Radarless Driving Report? Anyone?

[Zhe Wiz]

Makes me sad to hear that there could be possible disadvantages over the cars with radar, but still very much looking forward to my delivery in Sept!

If you’re watching some of the V9 beta FSD videos, it doesn’t look like TeslaVision will be much of a disadvantage. Sure there are a lot of logic issues to resolve, but the cameras are working quite well.

I worry about the car’s ability to “see” to the left and right at intersections, but radar wouldn’t help with that. I’m feeling pretty good about my Vision only Y after watching those videos.

 

[linux-works]

if 'vision' had PROPER REDUNDANCY I would not be as concerned about removing one of the 'senses' from this kind of car.

fact is, they cheaped out and did not provide heating elements or wipers or extra cameras per vision-site other than the front and that was not real redundancy, its just multi fixed focal lengths.

no, before you remove a valid working sensor, I'd want to see something else added - and no, software does not count as a 'soft pseudo sensor'.

we've seen enough 'camera blinded by sun, ap unavailable' haven't we? what makes all you vision believers think that we can really do anything more than level2 with the (now even more reduced) sensor array on teslas?

technically, what tesla says is pure BS. the simple fact that their radar is 'hard to fuse properly' probably just means they need new designers and architects.

I dont know how good the people are that tesla currently has, but throwing up your hands and saying 'its too hard' is a sucker's move. I dont respect them when they pull crap like that and try to make it seem that its not a turd sandwitch.

look, if you run short of parts say so. we're big boys and girls. we can take the truth. and when parts are back in stock, you could install them on cars that went without.

this kind of left-handed justification does not fly with me and does not fly with anyone who has done any software work. look, you cant figure out how to do this? fine - the door is over there. tons of people could take over and probably MAKE it work. I have zero doubt. but we wont find out since they basically threw in the towel.

 

[Dmirkin]

It is working much better now. I can use it up to 80mph on a clear day and it has worked fine in light rain. I’ve had not one phantom brake incident. It does warn me a lot, even when not engaged, that I am going to crash into the car in front of me when I feel I am no where close to that. It also warns me to take the wheel a lot-when nothing is engaged. I have a light touch on the wheel and maybe it thinks I am sleeping or inattentive. Is there a way to turn these warnings lower? Y with 1250 miles.

 

[JerseyShoreMY]

It is working much better now. I can use it up to 80mph on a clear day and it has worked fine in light rain. I’ve had not one phantom brake incident. It does warn me a lot, even when not engaged, that I am going to crash into the car in front of me when I feel I am no where close to that. It also warns me to take the wheel a lot-when nothing is engaged. I have a light touch on the wheel and maybe it thinks I am sleeping or inattentive. Is there a way to turn these warnings lower? Y with 1250 miles.

Much of our Tesla Vision experience has also been just fine on AP with no radar. No troubles here with rain either. Phantom braking more of an issue with some winding roads and shadows than on the highways.
Using the latest 18.10 OTA update for our MY, the vision length and action / reaction times seem more safe and sane now than before.
Recall earlier versions running you at full speed until the last moment up to the car at a stop in front of you. That is resolved.
Running at 80 MPH and no troubles on the high speed Parkways here, graphics seem better now. See this pic.

 

[Zhe Wiz]

if 'vision' had PROPER REDUNDANCY I would not be as concerned about removing one of the 'senses' from this kind of car.

fact is, they cheaped out and did not provide heating elements or wipers or extra cameras per vision-site other than the front and that was not real redundancy, its just multi fixed focal lengths.

no, before you remove a valid working sensor, I'd want to see something else added - and no, software does not count as a 'soft pseudo sensor'.

we've seen enough 'camera blinded by sun, ap unavailable' haven't we? what makes all you vision believers think that we can really do anything more than level2 with the (now even more reduced) sensor array on teslas?

technically, what tesla says is pure BS. the simple fact that their radar is 'hard to fuse properly' probably just means they need new designers and architects.

I dont know how good the people are that tesla currently has, but throwing up your hands and saying 'its too hard' is a sucker's move. I dont respect them when they pull crap like that and try to make it seem that its not a turd sandwitch.

look, if you run short of parts say so. we're big boys and girls. we can take the truth. and when parts are back in stock, you could install them on cars that went without.

this kind of left-handed justification does not fly with me and does not fly with anyone who has done any software work. look, you cant figure out how to do this? fine - the door is over there. tons of people could take over and probably MAKE it work. I have zero doubt. but we wont find out since they basically threw in the towel.

