Separate names with a comma.
Discussion in 'Autopilot & Autonomous/FSD' started by Bladerskb, Jan 6, 2020.
Unedited Ride in Mobileye’s Camera-Driven Autonomous Vehicle
Yandex autonomous car ride
I realize Mobileye’s one is HD map based, but it’s so much more advanced in handling different really hard and uncontrolled driving situations compared to Tesla’s autonomous day video which was basically an example of very bland and standard trip.
Low-speed maneuvers and taking turns with another car moving inch by inch to finally define who should yield is the most impressive.
And just look at the perfect parked cars detection!
Given that Tesla still cannot even break responsibly for the parked fire trucks, I feel a bit worried. We had autonomy day the last spring, and yet Tesla is still fighting with stop signals and signs.
I own Tesla and I don’t want to be pessimistic, but seeing such a tremendous achievement from the company Tesla ceased to cooperate makes me question the appropriateness of such step..
I really hope Tesla will publish some current-state AP footage from a place like SF or LA soon.
I definitely noticed a few glitches in cars on the display, like cars blinking in and out. But overall, extremely impressive. It is much more advanced than what Tesla has.
Yeah, it's a bummer. I love my Model 3 and I love Autopilot but I think the Mobileye demo is definitely proof that Tesla would be further along with FSD if they have continued working with Mobileye. The demo is essentially a taste of the FSD we probably would have today if Tesla had continued working with Mobileye.
I suspect that Tesla and Mobileye broke up because Mobileye knew that the HW2 hardware was not going to be good enough and did not want to partner with Tesla if they were going to insist on insufficient hardware. I say this because Mobileye's demo uses 12 high res cameras instead of the 8 low res cameras that HW2 uses. And if we look at the Lucid Air which is also partnering with Mobileye, it uses 8 cameras but also uses 6 radar and 4 lidar. So clearly, Mobileye's approach uses more sensors and better hardware than what Tesla wants to use.
I also suspect that the break up happened because Elon wanted to go with his vertical integration approach of doing everything in-house. So Elon wanted Tesla to develop their own perception NN and develop their own computer chip and not rely on an external company. Of course, the downside of this approach is that Tesla has had to basically start from scratch and redo a lot of what other companies like Mobileye have already done.
The end result is that Tesla is way behind the competition in terms of FSD. And Tesla may eventually still get there too but they took the unnecessarily harder path and took longer than they needed to.
And I don't think it would have required that much more. Go with HD maps, 12 high res cameras, add 4 radars, one in each corner for added redundancy for cross traffic and blind spot detection, and 2 EyeQ4 and based on this video, you would probably have something good enough for FSD.
The 3d ui display is not what the car drives or sees. It also has a very low refresh rate and its latency is very high. Plus it only displays some info. The actual bounding boxes view dont have these problem.
Although if another car completely obstructs another car then it wont show up.
Thanks for explaining it to me. Like I said, it is a very good demo, way better than what Tesla currently has. Thanks for sharing. I guess we can feel pretty confident about Lucid's FSD since they are going with Mobileye, right?
It is interesting that Mobileye did the demo with 12 high res cameras. It shows that a camera driven approach to FSD can work but you still need better hardware than what Tesla went with.
Do you think the demo would stop working if it had 8 cameras instead of 12? Why not 16 cameras?
Honestly I do wish Tesla had more, there's so much overlap in the front-facing views, like 5 different cameras all forward facing, and to the sides/rear you've effectively got... one camera covering the same field of view? I guess as long as they can do lane changes confidently they should be okay with the 8 cams.
You need enough cameras for redundant 360 degree coverage. Above that, it is overkill. Less than that, you will have dangerous blind spots. 16 cameras is too many. I don't know how the cameras are positioned. Depending on the position of the cameras, 8 cameras might be enough. I suspect that 8-12 is the right number. I would have to look at a diagram to know for sure. But you do need redundant 360 degree coverage.
Meh, it doesn't even have trashcan recognition.
Seriously, the computer vision is pretty impressive, especially considering that they don't even have radar to help with distance and speed measurements in front of the vehicle.
Very cool. How reliant is it on HD maps? Works for Waymo and MobilEye. Are they testing outside of Jerusalem?
The demo actually uses 8 cameras for the driving. The other 4 are parking cameras. The demo was done using 1 EyeQ5 chip but the car has 2 EyeQ5 chips.
Here is the presentation from CES 2020 where Amnon goes into great detail on how Mobileye does autonomous driving:
I noticed he uses the term "VIDAR", is this a typo or something new? Maybe it's just depth from vision?
Also I noticed they have a special stitching vision system for close-ups, partial captures of adjacent vehicles. Can Tesla ship that please? Adjacent or long vehicles are a weakness in the current firmware.
It stands for Visual LIDAR. Here is how Amnon describes it in the presentation:
"It's a Deep network that takes the multiple cameras that we have in the car and combines them altogether to do triangulation and get dense 3D percept".
Basically, it's a Deep Neural Net that combines images from all the cameras and creates an accurate 3D map. He says it is like LIDAR but much higher resolution.
Cool, ship it!
They will. The upcoming Lucid Air going into production later this year will use it.
Amnon goes into details about their vision software from 17 Mins to 38 mins.
This is a Must Watch!
Yeah, it is a great presentation. Mobileye has great stuff. I am excited that Mobileye looks to have L3/L4/L5 in consumer cars in just 5 years.
It will be production ready 2nd half of 2020 and it will ship on the Lucid Air, BMW iNext and currently an unnamed future VW model.
Yup but the most important part is who is doing the driving policy.
I feel bumper front camera is really a necessity in parking situation so you can see that nothing is still since the last time the car parked (toddlers, small objects, homeless people sleeping, etc).
Also this is why a separate redundancy system is needed. Tesla's rear view camera suffers during moderate rain for example, same with their rear facing cameras.
By the way the 4x parking cameras are fed into VIDAR so they definitely use it even outside of parking scenarios.