye.
Looks great, please add it to my Peugeot.
Does the vehicle have a rear camera as wel in order to evaluate changing line is safe?
yes it also have cameras and sensors/radar technology which covers the rear parts of the car but of course also the front and side angles. A good example is the situation with the emergency car 41:44 .
6:10 Note 1: The Car didn´t, technically, completely stop before crossing the line, it´s unlikely you´d get into trouble with Police but it could very much happen.
Violation 1: The Car entered the junction as another car was approaching.
7:10 Note 2: The Limit in the Display went to 60 while it still was 50?? Driving up to 10 over is mostly accepted but defenitly noteworthy, seems to be happening quite a few times too.
8:02 Violation 2: Driving 54 where only 30 is allowed, wow.
8:40 Violation 3: Limit of 60, in the Display set as 70, going up to 80 and the Car accelerating to 75
11:08 Violation 4: Crossing the Line with Traffic approaching.
12:24 Violation 5: Passing on the right.
15:10 Violation 6: Not moving over to the right Lane, even though there was no traffic and more than enough time to do so.
17:15 Violation 7: Limit of 60, driving 78 and display showing 80.
Pretty sure i´ve also seen some missing turn signals, stopped watching at 20 Minutes.
Considering that these are basic mistakes to make and that it´s been almost two years i hope it significantly improved.
I am keen on seeing the solution on snowy roads
@11:05 = balls of steel
Can it drop the passenger off, then go find a parking or drive around, then come back to pick them up later?
I think technology-wise it would be possible, but law-wise not (at least in Germany)
Very nice! But around minute 9 the car is going almost 80 km/h, even though the limit is at 60 km/h. You should fix that bug!
Is this EyeQ5?
Yes.
42:00 Stopped behind an ambulance being on duty, waiting that it starts. After two cars had overtaken the MobileEye car and the ambulance it seems convinced to start it own pass by task.
Most interesting part of the video. Goes to show how many edge cases there are in real life. Object detection/segmentation is the "easy" part. Modeling human behavior, a bit harder.
@@stri8ted the shown trip was only at the outer quarters of Munich (I‘m living here) and these main roads are mainly main road for flowing traffic in and out and around city center.
Interesting would be a trip innert the city, starting in -1 floor of a parking house, driving at 7:30am to an elementary school, then to office on the other side of city center (what I would do by underground) - best tangenting the pedestrian areas with a lot of walking workers (having their mind elsewhere or on smartphones ) heading for their work .... and all reverse at 4:00 pm.
It waited too long, and once it started to pass, it went too slowly. Once it had clear view of the road ahead, it should have sped up.
Yes that wasn't perfect but hardly something anyone would throw out self-driving for. Also often in those situations the back cars pull ahead because they cut out quickly making it dangerous to switch lanes until they pass.
Mobileye vs Tesla FSD is going to be really interesting to watch unfold.
Mobileye is real, tesla's fsd is just musk's scam based on a cheap imitation of mobileye's system.
@@whateverrevetahw6846 Only time will tell... I'm not tesla fan boy enough to think others might do a better job. Looks like tesla is making great progress.
@@monkeysezbegood Tesla's fsd blindly follows their flawed gps map through fences, it's a complete scam. When mobileye's supervision L2+ adas system is released on consumer cars it will vastly outperform 'fsd beta' and be a wake up call for the blind tesla fans.
@@whateverrevetahw6846 you obviously know nothing... this Mobileye video is impressive but they were using a Lidar based system at first. Tesla was the first to go all vision
@@henryabanto2155 Tesla's first autopilot system that musk started claiming would be fully autonomous soon was from mobileye, nearly everything musk says is copied straight from mobileye's plans minus the lidar redundancy.
Great. Speedsign recognition needs to be approved, but the system seems to be at a very good state.
Oh my! That is so remarkable
Wow. Great video.
Awesome Intel... :-)
This is not Intel, it is Mobile Eye. Israelian company, just bought by Intel.
