> Yah, all the ADAS test operators they've hired in the Phoenix area.

But they test the Autopilot, not self driving.

> The video is irrelevant: you could do a similar demo with Tesla EAP.

I highly doubt it. Are you really comparing that contrived/edited/cut marketing video to the demo I posted? One is a loop around the MobilEye building going through all different types of roads and traffic patterns, the other doesn't show you the same frame for more than 3 seconds. Imo, there is absolutely no way you could do this with EAP. EAP is merely adaptive cruise control with lane change and assist. That's literally it. Not sure if you saw the complete demo, but current MobilEye SAE level 3 systems respond to traffic lights, street signs, speed signs, emergency vehicles, localize based on visual reference points, etc etc. The most important part being RSS, it's used to assertively and safely navigate traffic with a fully parameterized driving policy that's adjusted dynamically based on many factors. Honestly, imo there is no comparison at all. We will see when Tesla goes to market with their FSD.

Also, I never said that LIDAR isn't important. You seemed to imply that Tesla is only behind because they are not relying on LIDAR. The public demo video I posted doesn't use LIDAR and it is operating in SAE level 3, which is beyond anything Tesla currently has. Just trying to make myself clear in saying that the reason Tesla is behind is not because they are taking some radical fundamentally different approach to this problem, like you seemed to imply.

I know there is a quote from Elon Musk with him saying that LIDAR is a "local minima" and a crutch. I want to make it clear that MobilEye's vision systems are more than up for the task, but MobilEye has a culture of safety. That's why AFAIK all SAE 3+ systems will have LIDAR as standard.

> There's the whole semis-with-few-features-blending-into-the-background problem that Tesla is maligned for. It's a rare case, but a key one where you don't want to rely on purely vision This is the kind of problem vision-only systems face: if you don't have enough features for coincidence finding, and you misclassify a truck-- pretty soon you'll have a problem.

Ah yes, the classic running into stationary fire trucks. This is one of the reasons why I have no confidence in Tesla/EAP/FSD. It's gotta be the way their software is written. Machine vision systems do make mistakes