Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

Tesla Vision vs Parking Sensors

This site may earn commission on affiliate links.
And yet you watched a video of tesla vision failing by driving into a person and hose reel, and ignore that it will not pickup something low that changes after you park.

Failed? interesting interpretation. Certainly wasn't the interpretation of the person who was doing the very same testing you are now usign to try to make your argument.

The video you are using as your 'evidence' certainly didn't make any such claim. In fact he pointed out a number of instances such as the chain link fence that USS doesn't normally detect, curbs, objects at height, etc, etc.

"Driving into a person" if you need the parking assist to tell you to stop when it is clearly showing a person in front of you (ignoring the fact they are in front of your face) then there is a much more serious discussion to be had. As you also saw in the video, approaching the same person with any speed resulted in an emergency brake in every instance so a completely false claim.

The USS car also struggled with the hose reel until he lined it up directly with the sensor... (which would be the same with a tow bar, pole or any other small object) Tesla vision at least showed it until it got very close meaning any semi capable driver, paying attention would know it was there well ahead of time. Any non-biased observer would have to give Tesla Vision the win on that one also.

Not much point in discussing when you've clearly come in with your mind made up well before you've even seen the system first hand.

Without adding extra cameras, 360 view systems are at their ceiling, likewise USS have several limitations. Tesla Vision is already as good or better by any fair minded comparison in it's first iteration, there is plenty of headroom for improvements (something your video also pointed out). Your video only strengthens my argument in every way, all systems have limitations but already Tesla Vision combines some of the best of both USS and 360 view in one system and is only going to get better.

The statement he made in the very beginning of the video is certainly apt in this case, nothing should replace eyes, mirrors and common sense. Seriously if someone is claiming that you need a parking assistance system to warn them about something right in front of them and they can't observe their surroundings before they get into the vehicle, then I would have to seriously question their fitness to hold a license.
 
Last edited:
Failed? interesting interpretation. Certainly wasn't the interpretation of the person who was doing the very same testing you are now usign to try to make your argument.

The video you are using as your 'evidence' certainly didn't make any such claim. In fact he pointed out a number of instances such as the chain link fence that USS doesn't normally detect, curbs, objects at height, etc, etc.

"Driving into a person" if you need the parking assist to tell you to stop when it is clearly showing a person in front of you (ignoring the fact they are in front of your face) then there is a much more serious discussion to be had. As you also saw in the video, approaching the same person with any speed resulted in an emergency brake in every instance so a completely false claim.

The USS car also struggled with the hose reel until he lined it up directly with the sensor... (which would be the same with a tow bar, pole or any other small object) Tesla vision at least showed it until it got very close meaning any semi capable driver, paying attention would know it was there well ahead of time. Any non-biased observer would have to give Tesla Vision the win on that one also.

Not much point in discussing when you've clearly come in with your mind made up well before you've even seen the system first hand.

Without adding extra cameras, 360 view systems are at their ceiling, likewise USS have several limitations. Tesla Vision is already as good or better by any fair minded comparison in it's first iteration, there is plenty of headroom for improvements (something your video also pointed out). Your video only strengthens my argument in every way, all systems have limitations but already Tesla Vision combines some of the best of both USS and 360 view in one system and is only going to get better.

The statement he made in the very beginning of the video is certainly apt in this case, nothing should replace eyes, mirrors and common sense. Seriously if someone is claiming that you need a parking assistance system to warn them about something right in front of them and they can't observe their surroundings before they get into the vehicle, then I would have to seriously question their fitness to hold a license.
Tesla shot themselves in the foot by releasing the circa-June version of TV park assist which was little more than a tech demo and wasn't actually useful. There's a lot of people carrying a grudge about vision based park assist now.

Last night I used it to parallel park up against a hedge in a very crowded scout car park full of moving cars and children in the dark. It was very useful. This morning I used it to park in a multistorey car park space with a concrete pillar in one corner and walkway 'hazard' markings marking the other side. The TV system rendered both perfectly.

The TV system is not USS and it's not trying to be, and thankfully neither is it trying to be one of those awful top down camera views. It needs to improve with its perception of low height objects to the front, but it's a genuinely good system and I think we're going to see a useful self park (perhaps not summon though, due to the issues with the blindspots at the present time) some time soon.