@@bernios3446 ya true.....Mobileye is an Intel company....
Any unprotected left turns within this video?
Yes
@@georgesideropoulos6704 If only you posted a timestamp :)
Pretty good. Interesting to see. I gather the HD mapping is generated from the cars since it was sparse and imperfect. It sees some phantom road work cones but I'm guessing that's from memory of prior passes. Close call at 53:40. Bad truck driver. It saw an opening car door. It did pretty good. Vehicle recognition to the side wasn't entirely stable.
funny tesla FSD would go full on evading here, and in this particular case it would be right thing to do,
but in other cases Tesla evasion without slowdown is dangerous, it seems like Tesla FSD always prefer to evade, and mobileye autonomous driving always prefer to stop.
@@sddndsiduae4b-688 I need to watch this more closely, but to me a cursory glance suggests that this Mobileye tech is ahead of Tesla. Am I wrong?
The only caveat is that Tesla is literally letting relatively unsophisticated users/owners of Tesla vehicles post videos without controlling them at all, there is not only no editing, there is no withholding of videos. Mobileye obviously has the ability to just choose the best out of many videos and then post it while claiming it is "unedited".
@@paulhamrick3943 they on equal mobileye better drive because of mapping but if some changes on the road ... i.e. in unmapped area Tesla tech is extremely ahead of others.
This is amazing!
The software seems to have a preference for the left lane though.
And the speed limit recognition seems to be a bit weird. There was no speed limit of 35 or 40.
I would like to see the system at night while it is heavily snowing.
Are you thinking in MPH or KPH? This is in Germany where speed limits are always in KPH
i think lane preference is written on HD map, i.e. easy to change globally or for particular street,
Tesla FSD for other hand always prefer right lane, especially on one way streets, because it may misunderstand line separator (two way or one way case).
Bei der Verkehrszeichenerkennung kann Mobileye noch nicht sehr weit sein. Ab Minute 8:00 fährt das Auto trotz Beschilderung einer Geschwindigkeitsbegrenzung auf 30 km/h in einem Baustellenbereich mit 60 km/h. Bei Minute 47:20 dann über eine längere Strecke mit 80 statt erlaubter 70. Da will ich kein Passagier sein wenn das nicht funktioniert!
Amazing. Mobileye looks better than Tesla Auto Pilot!
Does the system also detect when a truck is turning left while the car is making a lane change?
Sometimes you have to get back in the lane because some idiot drives right in front of the car to change lanes.
The drive through Jerusalem was much harder and impressive. There were only a handful of 90 degree turns in the whole vid.
Yeah but this was done with only 1 week of testing in Germany.
When is this software ready for mass market! Great job!
Poor editor, hope you muted the program monitor. I mean hey _I like_ this song, but after hearing it so many times... :P Great video you guys, looking forward to more!
Poor editor already had to focus on blurring all the license plates, no more time for finding more music! :D
@@DerKatzeSonne Oh yeah! I hope since it's intel they had some sort of AI solution for that, haha
@@JJRicks Sadly I don't think so, many of the trackers for the blurred field were doing weird stuff which definitely happens when humans edit... I would love to have a simple "blur license plates" button in my video editor, but I don't think such a thing exists, yet. Should be not very difficult to implement, though, as there are libraries out that for license plate detection.
@@DerKatzeSonne ahh well shoot, guess that's a serious bottleneck. Yea if there was a "license plate blocker" effect in premiere or something that would be useful. I think youtube can do faces automatically, thinking back
Quite capable the system, but too much driving on straight roads and avoiding tricky and complicated situations for example in residential areas, where it would need to make more complex decisions on its own to succeed.
I guess you didn't watch the whole video. Right near the beginning it made what I would consider a very challenging left onto an S turn street.
Awesome!
Viewing number even less than a popular Tesla fanboy video... So much underated
this is really true, a tesla video like this would have 30 million views
Das Limit war nicht korrekt in der Schleissheimerstr. sind nur 50 erlaubt.
