Hello, rear radar :-)

[chillaban]

Almost any radar patch array antenna looks like a black rectangle with a connector on it. That doesn't really help.

FWIW, the Tesla front antennas before the Conti switch was the Bosch MRR, not LRR.

 

[verygreen]

Agreed ... looks like the Mid-range radar MRR sensor to me

View attachment 281391

This aligns well with the fact taht 17.40 firmware got another bosch-based radar firmware added in addition to already existing Conti radar (added in 17.22) and a bosch radar firmware that was there since forever (also older version firmware there).

 

[MarcusMaximus]

So, it sounds like in the Model 3, they replaced one of the cameras(presumably backup camera) with a rear radar. The question now is: why? It could be because they don't think any rear camera will work, or it could be that they don't think the particular rear camera setup on the Model 3 will work. If it's the former, there's going to be hell to pay when it comes time to retrofit on all the previous FSD cars. If it's the latter, it's probably not a big deal.

Of course, there is always the third choice: They think a rear facing radar improves FSD, but isn't necessarily required for it. That would mean these newer cars get improved FSD, and the older ones are stuck with the older version.

 

[croman]

Of course, there is always the third choice: They think a rear facing radar improves FSD, but isn't necessarily required for it. That would mean these newer cars get improved FSD, and the older ones are stuck with the older version.

I think this is a red herring. Let me explain. Tesla would need to develop multiple levels of FSD software. Just for a legacy fleet? That makes no sense. So they have one version of FSD that assess rear by using a camera (amazing tech!) but the M3 and all new S/Xs use the radar (in addition, perhaps but doubtful). There is always a tension: which is the primary sensor. Its easy with AP2 (there is only 1) but becomes far more difficult when some (diminishing) portion of your fleet is supposed to be FSD without the help of radar support.

 

[bswn1]

3 forward, 2 pillars, 2 side markers.

Yes that's what I said in a later post.

However, it's 100% clear on the Tesla website and through statements they made that on the Model S / X, there are 8 total AP cameras (the ones you mentioned plus the rear camera).

As far as I know, it was assumed HW2.5 for S / X as well as the Model 3 used the same number of active cameras for EAP / FSD functionality.

This no longer appears to be the case.

 

[MarcusMaximus]

I think this is a red herring. Let me explain. Tesla would need to develop multiple levels of FSD software. Just for a legacy fleet? That makes no sense. So they have one version of FSD that assess rear by using a camera (amazing tech!) but the M3 and all new S/Xs use the radar (in addition, perhaps but doubtful). There is always a tension: which is the primary sensor. Its easy with AP2 (there is only 1) but becomes far more difficult when some (diminishing) portion of your fleet is supposed to be FSD without the help of radar support.

With traditional software engineering, that's true. It becomes less so with machine learning, though. No matter what, each type of car is going to need a different set of networks, due to differences in the positioning of the sensors(it's possible to make a single set of networks that can generalize between hardware configurations, but then you're wasting substantial portions of the capacity of the network on configurations any one particular car will never see). If you're already supporting different networks, then using radar vs a camera just becomes swapping out(or adding) one of the inputs and letting the network learn how to use it. Adding functionality to the existing networks is a matter of just adding additional layers or units in the latter part of the network and letting the training process determine if/how to use the radar vs. rear camera.

The most likely scenario, I think, is still the second one I gave: that some difference(maybe resolution or angle) in the rear camera made it more necessary to have rear radar on the Model 3, but the pessimistic one I gave here isn't as onerous as one might expect.

EDIT: I'll add this: I really don't think it's likely that this radar is something that's being determined to be necessary for FSD. Human drivers do fine the vast majority of the time even with the entire rear view blocked. It's not a terribly useful viewpoint, generally, because there's not usually any needed reaction to anything happening directly behind you. A simple camera should do fine here, and I've successfully used the rear camera display, myself, to stand in for a blocked rear view mirror multiple times. At absolute most, I could see the rear radar being used for some kind of rear-ending-avoidance functionality in the future.

 

[22522]

Platooning might work better with an active device there.

 

[lunitiks]

This aligns well with the fact taht 17.40 firmware got another bosch-based radar firmware added in addition to already existing Conti radar (added in 17.22) and a bosch radar firmware that was there since forever (also older version firmware there).

@verygreen when was 17.40 released...?

 

[BigD0g]

And 2.5

@verygreen when was 17.40 released...?

8/22 according to ev reports

 

[verygreen]

@verygreen when was 17.40 released...?

m... Cira Oct 12, 2017 judging but the build date. That said I do not have 17.38 sample so it might have been added a couple weeks earlier than that.

 

[JamesL6]

It seems like odd placement for rear radar. This would be analogous to the FOV of front radar, where it can read speeds of distant objects in all lanes and close-up objects right behind it.

Most cars implementing forward or rear cross traffic detection use directional radar on corners. Either front, rear, or all 4 corners like you see in the Audi Q7. The 4 corner solution also allows you to get rid of the center radar in the front and that effectively gives you redundant dual forward radar plus cross-traffic visibility and oncoming traffic discrimination.

