Anyone else concerned about fog without radar?

myco.rrhizae

Well-known member
First Name
Myco
Joined
Aug 9, 2020
Threads
10
Messages
115
Reaction score
162
Location
N. Cal
Vehicles
F150 4.9L, Chevy Volt, Tri-Motor Cybertruck
Occupation
Carpenter
Country flag
I dont have radar on my vehicles, and I'm not concerned about fog.
Sponsored

 

CyberGus

Well-known member
First Name
Gus
Joined
May 22, 2021
Threads
67
Messages
5,816
Reaction score
19,093
Location
Austin, TX
Website
www.timeanddate.com
Vehicles
1981 DeLorean, 2024 Cybertruck
Occupation
IT Specialist
Country flag

CyberGus

Well-known member
First Name
Gus
Joined
May 22, 2021
Threads
67
Messages
5,816
Reaction score
19,093
Location
Austin, TX
Website
www.timeanddate.com
Vehicles
1981 DeLorean, 2024 Cybertruck
Occupation
IT Specialist
Country flag
Straight ahead foggy progress notwithstanding, fate in ambient fog or frosty fog is dependent on the “other guy” seeing your car.
In sudden blackout conditions, I've found that traffic becomes Game Theory: too fast I hit someone, too slow I get rear-ended, pull over I hit the invisible tree.

I recall a story long ago about a man driving in heavy fog, who stuck his head out the window and just slowly followed the centerline. Unfortunately someone going the opposite direction had the same idea, and they had a "head-on" collision lmao
 

tidmutt

Well-known member
First Name
Daniel
Joined
Feb 25, 2020
Threads
8
Messages
603
Reaction score
992
Location
Somewhere hot and humid
Vehicles
Model Y Performance, Model X P100D
Occupation
Software Architect
Country flag
In sudden blackout conditions, I've found that traffic becomes Game Theory: too fast I hit someone, too slow I get rear-ended, pull over I hit the invisible tree.

I recall a story long ago about a man driving in heavy fog, who stuck his head out the window and just slowly followed the centerline. Unfortunately someone going the opposite direction had the same idea, and they had a "head-on" collision lmao
I confess, I too wonder about vision in bad weather scenarios. However, I think perhaps the right answer is that AI vision will do better than a human and that slowing down is really the best overall solution at least for now. So if a human has to slow down, and still makes more mistakes than a vision system at the same speed and then that vision system is better in normal driving conditions than vision + radar, then we are better off overall without radar, even if radar made fog easier to deal with.

Maybe when all cars are equipped will FSD and can see through fog effectively then we can let our cars just drive at higher speeds through those conditions.

One thing that always confused me is that when I'm driving through fog I have always found that flashing hazards made vehicles substantially more visible. Yet in many states of the US it's illegal to drive with them on even in fog or heavy rain (although Florida recently made it legal). It seems to me that this is something vision based systems could easily detect if it became common and indeed, even required. Or why not require manufacturers to flash the hazards at a different frequency when driving versus when stopped?
 


ajdelange

Well-known member
First Name
A. J.
Joined
Dec 8, 2019
Threads
4
Messages
3,213
Reaction score
3,403
Location
Virginia/Quebec
Vehicles
Tesla X LR+, Lexus SUV, Toyota SR5, Toyota Landcruiser
Occupation
EE (Retired)
Country flag
I’m going to put this on top of my Cybertruck so I will be safe in a fog.

1626669832800.jpeg
That won't do you any good. If you are following a car at constant distance that means you are going at the same speed and it will disappear from the screen!
 

Delusional

Well-known member
First Name
Phil
Joined
Dec 5, 2019
Threads
11
Messages
197
Reaction score
310
Location
Pittsburgh
Vehicles
F-150
Occupation
Construction
Country flag
The Fog Problem: (and rain, or smoke)
A little bit of signal, a lot of noise.

Filtering a little bit of signal out of a lot of noise is something computers are truly great at doing.
I guarantee that in fog conditions Tesla Vision sees more than you or I can see, and is able to make better sense of what it sees.
Is that enough to keep the vehicle rolling? That is another question.
 

ajdelange

Well-known member
First Name
A. J.
Joined
Dec 8, 2019
Threads
4
Messages
3,213
Reaction score
3,403
Location
Virginia/Quebec
Vehicles
Tesla X LR+, Lexus SUV, Toyota SR5, Toyota Landcruiser
Occupation
EE (Retired)
Country flag
Particulates (smoke, fog) scatter light from the target and the surroundings so that S (signal) goes down and I (interference) goes up meaning that S/(I + N) is low and detection performance drops off. This is because the particles are of size comparable to the wavelength of the radiation involved. With radar the wavelength is much longer and particulates don't scatter the rays. That's why radar can see where you, nor Tesla vision can. Tesla has decided that their cars don't need to be able to see in fog and smoke. They took the radar out.
 