When will the stock come to Frankfurt, Germany?
its Intel Stock they bought them
53:40 Slowing down towards an oncoming truck that is protruding into our lane --- wait ... I would want it to move over/to_the_right and not just slow down ... what good is just slowing down?!?
It probably recognized that the car would'nt be hit. But it's these "wierd" scenarios that makes it hard for autonomous vehicles. Picture a scenario where a truck is going to smash into you and the only way to steer out of it is by steering right into a pedestrian. Or a scenario where a motorcycle is going to hit you and you can only steer into a brick wall.
The in car screen with the animated top view and the zoomed view of it on the left hand side are different. Examples: Number of visible cars at 02:40, current speed (permanent). Why?
Overall it looks like smooth driving to me.
lol
The images are not in sync.
I like that animated display.
I couldn't quite tell, but if not....: a good idea would be to identify possible "threats" with the flashing colors; yellow for caution, red for warning.
A "threat" could be
-objects (people, animals, etc) moving randomly, such as kids on a playground or crowded sidewalks
-a bicyclist or runner on the road
-any "aggressive" vector (tail-gaters, lane-cutters, racers)
I'd love this tech to grow quickly, my eyesight is awful and I have ZERO desire to be uber-ed around by other people. Like anyone else, I value my independence of movement.
Different approach than Tesla, that's clear and known by all. Maybe there is a market for both of them. If I envision autonomous container trolleys on the Highway that drop their containers from e.e. Rotterdam to a Munich terminal at city boarders, this might be a working solution as well. Nevertheless, HD maps are cumbersome for 100% road coverage. The first car meeting any change (i.e. not on the map yet) must be able to handle it. Once handled and shared, the next one will have an advantage.
I haven't had time to read up on the differences between Mobileye's approach and Tesla's approach. What are those differences exactly? It looks like they are both vision based.
I must say, this video suggests to me that Mobileye is very far along, a cursory glance at this video suggests that they are as far along as Tesla is.
@@paulhamrick3943 Quite simple: Testa approaches each road as never seen that particular road before. No maps, no object DB etc. Like a human driving somewhere he has never been before. The computer AI should figure out how to drive based on his generic experience and "lessons learned".
Other insert mapping data into the system so it knows what to expect. Without the (HD) map data the system is less to not operating well. Like a really driver without his (sorry, her :-) navigator. No chance to win then.
@@oscarjager OK when Mobileye describes their tech as "camera-only"... they are omitting that they actually use HD maps, although maybe not the same type of HD maps that Waymo uses?
Obviously Waymo uses HD maps that are meticulously and intentionally compiled by driving their lidar scanners over every street (and ever parking lot?) in the area that is served (geo fenced). But it sounds like Mobileye is gathering HD maps in a more crowd-sourced way?
31:00 why so far behind the line?
to see the red light right over the car?
@ there are several traffic lights that are visible and also recognized by the system
Keeping the vehicle exactly in the middle of the lane is illegal in Belgium. Art. 9.3.1
That is an astonishing rule that I have never heard of anywhere else. Why is that? I can understand shifting to provide space for vehicles as needed, but not constantly driving off-center.
@@TROPtastic It creates more distance between vehicles. It helps to estimate when it's safe to overtake (yes, you are allowed to shift towards the left of the lane in order to check traffic, and then move back to the right). And before anyone says 'the distance between the 2 centres is the same as between the right edges': theoretically, yes, but in reality people seem unable to drive in the middle of the lane whereas following a white line and driving right next to it turns out to be easier.
Anyway: the point is that the manufacturers of self-driving cars don't seem to care about our traffic laws. Perhaps they are counting on it that the laws will be changed in order to align with their software, rather than the other way round.
What if a big object like a tree limb in the road?
woow, impressive !!!!!
I see you, sir! Watching.