You're guessing and opining about a lot of things here.

You're upset they removed radar and think the car will be worse because of it. OK.

You think they shouldn't remove a sensor without adding something. OK.

You think Tesla programmers are crap. OK

Thing is, you don't KNOW any of that.

Having said that, I don't disagree with you entirely. Clearly radar was removed because they couldn't get the units any more. Whether or not they were going to do that eventually anyway is a guess for all but a handful of people. I don't like that the most forward cameras are too far back. I don't like that the side cameras seem to fog up, get blocked sometimes. (I've not experienced that yet, but I only have 2K miles on the car) But how does radar help any of that? It doesn't help the side cameras AT ALL, and although it was further forward in the car, it can't help "see" far enough to the left and right to help with blind intersections.

Radar may help in rain/fog/snow. But you really want to use it under those conditions? If we're ever going to get to L5 we'd have to, but Tesla is no where near that now, and using AP in those conditions would be considered dangerous. With or w/o radar. I think a lot of things you're upset about you'd be upset about even with radar. Cameras will still get blinded w/o any help (side cameras). Programmers will still suck.

BUT, I'm not freaking out like you seem to be. I'm more "agnostic". I'll wait and see what happens. And frankly, the V9 beta videos are VERY encouraging. Do I think they prove that FSD is right around the corner, or even POSSIBLE with vision only? No, it does neither. But what it DOES show is that the missing radar isn't missed. The issues remaining with the V9 beta don't seem to be radar related. It seems to judge speed and location of environmental objects very well, and in cases where corners are involved, it seems to judge those things even better than when radar was used. Yeah, that could be programming, but does it matter? If vision sees everything it needs to see w/o radar, what's the big deal? Again, not saying that's true, just saying let's see. The initial feedback appears promising, but not definitive.

Mostly you just seem upset that they removed radar, but if they added radar back, what would really be better? Of your complaints, I'm not sure it solves any. Then again, my "goal" is to have good autopilot. FSD is too expensive, and I can't see it being L5 reality for decades. It's oversold, too many edge cases, and right now can't even handle normal situations like left turns. It's far away, and radar in the short term won't help it. L2 is good enough for me. That, ultimately, may be why I'm not as upset as you are. We probably have different end goals.

Don't be mad, at least not at me.

Jack

 

[linux-works]

in a nutshell, my complaint is that they gave up on something many of us think was an extra degree of freedom.

lets say they even get 'vision-only' working as well as the 'radar+vision' cars currently did, in non-ideal conditions. meaning, no need for bright light flashing or any other side-effects. suppose that you could not tell the difference, or that automation during test and simulation can't tell the difference between what we had with fusion and what we now are going to have, with vision-only. suppose they get to where we are now.

does that prove that the extra dimension, that of radar, was not needed? of course not. it means that they managed to get back to where they were before (and that remains to be seen). even if they do, they'd have to exceed it and maintain it during a long field test period (real people in real cars).

in fact, there's nothing you can do to say that if you had A+B and you removed B, that the new A will always be better than the previous sum. they are arguing that the B sometimes is negative and the overall result is lower because of it, and I'll buy that, perhaps; but that simply means that your choice of implementation of B may need to be re-examined. it does not mean that the whole idea of adding extra sensor fusion is bad. I still think its more useful to have extra sensor inputs.

the usefulness of the number of sensors does not keep going up. 10 sensors are better than 1 but are 100 better than 10? there's a limit to how much info you can handle and actually use. but to remove ONE of the important ones - that kind of goes a bit too far. we are not 'rich' with sensor data; if anything, we are starved for it. removing data now seems the opposite direction from how they should go.

but you are right about this - we dont have access to their code or their current devel status. but as a working software person in the field, it smells like a cop-out to me. I get that the problem is 'hard'. totally get that; but reducing data input seems like side-stepping the real problem. I just dont like 'solutions' like that.

 

[mark95476]

Yup. FSD beta v9 is vision-only as well. It's only going to get better.

