Tesla's vision-only autonomous driving system will be powered by a supercomputer with...

jsilva

Posts: 325   +2
In brief: Tesla is at the forefront of self-driving systems. Currently Tesla cars use cameras, radar, and LiDAR sensors to collect data that helps the system navigate safely, but the car manufacturer now plans to replace it with a vision-only system using cameras and a powerful supercomputer hosting a neural network.

Collecting data for a self-driving system using only cameras instead of radars, LiDAR, and other components may seem inferior, but there are some benefits from this approach. When cutting on the amount of technology packed inside the vehicle, two other things reduce proportionally: costs and weight.

Moreover, there's Elon Musk's argument on vision vs radar: "When radar and vision disagree, which one do you believe? Vision has much more precision, so better to double down on vision than do sensor fusion." He then added that vision is also faster than radar and LiDAR, concluding that as "vision processing gets better, it just leaves radar far behind."

For the neural network host, Tesla will be using a system called Dojo. The system is still under development, but during CVPR 2021, Tesla's head of AI Andrej Karpathy revealed the prototype system that will eventually be replaced by Dojo. The last-gen prototype of Dojo has 5,760 GPUs delivering up to 1.8 EFLOPS (exaFLOPS) and is equipped with 10PB of NVMe storage and a 1.6 TB/s connection. According to Karpathy, this system should sit at the fifth place of the TOP500 supercomputer list.

As for the car, each will be equipped with eight cameras capable of collecting footage at 36FPS. The collected footage is sent to the supercomputer, where it will be processed at a speed matching that of a human driver.

Compared to humans, the system will offer advantages like 360º awareness, better reaction times, and a non-distractible entity controlling the car. Karpathy also mentioned some cases where the system will get into action, including emergency braking to prevent a pedestrian from being hit and warn drivers about traffic lights.

Although the neural network part is still lacking, Tesla has already stopped equipping Model Y and Model 3 cars built in North America with anything besides cameras. As it seems, most of the work in Tesla's new self-driving system is done by the cameras, so the lack of the neural network isn't crucial to functioning accurately.

Permalink to story.

 
I would feel more comfortable with a combination of cameras and ultrasonic sensors (radar) ...

Ultrasonic isn't perfect: when vehicle jumps out of the way, sometimes my car's Adaptive Cruise Control will accelerate and you have to be prepared to cancel it. But, in some conditions with low visibility, ultrasound is actually best because it detects obstacles that may be below visual range.


Cameras can be obscured by various weather conditions. I think it best to keep both sensors but Tesla wants to make these cars cheaper and cheaper.
 
IMO, this sounds like Musky bloviating again. I bet his marketing department will call this something like Super Auto Pilot - and then blame drivers for "using it wrong" when it fails to meet expectations.

Maybe then he will turn all that computing power to mining crypto. 🤣 🤣 🤣
 
I think it best to keep both sensors but Tesla wants to make these cars cheaper and cheaper.
With all the other EVs coming into the market, that will be the only way that "outsider" Tesla will be able to make any money - IMO.
 
Okay, so how does the data get to the supercomputer? I'm assuming cell service but what happens if I don't have service?
Don't worry. Musky/Tesla will figure out a way to add that to the price of the car and/or monthly maintenance. ;)
 
Don't worry. Musky/Tesla will figure out a way to add that to the price of the car and/or monthly maintenance.
Ding ding ding! We have a winner. That seems to be exactly it. Otherwise, how you are going to maintain perfect processing and latency of feeds at all time as with increased load even this supercomputer will need upgrade?

I do wonder though, why buy a car at that point (insurance, electricity, maintenance, dozens of other small costs and now subscription)? Uber would make much more sense in future.

Also, on that point, it seems like Tesla will enter Uber market or some sort of hybrid rent/own system in future.
 
I would feel more comfortable with a combination of cameras and ultrasonic sensors (radar) ...
Ultrasonic used sound waves, not electromagnetic radiation… And ultrasonic sensors are affected by weather conditions, just not dirt or light. Radar seems to be what you’re talking about in your post, and is pretty useful for all the reasons you described.
Okay, so how does the data get to the supercomputer? I'm assuming cell service but what happens if I don't have service?
Apparently they flag certain situations and those clips are saved to be beamed back up to the mothership. But yeah it uses cell service which is paid for as part of the car’s offering. If you don’t have service later, they can’t train self-driving for that specific situation. But considering there are over a million Teslas on the roads with cameras and data connections, that situation is likely to come up somewhere else.
IMO, this sounds like Musky bloviating again. I bet his marketing department will call this something like Super Auto Pilot - and then blame drivers for "using it wrong" when it fails to meet expectations.

Maybe then he will turn all that computing power to mining crypto. 🤣 🤣 🤣
It was Andrej Karpathy in this situation, director of AI, but you might be right that they repurpose it to mine crypto at some point ;)
 
Don't worry. Musky/Tesla will figure out a way to add that to the price of the car and/or monthly maintenance. ;)
A new cell standard, Rg. A new rectal standard where you get bent over for how much it costs
 
I think the author has a total misunderstanding of what Dojo is. It is a supercomputer to train the vision AI algorithm, which is then pushed to a 'regular' computer within each car. Dojo is not a central server where each car's video is absorbed and processed in real time! Really poor article, that this fundamental detail was totally misunderstood....
 
I think the author has a total misunderstanding of what Dojo is. It is a supercomputer to train the vision AI algorithm, which is then pushed to a 'regular' computer within each car. Dojo is not a central server where each car's video is absorbed and processed in real time! Really poor article, that this fundamental detail was totally misunderstood....
So it is like DLSS?
Damn. That would make more sense as the latency and processing requirements were not adding up.

Techspot really dropped the ball in this one. Poorly worded and misleading article.
 
Looks like the days of questioning blame in an accident might also be coming to an end if they get all the cars cameras recording. 2 members of my staff have a new Model S, and they also have 8 cameras, so I guess its pretty standard stuff.
 
This certainly sounds like a path to disaster.

The car is collecting images and SENDING them to the supercomputer????? So, a little glitch in communications, and the car loses self-driving capabilities for 2-3 seconds????? On a highway??????

This is where the stupid 5G hype is leading us. To have a centralized computer for everything and rely solely on 5G to be the medium between the devices and central computer. A completely wrong path.

I wanna car that can self-drive even when the cell network is down. What Elon is trying to do probably can't even get a permit. You can't allow this kind of a torpedo on the streets. Not only is it dangerous for its passengers, it's dangerous for everyone else on the road.


 
This certainly sounds like a path to disaster.

The car is collecting images and SENDING them to the supercomputer????? So, a little glitch in communications, and the car loses self-driving capabilities for 2-3 seconds????? On a highway??????

This is where the stupid 5G hype is leading us. To have a centralized computer for everything and rely solely on 5G to be the medium between the devices and central computer. A completely wrong path.

I wanna car that can self-drive even when the cell network is down. What Elon is trying to do probably can't even get a permit. You can't allow this kind of a torpedo on the streets. Not only is it dangerous for its passengers, it's dangerous for everyone else on the road.
No, its not. See the discussion above.

The article is very poorly worded and is highly misleading. Techspot doesn't seem to care though.
 
Starlink?
Not effing joking, I think you're 100% right. Tesla is taking self driving AI out of it's cars to create a need and market for starlink since cellular infrastructure couldn't handle it.
 
No, its not. See the discussion above.

The article is very poorly worded and is highly misleading. Techspot doesn't seem to care though.

I've read the comments, there's no explanation what happens when the internet connection is lost. It relies solely on the central computer.
 
Back