It just seems odd to put a center radar in the back. Unless your rear camera view solution is really not gonna work, or you want to implement something like rear-ending-avoidance in the short run.

Some radars are now being made so you can position it in the center of the car and still get data from multiple lanes since they angle the sensors. I'm not saying Tesla is using this for the rear radar, but here is an example of what I am talking about. Maybe they chose this route because it was cheaper/easier to implement in the 3 instead of adding 2 more radars to the corners and gave them the same/around the same performance?

 

[lunitiks]

So let me get this straight... Software changed wrt radar back in August-October 2017?

Was this change only for Model 3?

 

[lunitiks]

Because that's about when S/X got the 2.5 hw

 

[chillaban]

Some radars are now being made so you can position it in the center of the car and still get data from multiple lanes since they angle the sensors. I'm not saying Tesla is using this for the rear radar, but here is an example of what I am talking about. Maybe they chose this route because it was cheaper/easier to implement in the 3 instead of adding 2 more radars to the corners and gave them the same/around the same performance?

I do agree you can theoretically make patch arrays like this, particularly with beamforming, but that's not something supported by the MRR antenna per se. In particular, the MRR page describes using two sensors rather than a single multi-field sensor:

The MRR rear version monitors the area next to and around the rear of the vehicle, reliably detecting vehicles in the driver’s blind spot as well as traffic approaching from behind. Two sensors, one to the left and one to the right, are concealed in the rear bumper of the vehicle. The sensors are angled outward at approximately 45 degrees and can detect objects at a distance of up to 80 meters.

This is the typical setup for blind spot monitoring corner radars.


IIRC, the MRR Bosch sensor actually does have 2 FOV's, but in the vertical axis. One shoots a "taller" beam for overpass detection, where if the taller beam has a stronger return then it implies that the return is an overpass and not a car. But, as we all know from driving the early days of AP2, it's not exactly foolproof.

Just because it's theoretically possible according to a marketing datasheet does not mean that in practice it works out as well. So far, nobody's brought to market a cross traffic detection scheme using a single center-mounted radar. Maybe Tesla will impress us by being the first.

 

[JamesL6]

I do agree you can theoretically make patch arrays like this, particularly with beamforming, but that's not something supported by the MRR antenna per se. In particular, the MRR page describes using two sensors rather than a single multi-field sensor:



This is the typical setup for blind spot monitoring corner radars.


IIRC, the MRR Bosch sensor actually does have 2 FOV's, but in the vertical axis. One shoots a "taller" beam for overpass detection, where if the taller beam has a stronger return then it implies that the return is an overpass and not a car. But, as we all know from driving the early days of AP2, it's not exactly foolproof.

Just because it's theoretically possible according to a marketing datasheet does not mean that in practice it works out as well. So far, nobody's brought to market a cross traffic detection scheme using a single center-mounted radar. Maybe Tesla will impress us by being the first.

Wow, great info there! We know how ambitious Tesla is with projects, especially AP I would sure hope that after the multiple stumbles with the platform that they would go with a solution they know would work. Lets hope they can pull that off!

 

[ValueAnalyst]

This aligns well with the fact taht 17.40 firmware got another bosch-based radar firmware added in addition to already existing Conti radar (added in 17.22) and a bosch radar firmware that was there since forever (also older version firmware there).

Quick question for ya: How many "3 months maybe; 6 months definitelies" before the FSD demo?

 

[BigD0g]

I do agree you can theoretically make patch arrays like this, particularly with beamforming, but that's not something supported by the MRR antenna per se. In particular, the MRR page describes using two sensors rather than a single multi-field sensor:



This is the typical setup for blind spot monitoring corner radars.


IIRC, the MRR Bosch sensor actually does have 2 FOV's, but in the vertical axis. One shoots a "taller" beam for overpass detection, where if the taller beam has a stronger return then it implies that the return is an overpass and not a car. But, as we all know from driving the early days of AP2, it's not exactly foolproof.

Just because it's theoretically possible according to a marketing datasheet does not mean that in practice it works out as well. So far, nobody's brought to market a cross traffic detection scheme using a single center-mounted radar. Maybe Tesla will impress us by being the first.

Or maybe they won’t....

 

[verygreen]

So let me get this straight... Software changed wrt radar back in August-October 2017?

Was this change only for Model 3?

Well. There were two changes, the HW2.5 front radar firmware was added in early June (as we were discussing the whole hw2.5 thing) and then this other Bosch radar firmware was added in October. Now it does not really say anything about positions of the radar or such, so it's also possible they did away with the conti and now are putting a different revision of the Bosch radar in (seems somewhat unlikely). Another notable change from about that time is the park system hardware got a new revision (the ultrasonics) with hw2.5.

also while we are discussing radars, did you all know that post-refresh the center front radar was moved to the side? Model S has it moved left and Model X has it moved right.

 

[BigD0g]

Quick question for ya: How many "3 months maybe; 6 months definitelies" before the FSD demo?

3 - 6 ?

 

[verygreen]

Quick question for ya: How many "3 months maybe; 6 months definitelies" before the FSD demo?

Who knows. Evidently their current attempts did not pan out to produce the results they were hoping for, so I guess they went back to the drawing board?