From the technologist perspective this is exciting because this is the first generation of a new technology with plenty of scope to improve, unlike USS and simple camera views which are at the end of their evolution and are not going to facilitate any advanced autonomous behaviour.
 
I've used 360 view systems in a number of different medium to high end vehicles for many years.

Having first hand experience with both systems I 100% stand by the statement that 360 view cannot handle this situation as well as Tesla Vision did (unless you want to stay hung up on the fact that the cars are perfectly represented as grey shapes rather than "pretty" (warped) pictures).

A 4 camera 360 view cannot handle when the cars are as close as they were in this example, the side cameras simply cannot see enough to warp the image correctly and you get the exact situation in the other video I'd posted on page 1 and perfectly demonstrates the limitations of a 360 view that is not present in Tesla Vision. You've got video evidence of a difficult situation where Tesla Vision worked perfectly and video evidence of a less difficult situation where 360 view failed miserably.

This is a dubious example because we only see what happened, not how it happened. Given how severely the guy cut the wheel my best guess is the cameras were showing everything just fine and he ignored it, and is now feigning surprise that the trailer is misrepresented in the image--well of course it is now, because there is no camera immediately next to where he hit it.

Considering my daughter did literally the exact same thing in my model 3 several months ago (Bumped garage door frame, $5k repair) I certainly have no hope in tesla's ability to avoid mishaps like this. I can't say she wouldn't have hit it in a car with a proper 360 system, but I can say she hit it in our tesla.
 
Considering my daughter did literally the exact same thing in my model 3 several months ago (Bumped garage door frame, $5k repair) I certainly have no hope in tesla's ability to avoid mishaps like this. I can't say she wouldn't have hit it in a car with a proper 360 system, but I can say she hit it in our tesla.
But this is the whole point, isn’t it? Current parking systems rely on point measurements from the USS sensors which become useless if the obstruction is sufficiently close to pass between the sensors.

A system that operates on a full 3D model of the world around the car has the potential to avoid incidents like this.
 
  • Like
Reactions: Jules22
But this is the whole point, isn’t it? Current parking systems rely on point measurements from the USS sensors which become useless if the obstruction is sufficiently close to pass between the sensors.

A system that operates on a full 3D model of the world around the car has the potential to avoid incidents like this.
"potential" sums up a lot of Tesla's approach to things :)
 
  • Like
Reactions: paulp
But this is the whole point, isn’t it? Current parking systems rely on point measurements from the USS sensors which become useless if the obstruction is sufficiently close to pass between the sensors.

A system that operates on a full 3D model of the world around the car has the potential to avoid incidents like this.
Exactly. That car wasn't using Tesla Vision so the example only highlights another USS limitation.
 
Tesla shot themselves in the foot by releasing the circa-June version of TV park assist
I only semi agree. As you point out it is very useful as it is.

There will always be some who will judge usefulness on how pretty the picture is but to the rest of us who judge on functionality it is clearly a step up from USS.

I think those sorts of people will get over it once it becomes pretty enough for them.
 
  • Like
Reactions: STUtoday
Is Tesla still using the ultrasonic sensors? A week ago I replaced my automatic truck opener in my 2018. When I put the bumper back on I forgot to reconnect the USS cable. I received a Park Assist not available until I reconnected it.
I thought they stopped using them in all cars, but maybe they just never corrected the error codes.
 
I only semi agree. As you point out it is very useful as it is.

There will always be some who will judge usefulness on how pretty the picture is but to the rest of us who judge on functionality it is clearly a step up from USS.

I think those sorts of people will get over it once it becomes pretty enough for them.
I think you misinterpreted me - I’m criticising the decision to release the half assed version that came out 6 months or so ago that attempted to replicate the ‘wavy line’ USS interface.

That was occasionally accurate but was wrong often enough that it just made it look like the tech wasn’t going to work.
 
I think you misinterpreted me - I’m criticising the decision to release the half assed version that came out 6 months or so ago that attempted to replicate the ‘wavy line’ USS interface.

That was occasionally accurate but was wrong often enough that it just made it look like the tech wasn’t going to work.
Ahh I get you, yes, I agree. People forget very quickly how many times Tesla have proven all of the 'experts' wrong time and time again.
 
