If you like MBLY use todays gift from "Trippin" Trip Chowdhry to add,
Tesla Autopilot Radar Focus Spells Mobileye Bad News|Trip Chowdhry
It seems to be very hard to find real, concrete information on the state of the development towards autonomous driving. Lots of hoopla. Lots of disagreement in the research literature and very little knowledge being shared about the systems themselves.
We should be at a point where we have some reasonable view of the various approaches that people had taken and whether or not they will lead to a development arc that leads to some sort of autonomous driving. But, I guess which should be expected, we have more divergence that not.
One of the biggest issues is whether or not one can just to Level 4 directly. Or what these levels really even mean:
A Critique of NHTSA and SAE "Levels" of self-driving
For example, it is possible for Google to deliver a SAE and NHTSA Level 4 bus that is not SAE Level 5 and does fully unmanned operation, but only along pre-selected bus routes. It would be Level 4, but not the Level 4 everyone is probably thinking what Level 4 means.
Geohot is working on Level 3, and he dismisses Level 4... as something that he knows nothing about, primarily because what does Level 4 even really mean. Certainly from TMC Connect 16, there is apparently significant divergence between those that believe in going straight to Level 4 and whatever that means and those that believe in releasing the intermediate steps into general operation.
The issue is, we might be at some sort of Level 3 for a very, very long time. And the only way to prove that a vehicle is actually Level 4 is to drive it around a *lot* at Level 3. And a lot means a whole crap ton of a lot here. And each implementation might need to achieve that level of driving around to be considered Level 4.
In any case, clearly Tesla's approach is not radar centric... the latest code allows for the radar to be the primary sensor in some cases. Before, it could not be... it was supplemental to the camera. And the camera currently used in Tesla's AutoPilot is not a full color camera. I think Tesla's approach of increment improvements in the field is going to allow them to achieve more than other approaches.
To bring that back to the short term thread, I would guess that the new AP sensor suite is still a bit away. I would guess it ships in the S/X before the 3 and it might make sense that there would be two levels of AP sensor suites. One that is the current level of hardware, and another with 4 radars, potentially 9 or cameras for a lot more money. So it makes sense to maximize the AP abilities within the current hardware setup. In the Mercedes DrivePilot system, for example, all that extra hardware doesn't really provide that much in terms of actual driving results.
I do look at the various other efforts that have LIDAR as part of the solution, that are claiming Level 4 in 2019/2020 and wonder where they are coming from... there must be something in the their roadmaps that make them think they can achieve it, but I don't see it.
As for MBLY, Geohot wasn't very charitable towards them. And I suspect that if MobilEye could get Tesla on the right development arc, Tesla wouldn't have had to drop them.