Better safety through radar

CyberGus

Well-known member
First Name
Gus
Joined
May 22, 2021
Threads
67
Messages
5,807
Reaction score
19,080
Location
Austin, TX
Website
www.timeanddate.com
Vehicles
1981 DeLorean, 2024 Cybertruck
Occupation
IT Specialist
Country flag
Karpathy showed how low-resolution radar would return inconsistent and/or low-value data, making it a net negative with vision. That does not preclude the use of high-res radar, or lidar, or other technologies in the future, many of which Tesla has tested. The rumor about "Tesla going back to radar" stems from their FCC application to use radar, but the application itself doesn't mean much.

The problem with introducing new technology for FSD is that Tesla would then need to provide free hardware upgrades to all current FSD holders. Even the looming HW4 computer slated for the Cybertruck has some worried that the older cars will be unable to keep up.
Sponsored

 

FutureBoy

Well-known member
First Name
Reginald
Joined
Oct 1, 2020
Threads
207
Messages
3,522
Reaction score
6,012
Location
Kirkland WA USA
Vehicles
Toyota Sienna
Occupation
Financial Advisor
Country flag
I wonder how much risk people will be willing to accept when they are no longer in control?
Humans are amazing at holding differing standards for self versus others. In my experience, the majority of humans can easily self-justify dangerous behavior that they do themselves. But if another human partakes in a behavior that is similarly dangerous it is easy to condemn the behavior as reckless endangerment. Now if a robot or other non-human were put in place with even the slightest bit of a danger profile, that would be irrationally condemned with extreme prejudice.

It generally will take one serious incident (especially with a horror result like 3rd-degree burns/death) before people will actually take notice though.
 

electricAK

Well-known member
Joined
Mar 5, 2020
Threads
8
Messages
230
Reaction score
547
Location
Haines, Alaska
Vehicles
Cybertruck dual-motor
Country flag
I'm afraid you are the victim of showmanship, technical naivete and Elon's faith in what AI can do.

....

The only thing that is potentially different is the claim that with AI they can process a temporal sequence of images to locate objects in 4D.
Your technical explanation about photodetectors is great. But the question really boils down to your point in the last paragraph: "The only thing that is potentially different is the claim that with AI they can process a temporal sequence of images to locate objects in 4D." This is exactly what they're doing with AI, and I believe they are succeeding.

Many more photons make it through a fog or a dark night than we can observe with our eyesight. In fact, Lidar is exactly this. The laser is shot in a scanning motion and only covers all points in the field over a period of time. You have to integrate all the returns over this period to form an image of what is in front of you. Most photodetectors don't do this, but they could! By integrating photo signals from multiple frames you can catch the occasional photons that make it through the fog or the dark or the dust. It's not perfect, but it's far better than what our eyes can see, and that's all that really matters.

Lately I've been working with nadir-pointing helicopter-flown aerial lidar data. The lidar is capable of seeing through the forest canopy and mapping the ground surface below. How does it do this? The laser itself doesn't go through the trees, it bounces off. But out of millions of laser pulses, since the helicopter is moving, some of the pulses over a period of time make it through the canopy through a little gap in the leaves, and bounce off the ground below. Thus you have to shoot millions and millions of points to get a good map of the ground below. But it works.

You can apply the same processing to a camera field to help ID objects that are obscured and just to weak to detect with a single image frame. With multiple frames, a signal starts to emerge from the noise. Again it's not nearly as good as Lidar or Radar, but far better than what we can see with our eyes.

All of those collisions were with radar-enabled Teslas.
This is part of the reason I'm getting the impression that radar is adding more noise to the system than the benefit of occasionally detecting an unseen object.
 

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
I don’t have radar. How can I have 20+ years of safe driving without radar?

Are we going to start kicking humans off the road for driving without radar?

Not against Radar, but I also don’t see it as necessary. A mixture of cars which drive crazy fast in near white out conditions and some which cannot sounds like a recipe for trouble.

Seems to me like slowing down to human speeds is the prudent solution.

At that point, why radar?
 