The big challenge with this tech is that they need to scan ALL the roads before it can work on ALL the roads. If something changes before it is scanned the vehicle will not work on that path. Excellent for often-used roads (highways, and major arteries), but Its unlikely to work anywhere else.
At 40 minutes, if real drivers had not paved the way around the ambulance, what would the car have done, and what was it doing on the lead up?
I don't think it followed the other cars. There must be some timer waiting for the stopped vehicle to move and after the timer times out without the other vehicle moving, it overtakes. It'd be interesting to know, though, if the camera could see far enough on the other lane.
39:30 it blocks the Mercedes SUV from parallel parking.
It waited to see if the SUV would back up or move forward. Once satisfied the SUV was moving forward at speed, only then did it continue to drive forward. I watched that several times as it is bookmarked at the beginning in the description of the drive (up top).
@@pats8597 It was obvious from much further back that the car was going to parallel park. It would have been nicer to give them a bit more room.
A car is right behind it. I can't say anyone ever courtesy reversed for me when I've parallel parked.
@@samuelboyle8627 It didn't have to reverse. It just had to stop sooner. The SUV was using the right turn signal to indicate it wanted to park there.
Awesome
What can I say ? It is fascinatingly boring to watch it under "fair weather situations"... I was wondering about many potential obstacles that are likely to arise under foul weather like gusty cross winds, fallen branches, homeless people by roadside, black ice , heavy rains,, heavy low fogs (my hometown every Januay) mud slides, any "nothing" obstacles that will make you stop needlessly and risk being whiplashed by cars hitting you from behind, you name the long list.. this wiii make it really fun watching.. Let me know when you made more unedited hours of your Mobileye autonomous vehicle rides. I will keep my eyes peeled..
Thank, you will see it soon
Great technology, I now wonder when would it be available in my Ford Fusion?
Great great great!
Music not corporate enough, sorry.
Automated driving. Coming soon
Super
On the Autobahn it is strictly forbidden to overtake on the right side. This happened one time.
All in all good job. But at 42' driving strategy is a bit too conservative in comparison to other cars which caused them to take over.
Some poor guy had to blur all the car signs :O
Some guy or gal had to program the computer to do that automatically.
Another reason I'm buying INTC while I can still get them for less. People think Intel is falling behind but they always make up lost ground.
That's definitely true. Intel is a great company, it will take time, but it is going to catch up to its competitors
This is the same reason I bought more intel today.
That is exactly what I have been doing from low 40 this year as well. Ppl seem to forgot about intel owns mobileye. My 2nd biggest position now.
Watch the time between 22.10 - 22.15. Awsome, Save. Great AI. Superb. It is nextgen folks.
???
You mean, just driving straight on the motorway without any manoeuvre is awesome? Or the part where it wasn’t sure where that truck was on the right?
Excuse me, but what happens at 30:32 when the car decides to unnecessarily swivel to the other lane? I dont think a car could fit into the remaining space..
The car stopping on the right was a lil in the lane, so it swiveled to leve enough space to it.
@@TBFSJjunior ooh I see now thanks for this explanation 😊👍
cool at 5:43 it detects door opening and person.
That guy doesn't know how lucky he was.
8:02 Speed limit to 30km/h and system didn't see it...
8:12 If not already done, maybe pay some attention to "bus stop with bus"-recognition. As far as I see that bus driveby with changing passengers was bit to fast and very close for the Strassenverkehrsordnung: Straßenverkehrs-Ordnung (StVO) § 20 Öffentliche Verkehrsmittel und Schulbusse
"(1) An Omnibussen des Linienverkehrs, an Straßenbahnen und an gekennzeichneten Schulbussen, die an Haltestellen (Zeichen 224) halten, darf, auch im Gegenverkehr, nur vorsichtig vorbeigefahren werden."
