Tesla.com - "Transitioning to Tesla Vision"

[orion2001]

Having gone through the AP2 migration, this is really bad news for new owners. Unless @verygreen says they have been collecting data for this for a while, seems like a hail Mary to save delivery numbers for the quarter. Was waiting for tax credit and updated headlights to take delivery on my Y, but now I'll get no radar and probably a crappy AP experience until they flush out all the issues. Also not sure how they can deal with fog and heavy rain without radar.

I'm definitely passing on my current MY order if the tax credit/rebate doesn't happen. In that scenario, I see no reason to pay more (all the recent price hikes) for a car with no radar and a gimped feature-set that will be achieving supposed parity "soon"*

I might as well wait for Austin to come online, EV competition to heat up and maybe drop prices or provide more interesting options, 4680 cells to become a thing (maybe) and to see if vision-only features on the 3 and Y end up achieving parity with older, radar-equipped models.

It leaves a bad taste in my mouth that the refreshed S and X will continue to retain Radar. If this was truly about changing the paradigm and moving forward, they would have done the same with those cars as well. Right now it feels a bit like experimenting with the Tesla "peasants" and deceiving them with typical marketing speak to make up for them cutting corners and trying to get around potential part shortages all while convincing their users how it is aligned with their best interests .

 

[JHCCAZ]

they are obviously not going to do it. The proof is literally in Tesla statements themselves: if you do not have radar you are having degraded experience. Also they'll force autohighbeams and autowipers on you if you don't have radar (when on autopilot, I guess when not on autopilot they don't care if you crash as much?)

I don't think this is proof nor the "obvious" interpretation. See my (admittedly speculative bit IMO reasonable) post above, and below is an alternative explanation based on that:

• Tesla presently has long-established AP features and modes that, for better or worse, incorporated and somewhat relied on radar.
• They have just now produced a large batch of cars without the radar module, cars that were undeliverable awaiting a mysterious update.
  • I'm postulating this radar deletion, this early on, was not the preferred plan and was forced upon them by supply issues.
  • Or maybe not absolutely forced but also opportunistically embraced & mandated by Elon, whichever.
• Consequently they've had to rush the re-writing of AP code sans radar, superseding the prior plan to roll out Pure Vision for City Streets beta and, only later, retrofit it into existing AP features under less-rushed conditions.
• So this would explain why they are restricting the operational envelope of older AP features. Under the circumstances, they're balancing the need to get the on-hold inventory of cars into a usable & deliverable state, against the concerns of deploying replacement AP software with less qualification testing than anyone would normally wish to see.
• The reduced AP-feature envelope is intended to be a temporary compromise pending additional testing and possibly tweaking.
• It is not obvious proof that the vision-based rewrite is ultimately inferior and will be a "degraded experience" long-term. It may or may not be, but at this point all we can say is that it's new, mandating extra cautions in this obviously pressured deployment.

Please understand that I'm neither defending nor condemning this course of action because I don't really know the specifics. I am saying that no interpretation, including mine, is obvious or proven. We're all free to assume good vs. bad motives that may inform our customer opinions and predictions.

 

[Bladerskb]

In fact the radar is no substitute for vision in inclement conditions. Automotive radar has come a long way, but it is far from presenting what we would think of as an "image" that you would want to rely on to guide the car. As an adjunct to cameras, it can certainly identify that there's something out there roughly x feet ahead, and in advanced systems roughly y degrees off-axis. So I'd say it can reinforce where not to go, but with blinded cameras it won't really tell you where you can go.

Another way to express this: if you are fool enough to plow ahead into fog or rain where you cannot see, the radar might save you from a collision or from running someone over. But you absolutely have no business proceeding ahead under those conditions, and neither does a sensor-loaded AV.

Proper conclusion: we can debate the radar Pros e.g. safety-backup warnings and unquestionably superior presence-of-hazard detection in poor-vision conditions, and Cons e.g. false positives causing phantom responses, but it's quite wrong to think that Radar can "take over" in case of camera-vision failure.

On the other hand, cameras with sophisticated frame averaging / noise reduction, and with some IR sensitivity outside human perception, actually do have the potential to see real images better than humans in poor weather. This is an interesting and very pertinent topic, and I wish it would share the discussion bandwidth with Lidar and Radar debates.

have you seen the output of an imaging radar with thousands of channels?

 

[JHCCAZ]

have you seen the output of an imaging radar with thousands of channels?