John K

Well-known member
First Name
John
Joined
Jan 2, 2020
Threads
41
Messages
2,803
Reaction score
5,768
Location
Los Angeles
Vehicles
Volt, CT reserve day 2
Country flag
I think we should add sonar plus Sony’s original infrared nightshot capabilities. (Don’t google at work)
 


CyberGus

Well-known member
First Name
Gus
Joined
May 22, 2021
Threads
67
Messages
5,807
Reaction score
19,080
Location
Austin, TX
Website
www.timeanddate.com
Vehicles
1981 DeLorean, 2024 Cybertruck
Occupation
IT Specialist
Country flag
Your technical explanation about photodetectors is great. But the question really boils down to your point in the last paragraph: "The only thing that is potentially different is the claim that with AI they can process a temporal sequence of images to locate objects in 4D." This is exactly what they're doing with AI, and I believe they are succeeding.
https://en.wikipedia.org/wiki/Object_permanence
"Object permanence is the understanding that objects continue to exist even when they cannot be sensed. This is a fundamental concept studied in the field of developmental psychology, the subfield of psychology that addresses the development of young children's social and mental capacities."​

The concept of "Object Permanence" does not fully develop in humans until 1-2 years of age (and never in some dog breeds that think I threw an invisible ball lol), but teaching this to a NN computer is kind of a big deal.
 

CyberGus

Well-known member
First Name
Gus
Joined
May 22, 2021
Threads
67
Messages
5,807
Reaction score
19,080
Location
Austin, TX
Website
www.timeanddate.com
Vehicles
1981 DeLorean, 2024 Cybertruck
Occupation
IT Specialist
Country flag


Crissa

Well-known member
First Name
Crissa
Joined
Jul 8, 2020
Threads
126
Messages
16,211
Reaction score
27,068
Location
Santa Cruz
Vehicles
2014 Zero S, 2013 Mazda 3
Country flag
Why now? Tesla tossed RADAR.

Its additive, layering and what purpose is PUREvision in need RADAR can fill?
Because it's the NTSB's job to find out 'why' even if it's late to the game. It's about people complaining to the NTSB, mostly.

-Crissa
 

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
Your technical explanation about photodetectors is great. But the question really boils down to your point in the last paragraph: "The only thing that is potentially different is the claim that with AI they can process a temporal sequence of images to locate objects in 4D." This is exactly what they're doing with AI, and I believe they are succeeding.

Many more photons make it through a fog or a dark night than we can observe with our eyesight. In fact, Lidar is exactly this. The laser is shot in a scanning motion and only covers all points in the field over a period of time. You have to integrate all the returns over this period to form an image of what is in front of you. Most photodetectors don't do this, but they could! By integrating photo signals from multiple frames you can catch the occasional photons that make it through the fog or the dark or the dust. It's not perfect, but it's far better than what our eyes can see, and that's all that really matters.

Lately I've been working with nadir-pointing helicopter-flown aerial lidar data. The lidar is capable of seeing through the forest canopy and mapping the ground surface below. How does it do this? The laser itself doesn't go through the trees, it bounces off. But out of millions of laser pulses, since the helicopter is moving, some of the pulses over a period of time make it through the canopy through a little gap in the leaves, and bounce off the ground below. Thus you have to shoot millions and millions of points to get a good map of the ground below. But it works.

You can apply the same processing to a camera field to help ID objects that are obscured and just to weak to detect with a single image frame. With multiple frames, a signal starts to emerge from the noise. Again it's not nearly as good as Lidar or Radar, but far better than what we can see with our eyes.



This is part of the reason I'm getting the impression that radar is adding more noise to the system than the benefit of occasionally detecting an unseen object.
Hence my comment regarding Samsung ToF cameras that give range information per pixel per frame and could use optimised radiation frequencies less sensitive to atmospheric disruptions.
 

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
As previously postulated, humans are poor at calculating risk.
That's obviously not true. Proof: you are here to post this. Thus you and your forebears have assessed risk adequately to have survived. QED
Autonomous vehicles need to exceed human capability to be considered "equal".
Autonomous automobiles must be virtually flawless because society is afraid of anything it doesn't understand and will pounce on any incident in which the automaton failed as a reason to snuff the technology. We've already seen that.
 

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
why radar?
Radar is just another sensor that can be added to a system to enable the system to make a better estimate of the nature of its surroundings.
 

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
The NTSB has a mandate to pursue Tesla, which are NOT based on comparisons between vehicles equipped or not equipped with driving aids, rather between only those that are equipped, of which Tesla vehicle represents the dominant majority of all vehicle, because not many others do it. Their reports have been weighted to ignore that completely.
Sponsored

 
 




Top