Even more important if hazard warning is enabled or schoolbus sign on it, than even for oncoming traffic:
die an Haltestellen (Zeichen 224) halten und Warnblinklicht eingeschaltet haben, darf nur mit Schrittgeschwindigkeit und nur in einem solchen Abstand vorbeigefahren werden, dass eine Gefährdung von Fahrgästen ausgeschlossen ist. Die Schrittgeschwindigkeit gilt auch für den Gegenverkehr auf derselben Fahrbahn
At least to give the priority if the bus continiues
(5) Omnibussen des Linienverkehrs und Schulbussen ist das Abfahren von gekennzeichneten Haltestellen zu ermöglichen. Wenn nötig, müssen andere Fahrzeuge warten.
Quite interesting, not only to read the STVO :)) but what it's about for human and machine to collect and process complex information.
WOW! 5:40
The system works so good
Wow
45:34 lost human tracking on the right (at least 0.3 seconds before she was partially occluded).
p.s. in average it is on equal with Tesla FSD now (at FSD beta start Tesla was worse on pedestrian detection),
need much more hours of ride with surround detection to decide which camera only system is better now.
55:30 positions of temporary road fences is far from perfect.
I disagree. Tesla FSD Beta could do this route without HD maps.
@@MsAjax409 we can only compare if tesla does this route 1000 times, and mobileye too, and then we can count absolutely necessary intervention (or crashes), for now it is very difficult to compare (not enough info from mobileye and no repeated runs from Tesla on same route), but i did see Tesla crashes on FSD in static obstacles (which would be mapped in HD maps).
Was the driver alone in this car?
Thanks
Is this more robust than the Tesla system?
yes.
we will see
39:19 me, showing my intentions, so everyone can equip themselfes.
The following.. :))
speed limit recocnition is very bad, when 30 often limit shown was 40 or higher. Thats not acceptable
please any proof of the video at which minutes this behaivour occured. thanks
Sorry i have problem with error and its working so how can i solve this problem or reset?
Nice, I would love to see it drive in within city roads without lane marking
Not a big challenge but it is still impressive,notr like New York
This is why I own more Lucid (CCIV) stock than TSLA. Mobileye will get there first with detailed maps and lidar. The safer option will get approved before we can rely only on AI.
Hello, Lucid is no longer working with Mobileye & Lucid is doing their own in house ADAS.
Mobile eye is also using AI besides the human created algorithm.
12:20 in Germany is not allowed to pass another vehicle at the right side, in this case a truck.
You're right, but I think in this case it's not illegal, because the lanes were already split up. Technically and legally this wasn't passing on the right.
@@Schreibtisch1 I think that if the lanes were already splitted up they would have to be marked with a continuous line. From my personal german driver perspective I would agree that this is a fault.
@@toniengel6908 Die StVo sagt das Gegenteil, laut § 7a Absatz 1 ist es erlaubt in diesem Fall erlaubt. Also das gilt technisch nicht als Rechtsüberholen.
@@Schreibtisch1 Ok, nicht nur bei durchgezogener, sondern auch "von Beginn der breiten Leitlinie rechts von dieser". Der Überholvorgang beginnt aber schon vorher bei exakt 12:26 und findet dann später links von der breiten Linie statt und auch ohne Abbiegen. § 7a Absatz 1 belegt eigentlich den Regelverstoß - das ist ja keine Riesensache und passiert öfters, sollte halt angepasst werden... Insgesamt ist das Fahrverhalten natürlich sehr beeindruckend!
The technology is still very primitive I'm guessing or else they would have shown us way more. But I like what I'm seeing.
What do you expect more?
@@joeyng1282 Edge cases. Ghost drivers. Demonstrations/jaywalking. Weird road construction with bad signs. Different kinds of debris on the road. Avoiding a redlight driver heading for collision. Car heading for rear ending. Heavy fog, rain and snowstorm. Dirt roads. Going through gates. Boarding a ferry. Underground parking lots. Bad roads with very low vision parts/merges. Not driving like a grandma that will cause traffic jams with its slow reaction times if everyone drived like it. The list goes on. Last but not least: video made by independent party also showcasing its biggest flaws rather than what we must assume is its near-peak performance.
