FSD rewrite will go out on Oct 20 to limited beta

[diplomat33]

I have asked this question a few time whenever I see LIDAR being mentioned. Please, someone tell me how much power a 360deg LIDAR that can see 300m away without interference consumes?

I have a a mini LIDAR that can only properly see 8m away with a 512x512 resolution and it consumes 12W.

Would Tesla be able to accelerate the switch to sustainable energy with a sensor suite that consumes maybe 10-20x more?

i would be willing to bet that the Wayne sensor suite & processing must consume at least 1000W vs ~100W of Tesla’s FSD solution. I would imagine the aerodynamic drag cause my the large sensor horns would also greatly diminish efficiency on the highway.

So can someone confirm how much power these long-range LIDAR systems consumes? Treating point clouds from a LIDAR is extremely expensive and matching LIDAR data with RGB data is also very expensive computationnally. My 512x512 point cloud çan at time overwhelm a 6-core i7 with 2080ti GPU when trying to reconstruct in into a 3D object. You need neural networks to break that data down in a more useful way. LIDAR for FSD is a brute force approach and if you don’t figure out the RGB NN stuff you’re kinda screwed long term anyways.

A lot less than 1000W.

The big spinning velodyne lidar with a range of 120 m consumes 60W. And the solid state lidar is less than 10W.

Velodyne LiDAR - Wikipedia

 

[murc585]

A lot less than 1000W.

The big spinning velodyne lidar with a range of 120 m consumes 60W. And the solid state lidar is less than 10W.

Velodyne LiDAR - Wikipedia

I know you know the Waymo system well. They say their latest sensor suite (in the I-Pace) has a 300m range and I assume they have more than one.

With what you know, what would be your guess on the current system’s power consumption? (including the computing power necessary)

60W for a single sensor with half the range Waymo states is already is not very promising. Imagine, two of these sensors could consume more than Tesla’s complete FSD system. That is a huge engineering challenge. The 5th generation hardware looks terrible right now, like a large add-on and we have not seen this addressed on any production car as of yet. There is still a lot of work to do. I still don’t understand what Waymo’s plans are to integrate this in a car with power consumption matching Tesla’s, aerodynamics matching Tesla’s and cost matching Tesla’s. This to me seems like a massive showstopper as their hardware solution is still many years away of making sense for a mass volume taxi.

 

[diplomat33]

I know you know the Waymo system well. They say their latest sensor suite (in the I-Pace) has a 300m range and I assume they have more than one.

With what you know, what would be your guess on the current system’s power consumption? (including the computing power necessary)

@Bladerskb knows Waymo better than I do. He can provide his own estimate.

But let me try to break it down as best that I can.

There is only 1 lidar that is 300 m, it's the big one at the top. That is a little over twice the range of the spinning lidar I quoted earlier with a power consumption of 60W. So let's say that Waymo's main lidar on top is 120W. The perimeter lidars are short range so probably around 100-150 m. Those are probably around 10W like I quoted earlier. There are 4 of those as you can see in the diagram, the 2 on the sides, and the front bumper and rear bumper ones. So let's say 40W. We see that there is a main long range forward facing radar at the top. My source suggests around 20 W. The perimeter radars are probably shorter range so maybe 10W. There are 4 of those. So that's 60W total. I think cameras are about 1W. Waymo has 29 cameras. So let's say 29W. The computer is probably 20 W, going by Mobileye plans to use 2 eyeQ5 chips of 20W total.

Waypoint - The official Waymo blog: Introducing the 5th-generation Waymo Driver: Informed by experience, designed for scale, engineered to tackle more environments

So my completely educated guess based on other, similar hardware, would be a total power consumption of the entire FSD hardware including the computer to be around 269W. Of course, we don't have Waymo's exact numbers. Waymo custom builds their hardware in-house. It is possible that Waymo has achieved a lower power consumption.