JBee

Well-known member
First Name
JB
Joined
Nov 22, 2019
Threads
18
Messages
4,752
Reaction score
6,129
Location
Australia
Vehicles
Cybertruck
Occupation
. Professional Hobbyist
Country flag
I've been driving with radar for over a decade, and even by itself its saved me by doing an emergancy brake at least 5-6 times. Most of the cases where actually with kangeroos, only twice with a car stopping in front of me.

Typical thing with roos is they are curious cats and like to come and have a look if something is going on. They also always come in pairs or more, so while you're swerving one the next one is already eyeballing to hit you. Some can be as big as a calf...just airbourne. Having a roo/bullbar doesn't always help if they're in the air. Just don't land them on the windscreen if you can, at best their insides like sqirming and spraying through the cracked glass, at worse they'll land in your lap, and you'll be in the backseat. Seatbelts and airbags don't help if you have kangas hitching a ride. Hasn't happened to me, but I've seen it plenty. Not pretty.

They also have poor vision when being zapped with LED light bars so their ears work overtime. Having some whistlers helps heaps, they xan detect direction better with sound, but a blast from a airhorn gets them moving from a far off if you can see them first.

This is all compounded by the fact that our roads are all well drained, so the best feed in town, in our dry country, is exactly along side the roads where the little bit of rain we have drains to and grows the bushes.

The other thing they like to do is stand on the side of the road then jump in the middle of the road to play chicken with you, then spin around and jump straight back out. But the best one is when they sit on the edge and stare you down, then just when you get close they think they have to escape to otherside of the road, instead of staying put.

This is wear radar has saved me a few times, so whilst driving with low beam because of oncoming traffic, before I had the chance to do anything, radar had triggered emergency breaking, seat belt tensioners would go off pulling us back into the seat etc with the car making things pre- crash safe. Same for dead roos on the road you can't see but would of wiped out the bottom of the car. Its weird, but cool.

Its not just fog it helps with, its anything you can't see in the dark, or in high contrast or reflective night time driving situations, or when you're passing other vehicles at night.

The other thing with our radar cruise is that it could always accurately measure the speeds of large things and distinguish things that are moving and things that aren't. Technically this is pretty obvious because it can compare the dopplar data with the vehicles own speed sensor, and by doing so can more accurately decelerate and accelerate with traffic, because it knows before you can see a change, that the vehicle ahead is slowing down or speeding up. All whilst maintenaining a safe distance giving you ample time to intervene. Most rear enders happen because people can't tell how fast the car in front is going before its to late. Radar measures it accurately, so it knows exactly and slows at the exact right rate to match before you can.

Obviously you can mostly do the same with CV, but the radar on our Toyotas never glitched on bridges or traffic, and only switched off in really heavy rain when even the wipers didn't help.

The question is just how good are the Tesla cameras and how much better will they be with the new Samsung deal.

Samsung also have ToF camera sensors (on some of their phones too) that can be used as 3D scanners, because they can measure the flight time of light, so range accuracy is high.

They also have embedded AI camera silicon that should be able to delegate some CV processing reducing cpu bandwidth on the main FSD loop. The trick with good CV algorithms is to learn how to see how the machine sees, rather than teach it how we see. Another factor is to exclude as much irrelevant data to get you processing frame rate up. Things like plain road or building surfaces, or trees etc can simply be ignored. But things on them can be tracked.

Obviously the biggest improvement is reaction time with CV and radar, and the ability to modulate the controls for smooth effective response beyond what is humanly capable. Those PID loops really need to be tuned though, which can be an art in itself. Of course nowadays they have autotune. And no its not for the music blasting on Spotify (or radio for those that like commercials stiil)

And that leaves me with some things I think Tesla should add. One is passive SDR radar, where they can monitor EMF and its reflections from other vehicles and obstacles to generate 3D maps using 2 or 3 antennas on the car, that could also double for 4/5G comms. The other is passive sonar, where they can do the same using tyre noise etc to monitor tyre performance, traction and road surface along with obstacle and bump detection.