@@Muskar2
The next test will be in India.
You will see what you want to see.
@@gsopoagmle You indirectly accuse me of confirmation bias, but I'm just being is skeptical. And trying to voice that this clearly isn't enough information to conclude much. Being skeptical of extraordinary claims is by no means unreasonable. And I'm sure their test in India will go great. But we've had the technology to build localized autonomous driving for more than a decade, so I'm more interested in understanding its limits than seeing it work in single scenarios.
@@Muskar2
Mobile eye founder is an expert in artificial brain and cognitive intelligence. He earned his PhD on this from MIT in 1993. Six years later he founded his company.
I think he knows what the challenges are in this area, certainly because he lives in Israel. Their highways are excellent but the inner cities road and traffic are completely different compared to for example the Netherlands. I know that because I know both countries quite well.
Yes you can expect that he knows what bad road conditions are.
The fact that car drives like a grandma is completely understandable. The most stupid thing you can do during the test is getting your company and technology canceled because your car caused a deadly accident. The car looks also slow because it is driving according to the speed limit while many other drivers always ignore the limit.
Way too much positional pop-in.
They should do a better image recognition and classification, and then tie that in and correlate with actual lidar sensor.
Use AI.
Watch out Tesla
Check out FSD Beta 10.8 videos and say that again ^^
@ Lol I've seen what 10.8 had to offer and have to laugh.
not bad, not great
HD maps are dumb... You just had exactly this one route mapped. All side-streets are un-mapped, so you can't drive on them. Elon is right when he says that HD-maps are cheating.
@dimonat yes. And what they say is wrong. HD maps is a dead end.
This isn't a level 5 autonomoous. It still has a driver and a steering wheel.
If you had listened well, they talk about level 4. Level 5 has to pass regulation, and as far as I know, it hasn't happened anywhere in the world.
Funny, you never see a Tesla capable of doing this. Better buy Intel stock asap.
The problem Tesla is working on solving is 100 times more difficult than this HD map stuff.
@@TheKimNeeper lol. only a fanboy would believe that.
@@acemav Well, it's actually true. Refusing to use LIDAR and only relying on cameras + AI does make the computational part much more complicated, but the hardware is cheaper and less complex and some LIDAR systems do look very ugly. The question is whether that is a good trade-off. Elon Musk seems to think so, I am less convinced ...
@@Everest314 I believe there is no lidar in this mobileye car. This is impressive what they do. However you have to keep 2 things in mind:
1) tesla does not share info between cars to have updated HD map just yet. They have all they need to implement that in the future if needed, but the fact Tesla's doesn't know the road ahead except for the google maps low detailed features is way more impressive.
2) you can't compare a commercial footage made by the company that makes the system to the youtube FSD footage where youtubers tries to see the limitation of the system. Actually, Tesla have shown multiple long footage where FSD behave properly, as good as mobileye shows here if not better.
@@bboybasics2 So only radar and cameras? I don't know the specifics of this particular test, but afaik there is no plan for Mobileye to drop the lidar.
1) Well, in the end this is not about impressing someone. That Tesla makes it more difficult for themselves than they have to is imo more foolish than impressive. (Not saying their end product is bad, just questioning some decisions without which it should be even better.)
2) I am not reading anything into promotional material. In fact (having only skipped through this video), I haven't seen any situation here that should test the limits of a finished FSD system.
Other cars in the view are glitching in and out, getting longer and shorter, twisting and turning, and changing from trucks to buses and vans.
I think the cameras that detect other cars are too low, so they get blocked easily.
The number of robotic commenters here saying things like, "Awesome! I would like to purchase this immediately! We agree this is amazing technology!" Is a fucking embarrassment.
extremely impressive mate.
At least 2-3 years behind Tesla FSD plus DOJO system.
Clueless fanboy
Tesla will win fsd without breaking a sweat...
Dream On.
How is it possible that you are overtaking a truck on the right lane? These rules are extremely simple and you failed!