Well, yes but how is that relevant? If you tell me that you can heat a cup of soup faster in the microwave than on a gas range, should I ask you "have you ever seen a blast furnace?"

What you raise is surely interesting technology in diverse but mostly quite unrelated applications, and to my knowledge changes not one thing I wrote in the post.

My purpose was to helpfully clarify, given a number of users' references to the car's radar "seeing" through fog and rain, that it isn't really "seeing" a safely navigable image. At all.

Now, how many channels do you think are on the Tesla radar module under discussion?
Do you know the wavelength, tange, velocity and angular resolution and how these could paint an image or, more realistically, assemble a roadscape model that could in any way be navigable?
Have you seen the output of any production automotive radar that you would recommend to use as the primary perception stack to pilot the car without cameras or Lidar, in any weather?
That can be used in any relevant scenario under discussion?
If that were a viable design choice, then why would all your favorite players be using cameras and often Lidar?

See I can ask questions too, perhaps you can answer them. However, to tone down the argument, I would be satisfied if you would re-read what I wrote, consider the content and point of the message, and engage based on that.

If it makes you feel better, I'll clarify that I'm quite enthused about radar technology past, present and future. Including on cars, where the engineers have used it to detect hazards. And I sometimes use it to heat my soup, I'm very open-minded that way. But I haven't yet met the radar that can guide my car to the store in ideal conditions, much less in a rainstorm.

 

[qdeathstar]

I think there is a greater than 50 percent chance that we look back at 2021 as the year when it all fell apart of Tesla. Making a lot of bizarre, out of the blue, consumer unfriendly decisions lately, not just with radar, but with thier solar projects and the whole bitcoin debacle....

 

[Microterf]

I think there is a greater than 50 percent chance that we look back at 2021 as the year when it all fell apart of Tesla. Making a lot of bizarre, out of the blue, consumer unfriendly decisions lately, not just with radar, but with thier solar projects and the whole bitcoin debacle....

I hope not, but I could see that happening

 

[Darblish]

I'm definitely passing on my current MY order if the tax credit/rebate doesn't happen. In that scenario, I see no reason to pay more (all the recent price hikes) for a car with no radar and a gimped feature-set that will be achieving supposed parity "soon"*

I am tempted to agree. (I’ll brace myself for the inevitable thrashing we’ll get for daring to doubt The Company™ and their Plans™, and informing us that we’re not the right kind of Tesla owners.) I really liked the Model Y, much more than any other cars I looked at when I was choosing. I liked it enough that when the store said “you probably won’t get one until September,” I thought that was worth it.

I’m not getting FSD. It puts the car well out of my budget, and potentially I could look into EAP or the FSD subscription later if there was something that really made me think it would be worth that kind of money. Right now there isn’t, and that’s ok. Regular old autopilot matches pretty closely the ProPilot system on my wife’s Nissan, but I think it did a better job of keeping centered in the lane during my test drives than the Nissan does. The only thing it doesn’t have (in comparison) is an automatic re-engage of the steering after a lane change, which was frustrating mainly because I wasn’t used to it yet.

If I had test driven a Model Y and found that what autopilot features it does have won’t fully engage at 76mph, and the lane departure safety feature is disabled indefinitely, I would not have evaluated it so positively in comparison. I’m especially disappointed because it was all there when I paid Tesla their $100, and now it’s been removed. Maybe the new system will be working and reach feature parity in a couple of weeks; we’ll have to see what actually happens with the update, and reports from those who are taking delivery this week. If it’s business as usual by the time my car is built — and that could still be 1-3 months for me, so lots of time — then I’m probably going to feel OK about it. If it’s still limited, I don’t really have the kind of money to throw this much cash into a vehicle that is missing features found in nearly every competitor, especially when I could go buy an ICE competitor (without the same limitations) and drive away the same afternoon.

I’m not ready to cancel right now, but it’s certainly on the table in a way that it wasn’t yesterday morning.

 

[qdeathstar]

Not

I am tempted to agree. (I’ll brace myself for the inevitable thrashing we’ll get for daring to doubt The Company™ and their Plans™, and informing us that we’re not the right kind of Tesla owners.) I really liked the Model Y, much more than any other cars I looked at when I was choosing. I liked it enough that when the store said “you probably won’t get one until September,” I thought that was worth it.

