[uk] UltraSonic Sensors removal/TV replacement performance

[Mullermn]

I see people pondering over 'sensor disagreement' as if Tesla are struggling with something completely new and unexplored. Fact is it's nothing new at all. Sure an autonomous car for public roads is a new(ish) application, but complex autonomous systems with multiple sensors have been around for decades.

This is more relevant than you might think. Data is never perfect and you have to program assumptions to work around it's short comings, but as you expand the range of assessments you want a system to make you can have scenarios where those assumptions start to conflict.

Here's an example.. You're driving down the road following a vehicle. Your radar sensor is happily reporting that vehicle is moving at 70mph. Suddenly you get a reading that says the vehicle is stopped. Do you trust that and slam on the brakes or do you filter it out? After all, vehicles can't literally go from 70mph to stationary instantly, can they? What you saw is probably a bridge over the road (traditional radar has very limited vertical resolution, remember). 99% of the time you'll be correct.

Now lets say you're driving along behind a 70mph vehicle and a tree falls in to the road due to freak wind. Your radar sensor reports that a vehicle decelerated in a hugely improbable fashion so you throw the sensor reading out, and as a result hit the tree.

The point is not that the specific scenario I made up is a problem, but the general case that the more nuanced you expect your system to be the less you can afford to smooth the data because you start wiping out genuine data points. It could be that every emergency braking radar on cars today would fail the 'tree' test above but we never find out because such an occurrence is freakishly rare (and 99% of cars don't have forward facing cameras recording all the time for internet dissection), and other manufacturers are maybe happy to live with these limitations because they don't also intend to build a self driving system that SHOULD be able to spot an object has fallen in to the carriageway and navigate around it.

Andrej Karpathy has a segment on this in this video:

 

[Boza]

But yet a deaf person is still allowed to drive.

It is considered impairment and requires special equipment in the vehicle. So, legally, relying on vision only is considered less than optimal.

 

[Boza]

This is more relevant than you might think. Data is never perfect and you have to program assumptions to work around it's short comings, but as you expand the range of assessments you want a system to make you can have scenarios where those assumptions start to conflict.

Here's an example.. You're driving down the road following a vehicle. Your radar sensor is happily reporting that vehicle is moving at 70mph. Suddenly you get a reading that says the vehicle is stopped. Do you trust that and slam on the brakes or do you filter it out? After all, vehicles can't literally go from 70mph to stationary instantly, can they? What you saw is probably a bridge over the road (traditional radar has very limited vertical resolution, remember). 99% of the time you'll be correct.

Now lets say you're driving along behind a 70mph vehicle and a tree falls in to the road due to freak wind. Your radar sensor reports that a vehicle decelerated in a hugely improbable fashion so you throw the sensor reading out, and as a result hit the tree.

The point is not that the specific scenario I made up is a problem, but the general case that the more nuanced you expect your system to be the less you can afford to smooth the data because you start wiping out genuine data points. It could be that every emergency braking radar on cars today would fail the 'tree' test above but we never find out because such an occurrence is freakishly rare (and 99% of cars don't have forward facing cameras recording all the time for internet dissection), and other manufacturers are maybe happy to live with these limitations because they don't also intend to build a self driving system that SHOULD be able to spot an object has fallen in to the carriageway and navigate around it.

Andrej Karpathy has a segment on this in this video:

Ground proximity radar and barometric altimeter disagree most of the time, yet somehow airplanes keep using both.

 

[Zilla91]

Single stack

Everybody drink!!!

 

[LiamPope]

This is more relevant than you might think. Data is never perfect and you have to program assumptions to work around it's short comings, but as you expand the range of assessments you want a system to make you can have scenarios where those assumptions start to conflict.

Here's an example.. You're driving down the road following a vehicle. Your radar sensor is happily reporting that vehicle is moving at 70mph. Suddenly you get a reading that says the vehicle is stopped. Do you trust that and slam on the brakes or do you filter it out? After all, vehicles can't literally go from 70mph to stationary instantly, can they? What you saw is probably a bridge over the road (traditional radar has very limited vertical resolution, remember). 99% of the time you'll be correct.

Now lets say you're driving along behind a 70mph vehicle and a tree falls in to the road due to freak wind. Your radar sensor reports that a vehicle decelerated in a hugely improbable fashion so you throw the sensor reading out, and as a result hit the tree.

The point is not that the specific scenario I made up is a problem, but the general case that the more nuanced you expect your system to be the less you can afford to smooth the data because you start wiping out genuine data points. It could be that every emergency braking radar on cars today would fail the 'tree' test above but we never find out because such an occurrence is freakishly rare (and 99% of cars don't have forward facing cameras recording all the time for internet dissection), and other manufacturers are maybe happy to live with these limitations because they don't also intend to build a self driving system that SHOULD be able to spot an object has fallen in to the carriageway and navigate around it.