  • Like
Reactions: Mullermn
Is Tesla still using the ultrasonic sensors? A week ago I replaced my automatic truck opener in my 2018. When I put the bumper back on I forgot to reconnect the USS cable. I received a Park Assist not available until I reconnected it.
I thought they stopped using them in all cars, but maybe they just never corrected the error codes.
They still work on vehicles that have them. Newer builds (depending on where you are in the world - In Australia only the latest Model 3 at this point) don't have them.
 
  • Like
Reactions: STUtoday
Is Tesla still using the ultrasonic sensors? A week ago I replaced my automatic truck opener in my 2018. When I put the bumper back on I forgot to reconnect the USS cable. I received a Park Assist not available until I reconnected it.
I thought they stopped using them in all cars, but maybe they just never corrected the error codes.
The cars that have USS hardware still use it.

It's the 1D radar that isn't used anymore, even if you have the hardware.
 
Removing USS sensors was primarily a cost saving measure. (Check out Sandy Monroe's YouTube on the subject). Having just finished Walter Isaacson's autobiography on Elon, it's clear on his obsession with removing all parts and processes that aren't absolutely essential. Hence removing radar, steering wheel stalks, etc. Hey, he even asked the guys installing his solar tiles to only use one nail not two, until the tiles starting rotating and then he reluctantly agreed that two were better.
 
The most gimmicky nonsensical feature in the Tesla.

All we need is ultrasonic sensor and front bumper cameras. We dont need a visual map. We don’t know a 3d diagram. Parking is the simplest action that was already solved 25 years ago. Tesla is making the act of parking more difficult and complex.

A half baked feature with zero value.
Cars today should be making it easier to do monotonous stuff like park (they are). Front bumper cam on newer Teslas would make parallel parking and pulling into certain spaces better. Plus, vision wouldn't have to remember objects because, well, it could see them. But seems they don't want to go that route. Oh well.
 
The cars that have USS hardware still use it.

It's the 1D radar that isn't used anymore, even if you have the hardware.
But this has to go away in version 12. When will all be on one single stack? or are there gonna be two versions of 12 with and without ultrasonic? Eon said vision only, did he meant vision plus ultrasonics on early adopters? How are mine being used?
Hopefully someone will sell a third-party box that I can plug into my radar and put on a display. A nice piece of equipment to be left dead in the bumper. I’ve always wanted to buy a radar for R&D, I’m glad I could help Tesla decide what they sold me was stupid and garbage. But now they have seen the light, less equipment is more, , I mean more profit. Maybe they could incorporate the radar into the gaming.” Ultrasonic, cat quest, radar edition”.. (every input is an error or catnip)
 
Last edited:
  • Disagree
Reactions: STUtoday
But this has to go away in version 12. When will all be on one single stack? or are there gonna be two versions of 12 with and without ultrasonic? Eon said vision only, did he meant vision plus ultrasonics on early adopters? How are mine being used?
Hopefully someone will sell a third-party box that I can plug into my radar and put on a display. A nice piece of equipment to be left dead in the bumper. I’ve always wanted to buy a radar for R&D, I’m glad I could help Tesla decide what they sold me was stupid and garbage. But now they have seen the light, less equipment is more, , I mean more profit. Maybe they could incorporate the radar into the gaming.” Ultrasonic, cat quest, radar edition”.. (every input is an error or catnip)
USS has nothing to do with FSD.

USS is only useful at very close distances for parking, it's never been used at driving speeds and distances to cars on the road.
 
You were talking about two versions of V12 (FSD) one for with Ultrasonic Sensors, one without.

FSD has never used Ultrasonic sensors, so there is no need for two versions of it.
I am talking about version 12 doing parking assist, how complicated can this be? So are you saying my car will always use ultrasonic sensors for Park assist and not use vision only? Please educate Elon or yourself on how vision only works. I was not talking about two versions of FSD.
Are you saying the full self driving module does not park the car? Is that another computer? Is it using dojo to park the car or are we waiting on the Tesla bot. I get so sick of the Tesla Shills just saying how great everything is and not providing any explanation. If Tesla would ever release the software product I bought years ago we could discuss how the hell it works. The product has only been a release to employees and shills. Releasing to employees is not a release, that is called product testing. I know the world has changed since we have a kickstartarter carnival huckster instead of an engineer selling future tech to the uneducated.
But we can continue to discuss two different versions of FSD if you like