I’m not getting FSD. It puts the car well out of my budget, and potentially I could look into EAP or the FSD subscription later if there was something that really made me think it would be worth that kind of money. Right now there isn’t, and that’s ok. Regular old autopilot matches pretty closely the ProPilot system on my wife’s Nissan, but I think it did a better job of keeping centered in the lane during my test drives than the Nissan does. The only thing it doesn’t have (in comparison) is an automatic re-engage of the steering after a lane change, which was frustrating mainly because I wasn’t used to it yet.

If I had test driven a Model Y and found that what autopilot features it does have won’t fully engage at 76mph, and the lane departure safety feature is disabled indefinitely, I would not have evaluated it so positively in comparison. I’m especially disappointed because it was all there when I paid Tesla their $100, and now it’s been removed. Maybe the new system will be working and reach feature parity in a couple of weeks; we’ll have to see what actually happens with the update, and reports from those who are taking delivery this week. If it’s business as usual by the time my car is built — and that could still be 1-3 months for me, so lots of time — then I’m probably going to feel OK about it. If it’s still limited, I don’t really have the kind of money to throw this much cash into a vehicle that is missing features found in nearly every competitor, especially when I could go buy an ICE competitor (without the same limitations) and drive away the same afternoon.

I’m not ready to cancel right now, but it’s certainly on the table in a way that it wasn’t yesterday morning.

Well, let’s not overstate things. Nearly every other manufacturer does not have smart summon or autosteer above 75mph... the lane departure warning thing is weird though, I would think tha would be vision only already.

 

[Darblish]

Not


Well, let’s not overstate things. Nearly every other manufacturer does not have smart summon or autosteer above 75mph... the lane departure warning thing is weird though, I would think tha would be vision only already.

I didn’t mention smart summon, and in fact I did say I was not getting FSD. It’s certainly disappointing to some, but that isn’t weighing in my decision.

I can get into my wife’s 18-month-old mid-level trim Rogue and turn on its autosteer equivalent at 76+mph with no problem (I think the fastest I’ve set it to was 80mph to keep pace with traffic). For that car, we paid almost exactly half the cost of the Model Y configuration I ordered. Every other car (ICE and EV) that I was considering had their own equivalent of autosteer (“ProPilot,” “LKA,” “Copilot,” etc.), and none of them have the 75mph limitation. They also all had lane departure warning/assist systems.

Again, if this is all over and done within a couple of weeks then I will happily swallow the doubt that I have today and move on, and I hope to enjoy the Model Y as much as I did on the overnight test drive. Tesla does not have a track record of promises following the same timeline as reality, though. That’s an OK tradeoff when we’re talking about a redesigned interface or something else shiny and new, but not when we’re looking at “basic” functionality for the market segment.

 

[Simplepotamus]

This is a dangerous change and completely disingenuous messaging by Elon and Tesla. Here is why:

FOG: Growing up on the Oregon coast, I have driven in my share of thick fog. On occasion it is so thick that you cannot see more than 3 car lengths ahead of you. The cameras will not be able to see any better. But radar can detect cars far down the road, even in fog.

SNOW: I now live in the mountains of Utah where we routinely get whiteouts during winter storms. See comments on fog above. Snow also can stick to the front of the car or windshield, blocking camera view.

NIGHT: News flash: Night is dark. When it is dark, it is hard to see. However, I recognize that cameras these days can have a wider range of light sensitivity than the human eye. My dashcam, for example, shows things much better at night than I can see. Nonetheless, radar again provides much earlier recognition of slow or stalled vehicles, moose (which we have here and which are surprisingly dark in color), black panthers, etc.

NIGHT + SNOW + FOG: Been there. I'll plan on driving the Volvo on those days.

My prediction is that the first comprehensive test comparing autopilot-type systems across vehicles will rank Tesla Vision dead last unless driving during daytime with a clear view down the road. Accidents and associated lawsuits will force Tesla to acknowledge that the system is dangerous, and they'll push out Tesla Magic Vision that brings back radar.

To me, it would make sense to just acknowledge that camera-based is limited and to sell that as "basic autopilot" as the default. If someone wants to have a step up, add radar and call that the new Enhanced Autopilot. Then if someone wants true FSD, add more cameras, additional radar, lidar, etc to get as much input for decision making as possible and charge even more for that. Right now, the FSD is solely a software upgrade. It would be easier for people to stomach paying $10k extra if they knew that it meant that they were getting more sensors in the car. That would of course mean that the decision would have to be made upon order, not as an after-purchase add-on. But that would be okay.

For the record, I'm waiting on a Model 3 LR that I ordered mid-April and I am not enthused about this change to Autopilot.