I would also point out that Waymo has ordered 20,000 Waymo I-Pace robotaxis. Surely, they would not have ordered that many if the power consumption was too high and the range was too low. They must be confident that the cars will work for their ride-hailing needs.

60W for a single sensor with half the range Waymo states is already is not very promising. Imagine, two of these sensors could consume more than Tesla’s complete FSD system. That is a huge engineering challenge. The 5th generation hardware looks terrible right now, like a large add-on and we have not seen this addressed on any production car as of yet. There is still a lot of work to do. I still don’t understand what Waymo’s plans are to integrate this in a car with power consumption matching Tesla’s, aerodynamics matching Tesla’s and cost matching Tesla’s. This to me seems like a massive showstopper as their hardware solution is still many years away of making sense for a mass volume taxi.

I know it is natural to want to compare to Tesla since we are Tesla owners on a Tesla forum. But why does the Waymo FSD hardware have to match Tesla in cost, aerodynamics and power? Waymo just needs a city robotaxi. Waymo just needs FSD hardware that can do driverless FSD, at an acceptable cost and range for a city robotaxi. Waymo is not trying to sell cars at a profit like Tesla is. Waymo is not trying to outsell Tesla. As long as Waymo has a city robotaxi that works for their needs who cares if it uses more power or the sensors looks like an add-on?

Personally, I think this pic of the Waymo Jaguar I-Pace looks pretty well integrated into the car. Everything, except perhaps the roof pod, is very well integrated into the car. It definitely looks more integrated than the Chrysler Pacficas. And Waymo has made it more aerodynamic than the hardware on the Chrysler Pacifica's. Sure, the roof pod still looks like an add-on but who cares if you are ride-hailing? Honestly, this pic looks fine to me.

I would also remind you that Tesla's hardware has not proven it can do true driverless FSD yet. If it cannot do that, cost, power and aerodynamics are a moot point. You can have the cheapest, lowest power, most aerodynamic car, but if the hardware cannot do true FSD, then you don't have a FSD car.

Of course, it might not matter for Tesla. Some would argue that cheap, low power hardware that delivers advanced L2 on millions of cars everywhere is still more valuable than more expensive, higher power hardware that delivers true driverless on a limited number of city robotaxis. It depends on what your goals are.

 

[mikes_fsd]

What today, there is lidar available ranging from $100-$1000 all of which would add value to tesla autopilot / fsd... the $100 would add a small amount of value and the $1000 would add much more... obviously.

But you're missing the bigger picture.
Tesla can have over a million FSD capable cars on the road, while perfecting camera vision.
Whereas, with lidar that is economically feasible they would ONLY be able to start from, say, model year 2021!
They would have missed out on a ton of data gathered, learning experience of scaling vision, because internally they would have the false hope "lidar is coming".

Just like Elon said, it would have been a crutch.
But even worse internally, engineers would take the easy way out - "with lidar we can get an accurate enough measurement, so we don't need to worry too much if this is off 10 or 20% of the time" etc etc.

 

[mikes_fsd]

There is only 1 lidar that is 300 m, it's the big one at the top. That is a little over twice the range of the spinning lidar I quoted earlier with a power consumption of 60W. So let's say that Waymo's main lidar on top is 120W. The perimeter lidars are short range so probably around 100-150 m. Those are probably around 10W like I quoted earlier. There are 4 of those as you can see in the diagram, the 2 on the sides, and the front bumper and rear bumper ones. So let's say 40W. We see that there is a main long range forward facing radar at the top. My source suggests around 20 W. The perimeter radars are probably shorter range so maybe 10W. There are 4 of those. So that's 60W total. I think cameras are about 1W. Waymo has 29 cameras. So let's say 29W. The computer is probably 20 W, going by Mobileye plans to use 2 eyeQ5 chips of 20W total.

Are you pulling these out of thin air?
Can you please link to the sources, I cannot find anything on Waymo blog re power consumption of their system.

 

[diplomat33]

But you're missing the bigger picture.
Tesla can have over a million FSD capable cars on the road, while perfecting camera vision.