And the third is to use those pretty phased array antennas from the starlink dish so they can beamform radar signals around Crissas massive head so I can see that next off topic political arguement coming on this thread.... lol just kidding bud. :)
 
Last edited:


Ogre

Well-known member
First Name
Dennis
Joined
Jul 3, 2021
Threads
164
Messages
10,719
Reaction score
26,998
Location
Ogregon
Vehicles
Model Y
Country flag
There is this growing problem where the internet makes everyone think they can be an expert on any topic by spending 15 minutes with a search engine.

I wonder if these instant experts go to their proctologist's office and argue with their proctologist about what finger is the most sensitive.

"Was that your index finger? Because when I googled this earlier, I read that the thumb was has 40% more nerves than the index finger and is better for this procedure. Maybe you should try your thumb instead."

I'm a strong believer in expertise. I hire a doctor I trust and don't second guess her expertise. I hire a lawyer I don't trust, but who I know will lie well for me. I hire Tesla to figure out AI/ computer vision for me because I sure as hell aren't going to figure out more about than they know with 15 minutes of Google.

If you trust Tesla to be your computer vision people, then you have to accept their opinion on what the best way to accomplish that goal. If you don't trust Tesla to make this choice.... then why in gods name are you buying a car with safety features designed by Tesla?

There is no amount of armchair expertise in either direction that is going to replace the depth of knowledge and experience Tesla's (and GM, and Ford, VW, etc etc) have.

If 20 minutes with Google could answer these questions, then self driving vehicles would have been solved a decade ago.
 

Crissa

Well-known member
First Name
Crissa
Joined
Jul 8, 2020
Threads
126
Messages
16,227
Reaction score
27,087
Location
Santa Cruz
Vehicles
2014 Zero S, 2013 Mazda 3
Country flag
So, alot are asking, 'why not use radar?'

There are two main branches here: How radar is less good because of how driving works; and how fused sensors are less good because how AI works.

We've discussed how signposts, overpasses, freeway barriers, all look like cars on radar, and hills/turns are completely invisible to radar.

But next, we haven't discussed how the AI uses this data. Now FSD Beta isn't yet programmed for weather. It will be, but it's still in beta and is basically a teen driver. It need to learn. So all this applies to the method it uses to see the world around it.

The AI, when it processes the data, is basically always using all of its attention. It doesn't get target fixation like a human, it doesn't get complacent. The visual processes are identifying objects and shapes and edges every frame of video - then comparing how those edges change between frames and between camera angles.

So with one camera, between each frame, the AI can measure the relative speed of an object by this size-change. And it does this for every object it sees.

The AI also can tell how far these objects are away using the parallax between its multiple camera views. That gives the AI a position in space to assign where these objects are around it. This short-term memory is the hardest thing for the AI.

Then it has to identify what these objects are to predict what they'll do. This is mixed into the find-edges stuff, so there are parallel processing of the image.

But that's how the 'mind of FSD' is finding things. This isn't how it chooses a path.

The radar bounce coming back only have distance information. The vector information is fuzzy. So when this information is fused into the visual data, it's a bunch of ghost surfaces the AI has to match to visual surfaces. And by that point, it has identified more objects with visual, the radar false positives just create extra work for no additional benefit.

The AI never slacks off. It doesn't get distracted. It doesn't need a 'look over here' alerts. Because it's already looking. It's hyper-aware. (Which isn't always a good thing!) The radar is only reliable in certain places on the map, because of terrain. Should it start remembering where to ignore its radar? Then what good is it?

-Crissa
 
Last edited:

Ogre

Well-known member
First Name
Dennis
Joined
Jul 3, 2021
Threads
164
Messages
10,719
Reaction score
26,998
Location
Ogregon
Vehicles
Model Y
Country flag
For what it's worth.

Even if you are not using "FSD", all of Tesla's automated driving and safety features will be using the "Mind of Car" engine that FSD offers. That includes, all the collision avoidance, autopilot, traffic aware cruise, everything.

So even if you "Don't use FSD", you will be using the same world view that FSD uses.
 

ajdelange

Well-known member
First Name
A. J.
Joined
Dec 8, 2019
Threads
4
Messages
3,213
Reaction score
3,403
Location
Virginia/Quebec
Vehicles
Tesla X LR+, Lexus SUV, Toyota SR5, Toyota Landcruiser
Occupation
EE (Retired)
Country flag
, they xan detect direction better with sound, but a blast from a airhorn gets them moving from a far off if you can see them first.
In my very limited experience with this I always found that just blowing the horn would scare them off. You have to see them, though.


Those PID loops really need to be tuned though, which can be an art in itself. Of course nowadays they have autotune.
Fuzzy and neural.
Sponsored

 
 




Top