Andrej Karpathy has a segment on this in this video:

Your example just demonstrates the folly of relying on a single sensor, in your case radar only. Your radar car doesn't know what to do. A sophisticated system gets a second opinion from cameras, third from lidar, and between them reliably arrive at the correct interpretation. It's difficult to implement, buy way better than just throwing away data because you just can't deal with that. That's curl up into a ball and hope thinking.

 

[MP3Mike]

It is considered impairment and requires special equipment in the vehicle. So, legally, relying on vision only is considered less than optimal.

It varies by state. Oregon doesn't require any special for deaf drivers. No special equipment required... You don't even have to report it to the DMV. (You can optionally report it and ask to have it listed on your license/registration so that police know, but it isn't required.)

 

[CWT3LR]

It could be that every emergency braking radar on cars today would fail the 'tree' test above but we never find out because such an occurrence is freakishly rare (and 99% of cars don't have forward facing cameras recording all the time for internet dissection), and other manufacturers are maybe happy to live with these limitations because they don't also intend to build a self driving system

Most of the companies in the EV space are carrying cameras that are multiples of those on Tesla. The Tang, for example, has a full 360 degree camera suite with something like 15 cameras, as well as 11 radars. Waymo similarly has multiple radars, more cameras - and Lidar too. And since they are the closest anyone has yet come to true FSD, I trust their tech more than just a few low res Tesla cams that get spooked by early evening sun.

We will see, now that the competition is heating up. Intel and Apple bring almost access to almost unlimited capital if they desire it, they are at least as prestigious if not more so, so can poach talent from whoever looks closest to a real world solution, and are likely to be able to design a UI that is elegant and user friendly. They might even let people specify USS if they are so outrageous as to request it....

 

[GlynG]

I really don't know what it is your saying except a single source of data shouldn't determine an action - or should it?

From my perspective the quest for FSD is a long way to go and perhaps Teslas use of the phrase and peoples general understanding of what FSD means is completely different. Driving is super complex for a computer but a cinch for the human brain - though as humans we get distracted or take chances and that results sometimes in an accident.
It doesn't help when Elon makes comments regarding Robotaxi - or his latest - the car will go and find its own parking space - as it reaffirms our understanding of what he says Tesla will be able to do.

The point regarding USS - the quest for vision only is good and therefore the removal of USS is sensible - but only when the AI vision system works to a similar standard of the old trusted tech of USS, I can understand people not being happy with no working system at all.
People could have a cray supercomputer to use as a calculator - but its a sledgehammer to crack a nut and would be plain stupid - likewise adding further complications and additional inputs of data into a computer that simply cannot process all the variables it already has to handle doesn't make sense.
We have a computer in the car that is supposed to be able to handle FSD - it hasn't worked so far and at best is an advanced cruise control in some conditions, it cant handle reliably the inputs for auto wipers or auto main beam either and yet Tesla now want the same computer to also handle the data from cameras for low speed parking, Why not simplify the tech, use stand alone proven technology like USS, rain sensors and Auto main beam and by doing so will free up some processing power of the CPU, stand alone means no data conflicts, people would be happy with the performance of the USS, the wipers and main beam and wouldn't even think or care how its done - it doesn't matter as long as it works.

By trying to be too clever too soon and with inadequate processing power Tesla are antagonising their customer base - they are making us unhappy despite some super wonderful battery and motor technology they have in their cars.

Besides the charging network, the rather nice battery management and rather clever dual motors in my car the things I'm most pleased with are the aftermarket Hansshow power boot and frunk kits I fitted, the illuminated sill plates, the illuminated rear reflectors with sequential indicators - add ons that simply work - every time. If Hansshow released a modification to control my wipers and auto main beam Teslas connection to them would be disconnected. I keep my fingers crossed that in the future Tesla doesn't switch off my USS for their vision system like they did my radar and made things worse.

 

[Boza]

It varies by state. Oregon doesn't require any special for deaf drivers. No special equipment required... You don't even have to report it to the DMV. (You can optionally report it and ask to have it listed on your license/registration so that police know, but it isn't required.)

Sure. There is always exception. Vatican does not even require driver license.

 

[CWT3LR]

Ground proximity radar and barometric altimeter disagree most of the time, yet somehow airplanes keep using both.

That's why the RAF always matched a navigator with a pilot. In the event of baro and prox radar disagree, the navigator dangled a plumb line out and then hauled it in to give a true reading.