The 1 millions Teslas are theoretically capable of FSD. Tesla does not have FSD yet.

They would have missed out on a ton of data gathered, learning experience of scaling vision, because internally they would have the false hope "lidar is coming".

Just like Elon said, it would have been a crutch.
But even worse internally, engineers would take the easy way out - "with lidar we can get an accurate enough measurement, so we don't need to worry too much if this is off 10 or 20% of the time" etc etc.

No, because companies like Google developed camera vision too. They did not stop their camera vision work because of lidar. They never tried to do FSD with just lidar. They are doing FSD with advanced camera vision AND lidar. They actually have camera vision that is just as advanced if not more advanced than Tesla. So the idea that lidar is a crutch because it will make FSD companies think they don't need camera vision is silly.

 

[diplomat33]

Are you pulling these out of thin air?
Can you please link to the sources, I cannot find anything on Waymo blog re power consumption of their system.

Like I said, Waymo does not provide any info on power so I pulled from a variety of other sources for hardware similar to what Waymo is using, in order to provide an educated guess. So my numbers are based on other available data. I can provide my direct sources for every little detail if you really want.

As I said:

So my completely educated guess based on other, similar hardware, would be a total power consumption of the entire FSD hardware including the computer to be around 269W. Of course, we don't have Waymo's exact numbers. Waymo custom builds their hardware in-house. It is possible that Waymo has achieved a lower power consumption.

 

[jlbzz]

We are two days out from October 20th. What’re the odds we actually see a private tester upload a video of the rewrite in action?

I’m sure they have to sign NDA’s but will this stop them from anonymously posting?

 

[diplomat33]

We are two days out from October 20th. What’re the odds we actually see a private tester upload a video of the rewrite in action?

I’m sure they have to sign NDA’s but will this stop them from anonymously posting?

I think chances are excellent that we will get a leaked video. We got leaks with Smart Summon even though Early Access was under NDA.

Honestly, I think Elon secretly wants a video to leak because it creates buzz. It also allows Tesla to get feedback. Tesla can see what the public reaction is to the video leak to gauge if there is something they need to tweak or fix. We saw that with Smart Summon where Tesla pulled the first version back when first reactions were negative.

 

[mikes_fsd]

They did not stop their camera vision work because of lidar.

Neither would Tesla, please look up the definition of a crutch.
When you give an engineer an easier out, they will take, that is just human nature.

 

[drtimhill]

it's overlooked because its not a big issue

As I made clear in my posts it’s not an issue today because so few cars have it. It is likely to become an issue if it becomes common on cars.

 

[drtimhill]

A lot less than 1000W.

The big spinning velodyne lidar with a range of 120 m consumes 60W. And the solid state lidar is less than 10W.

Velodyne LiDAR - Wikipedia

lol 1000W lidar isnt lidar .. it’s a death-ray!

 

[NHK X]

As I made clear in my posts it’s not an issue today because so few cars have it. It is likely to become an issue if it becomes common on cars.

It’s a great question, the few papers I read about it seemed to minimize the amount of interference, one out of Korea also seemed to think the effect was minimal but could make the system question the possible detection of an object.

Also sounds like there are various techniques to mitigate interference, ie pulse modulation to reduce impact.

That said, not sure if I saw anything that really tests what would happen if there were 10-40lidar units (or more) of various specs surrounding a small space (intersection). If solid state does become super cheap, it is crazy to think there could be multiple devices on each car and it is interesting to wonder how much noise that would introduce.

 

[Daniel in SD]

As I made clear in my posts it’s not an issue today because so few cars have it. It is likely to become an issue if it becomes common on cars.

Quick internet search says the pulse duration of a LIDAR laser is 6ns (6 billionths of a second!). For 300m range there would be 1us window (300m/(speed of light)) that the unit would be looking for each return pulse. Intuitively it seems like it would not be problem. Is there any research suggesting it would be?