Another nail in the coffin for doubters...... Tesla Vision-powered active safety features found to be at least as good as with radar in new test - Electrek

 

[linux-works]

I looked at the electrek article. maybe I missed the details or maybe they didn't list them. what kind of lighting changes did they test under? what kind of weather? splashes on cameras, like in real life? did they keep all the cam lenses clean, as a finger on the scale, or where they real-world fair?

if I missed the details, my bad. but all I see is the summary and I'm not buying it until I see the test conditions.

 

[iustin]

fact is, they cheaped out and did not provide heating elements or wipers or extra cameras per vision-site other than the front and that was not real redundancy, its just multi fixed focal lengths.

This makes sense, indeed. However:

technically, what tesla says is pure BS. the simple fact that their radar is 'hard to fuse properly' probably just means they need new designers and architects.

No, lol. Companies that hire "(software) designers and architects" are the ones that have the bad hierarchical structure that left them in the 2000's, and we are in the 2020s. I'd argue that even in 2000's, the "certified software architects" were a bad hire.

I dont know how good the people are that tesla currently has, but throwing up your hands and saying 'its too hard' is a sucker's move. I dont respect them when they pull crap like that and try to make it seem that its not a turd sandwitch.

By this measure, we shouldn't abandon any technlogy ever, since it's always possible to work around its deficiencies. Segmented memory? Why not, it was cheaper than replacing it with an architecture with an MMU, and everybody was using it. Or take whatever other example - some things are worth keeping, some not. It's absurd to say that deciding doing X is not worth means you don't know how to do X. Sometime, people with knowledge decide it's not *worth* working around the deficiencies of X. Everybody was using at one point expensive hardware (choose: mainframes, expensive routers for small office, hardware radios, whatever), only to transition in time to more and more "software-defined" solutions. I see radar to vision exactly the same - replace specific hardware with a more software-defined solution that interprets better the camera output.

Remember that Subaru is using vision-only for their safety features for about ten years now. Tesla is not the only company doing vision-only.

 

[PackMan730]

Either way, this car is going to be a huge upgrade from my 2007 Lexus GS 450H, which does have auto cruise control, but shuts off when going slower than 35 MPG automatically. Can't wait!

 

[WADan]

This makes sense, indeed. However:



No, lol. Companies that hire "(software) designers and architects" are the ones that have the bad hierarchical structure that left them in the 2000's, and we are in the 2020s. I'd argue that even in 2000's, the "certified software architects" were a bad hire.



By this measure, we shouldn't abandon any technlogy ever, since it's always possible to work around its deficiencies. Segmented memory? Why not, it was cheaper than replacing it with an architecture with an MMU, and everybody was using it. Or take whatever other example - some things are worth keeping, some not. It's absurd to say that deciding doing X is not worth means you don't know how to do X. Sometime, people with knowledge decide it's not *worth* working around the deficiencies of X. Everybody was using at one point expensive hardware (choose: mainframes, expensive routers for small office, hardware radios, whatever), only to transition in time to more and more "software-defined" solutions. I see radar to vision exactly the same - replace specific hardware with a more software-defined solution that interprets better the camera output.

Remember that Subaru is using vision-only for their safety features for about ten years now. Tesla is not the only company doing vision-only.

Also people kept on speculating that Tesla removed radar due to shortage (which happened only recently) when they have been saying “vision only” for quite a few years now.

 

[Zhe Wiz]

Also people kept on speculating that Tesla removed radar due to shortage (which happened only recently) when they have been saying “vision only” for quite a few years now.

Yes. As far as I can tell the lack of parts accelerated their change to Vision only, but didn’t cause it. It appears they had planned to do it anyway, just not quite so soon.

 

[linux-works]

the cost-cutting really bugs me.

the fact that vision sensors are not redundant says a lot about elon's approach toward car safety vs cost/proft/money issues.

I understand that companies have to make a profit. but I'm not willing to cut on things I consider safety related. I'd add more sensors, not take existing ones away.

we'll never get to level 5 in this current array. no way, guys. dream all you want, but elon is cutting cost. not a good sign at all.

 

[Occhis]

Has Tesla discontinued radar on S and X also?

 

[Zhe Wiz]

Has Tesla discontinued radar on S and X also?

Not yet!