 

[Danny_F]

Some very nice cars being developed in China now. I was struck by the fact they have no shame in ripping off other people's ideas and designs. View attachment 893880

That is just one of the reasons why China is becoming persona non grata within the world. They use their intellect to steal intellectual property and are very open and brazen about it. I can see that China will be shunned by the world with consequences for all. If you think things are difficult now just wait for the backlash at China and the resulting difficulties as we have grown so dependent on Chinese goods we will struggle without them.

 

[Battpower]

Has there actually been any confirmation what this radar is for? Everyone's making the assumption that it's a drop in replacement for the old radar that was removed. Last I heard, it's possible it's not even intended for use on Tesla's cars?

I agree the function is not known.

I typed a comprehensive response that my phone just ate. Any way, most has been said before. If we had responses from Tesla that aligned with user experience, we would have far less discussion and conjecture here.

When FSD, AP, TACC etc are dependable, consistent, fully functional and stable, it won't matter what the sensor suite is.

I've just driven days with blinded cameras and AP regularly struggling to find correct lane position in totally clear weather conditions on high standard French autoroutes (with multiple phantom braking events such that my passengers demanded I stopped using AP - accepted on older sw version, but essentially unchanged behaviour over 3 years of updates).

EM has made positive noises about HD radar, but what can we conclude from his statements?

While the cars have a deficiency as seems evident with the removal of USS and continued reports of phantom braking, and Tesla are apparently looking at HD radar for some application somewhere, it seems to me like a reasonable avenue to explore in these discussions that these might be related.

 

[Type2]

That is just one of the reasons why China is becoming persona non grata within the world. They use their intellect to steal intellectual property and are very open and brazen about it. I can see that China will be shunned by the world with consequences for all. If you think things are difficult now just wait for the backlash at China and the resulting difficulties as we have grown so dependent on Chinese goods we will struggle without them.

It would be great if China were shunned and unable to sell their goods abroad if only because it would weaken a dangerous country. However, if they start to produce excellent cars (using stolen intellectual property) at a low price I’m sure they won’t be short of customers. People tend to look after themselves rather than looking at the greater good. It would take action at government level, and where is the appetite for that?

 

[Danny_F]

At the moment I don't think there is government appetite for it, but what with the government taking action in relation to Huawei and Newport Wafer Fabs I wouldn't want to guess where they will wave their sword next? The country has had their heads buried in the sand for years in relation to China however this may be slowly changing.
However you are 100% correct, people will continue to purchase cheaper Chinese copies as money is more important than morales.

 

[Boza]

That's why the RAF always matched a navigator with a pilot. In the event of baro and prox radar disagree, the navigator dangled a plumb line out and then hauled it in to give a true reading.

See? More sensors are needed, not fewer

 

[Battpower]

this may be slowly changing.

The need for change is very evident. China has always focused on the initial purchase rather than support and longevity. As witnessed by this thread, it is very challenging to maintain support for products over an extended time. However, challenges often equate with opportunity.

money is more important

I would happily pay for keeping my car viable over a long period. It was one of my initial high hopes for Tesla that a business model based around a relatively small, complementary range of vehicles with a commitment to ongoing updates would equate to more sustainable motoring.

 

[yessuz]

Left pillar camera blinded so it's going to be an excellent exercise and flood of threads about parking in the dark and no USS

 

[DrJFoster]

That a ridiculous argument. The radar worked well for basic autopilot. At some point the software outgrew it and it stopped being useful. It really isn’t hard. If you‘re gonna quote me I’ll reply all I like, thanks.

Far from it. The goal was always FSD, Tesla even claimed "buy a car and it has all the hardware you need for FSD". They're even looking at a new better Radar. Tesla have even admitted they could never get it working well.

 

[CaptFish2003]

Had my Model3 in for a service and took a look at the Model Y. Nice car, but from the inside you can't see where the bonnet ends. I know that this is common with other cars and older cars I drove without USS. However I am now so used to the USS which I rely on daily in a tight multi storey car park, that I could/ would not get another Tesla without USS.

 

[CWT3LR]

Had my Model3 in for a service and took a look at the Model Y. Nice car, but from the inside you can't see where the bonnet ends. I know that this is common with other cars and older cars I drove without USS. However I am now so used to the USS which I rely on daily in a tight multi storey car park, that I could/ would not get another Tesla without USS.

Funnily enough, in my experience in the Y, it is not so much the front that USS serves admirably to protect, but the rear three quarters / wheel arch area. Unlike most cars I have had, there is a pronounced haunch to the car that I struggle to place with cameras alone.