 

[mikes_fsd]

In a consumer world, a capitalist system, when I fully paid for my dinner party, I should not have to wait for the restaurant owner to get regulatory approval

Except, in the case of Full Self-Driving Capability - you did not "pay for a dinner party" but rather you paid for the "process of how to plan/setup and host a massive rave" -- you would get to be at the practices/rehearsals with the final delivery to actually participate in the rave.... and you were warned that "regulators will be involved in all permitting needs"

That is what you bought! Refusal to understand that does not change that fact.

2016 - Autopilot
2021 - Autopilot

 

[mikes_fsd]

Right from a Kaparthy video.
It may all matter what they classify as an "automated lane change".

Same exact slide was shown by the same person in a presentation in February 2020

First, hard to believe from in 4-5 months that number stayed the same
Secondly, they added a million cars on the road since then!

One more thing, they introduced automated lane changes (not just suggestions) in April 2019 - Introducing a More Seamless Navigate on Autopilot

So, with the smaller fleet before the 2/2020 presentation by Karpathy, they had the number of 200k automated lane changes.

 

[mikes_fsd]

Unsaid is the fact that those with HW3 and paid FSD (our case, $8K and $6K in 2020 and 2018, respectively) will NOT get "Tesla Vision." With the abandonment of radar-equipped FSD betas (the 8.x city FSD), we are pushed to the back of the line.

I do not think they are saying that at all:

 

[texas_star_TM3]

there is clearly a supply chain issue with the radar sensor or massive cost cutting pressure to stop adding them to vehicles asap... because not including them in 3 / Y *before* "Tesla Vision" is as capable (or ideally: better) than current AP is crazy. As an example: limiting autosteer to 75mph will be fun for long road trips. Emergency lane departure avoidance not available is a safety feature most other OEMs have... and what "weeks ahead" means in software releases is up to anyones guess. "weeks" = Elon time or normal time?

 

[Stealth.Pilot]

75mph limit on autopilot is very poor. Mercedes has a 120mph limit on their equivalent system.

what-is-mercedes-benz-distronic-plus-with-steering-assist

 

[ItsNotAboutTheMoney]

there is clearly a supply chain issue with the radar sensor or massive cost cutting pressure to stop adding them to vehicles asap... because not including them in 3 / Y *before* "Tesla Vision" is as capable (or ideally: better) than current AP is crazy. As an example: limiting autosteer to 75mph will be fun for long road trips. Emergency lane departure avoidance not available is a safety feature most other OEMs have... and what "weeks ahead" means in software releases is up to anyones guess. "weeks" = Elon time or normal time?

Or just timing.
They have to place orders in advance, want to eliminate the radar cost and it depends on the crawl of Autopilot progress.
Guess wrong and you have a limbo period.

We often see how bad guesses affect production rollouts.

 

[Eno Deb]

Six months ago they wanted to use "millimeter-wave radar" ...

Tesla files to use new 'millimeter-wave radar' in electric cars

Tesla has applied with the FCC for the use of a new “millimeter-wave radar” inside the automaker’s electric cars. New...

electrek.co

 

[diplomat33]

Six months ago they wanted to use "millimeter-wave radar" ...

Tesla files to use new 'millimeter-wave radar' in electric cars

Tesla has applied with the FCC for the use of a new “millimeter-wave radar” inside the automaker’s electric cars. New...

electrek.co

I think that was for a driver monitoring system. But I could be wrong.

I know back in 2016, Elon was gushing about radar using point clouds that would be better than lidar:

Tesla Again Spurns Lidar, Betting Instead On Radar

Musk says radar alone can produce a lidar-like “point cloud” to help Autopilot navigate

spectrum.ieee.org

Obviously, he's dropped that idea now.

 

[MP3Mike]

Six months ago they wanted to use "millimeter-wave radar" ...

Tesla files to use new 'millimeter-wave radar' in electric cars

Tesla has applied with the FCC for the use of a new “millimeter-wave radar” inside the automaker’s electric cars. New...

electrek.co

That is unrelated. That was for radar inside the cabin.

 

[DarkandStormy]

Subaru has been offering a vision/camera-based ADAS for a long time now. It was announced in 2008 and the current version looks to have started in 2013. From what I can tell, it doesn't offer Summon, AutoSteer, or City Steer (whatever jargon the supposed FSD is called in non-highway settings). Costs $845 on the sedans and a little more than $1,300 on the SUVs.