 

[zecar]

Lidars seem to be in in the 10-30 watt range. Energy required is expected to go down substantially. But a commercial vehicle consuming even a kW should work financially.

Waymo could be using 500w total today, but energy use will go down as they refine hardware.

 

[drtimhill]

Quick internet search says the pulse duration of a LIDAR laser is 6ns (6 billionths of a second!). For 300m range there would be 1us window (300m/(speed of light)) that the unit would be looking for each return pulse. Intuitively it seems like it would not be problem. Is there any research suggesting it would be?

I can find very little discussion on this except several companies selling solutions .. which suggests it’s a problem.

you are right about an individual pulse but any sensor sends out a lot of pulses in a wide grid .. current state of the art seems to be about 16000 grid points and 100Hz rate, but that seems to be advancing quite fast. Multiply the rate by (say) 10 cars ... and it starts to sound problematic to me ... especially if you think about beat frequencies (when two lidar are pulsing in phase at the same rate)

 

[Snuffysasa]

But you're missing the bigger picture.
Tesla can have over a million FSD capable cars on the road, while perfecting camera vision.
Whereas, with lidar that is economically feasible they would ONLY be able to start from, say, model year 2021!
They would have missed out on a ton of data gathered, learning experience of scaling vision, because internally they would have the false hope "lidar is coming".

Just like Elon said, it would have been a crutch.
But even worse internally, engineers would take the easy way out - "with lidar we can get an accurate enough measurement, so we don't need to worry too much if this is off 10 or 20% of the time" etc etc.

uh... just because you plan to use Lidar in say 2021 doesn't mean you can't work on collecting data, learning experience of scaling vision, etc... this is a silly meaningless argument.

We are two days out from October 20th. What’re the odds we actually see a private tester upload a video of the rewrite in action?

I’m sure they have to sign NDA’s but will this stop them from anonymously posting?

I think our odds will go up when it goes into early access which will be sometime after the closed beta.

 

[powertoold]

If the rewrite is released to non-Tesla employees, a video will leak for sure, like it has with every major new feature.

Like with any Elon announcement, I still have my doubts for this Tuesday. If it is Tuesday, it'll be at 11:59pm lol.

 

[mikes_fsd]

lol 1000W lidar isnt lidar .. it’s a death-ray!

The person was talking about the consumption of the entire system...
see the post for reference.

I have asked this question a few time whenever I see LIDAR being mentioned. Please, someone tell me how much power a 360deg LIDAR that can see 300m away without interference consumes?

I have a a mini LIDAR that can only properly see 8m away with a 512x512 resolution and it consumes 12W.

Would Tesla be able to accelerate the switch to sustainable energy with a sensor suite that consumes maybe 10-20x more?

i would be willing to bet that the Wayne [I think OP meant Waymo here] sensor suite & processing must consume at least 1000W vs ~100W of Tesla’s FSD solution. I would imagine the aerodynamic drag cause my the large sensor horns would also greatly diminish efficiency on the highway.

So can someone confirm how much power these long-range LIDAR systems consumes? Treating point clouds from a LIDAR is extremely expensive and matching LIDAR data with RGB data is also very expensive computationnally. My 512x512 point cloud çan at time overwhelm a 6-core i7 with 2080ti GPU when trying to reconstruct in into a 3D object. You need neural networks to break that data down in a more useful way. LIDAR for FSD is a brute force approach and if you don’t figure out the RGB NN stuff you’re kinda screwed long term anyways.

 

[Kmartyn]

I think the rewrite will bring enhanced L2. Elon has suggested the Dojo will be needed to tune it, and that is a year away. This will happen about the time HW4 arrives. I believe at this time they will also enhance the radar. The number of points an automotive radar can see, has been dramatically enhanced in the last year. From coarse cloud, to vision radar. Tesla has already upgraded once, from Bosch to Continental. TI is one of the companies with vision radar development kits out to automotive customers. It would be foolish for Tesla not to take advantage of this. More detail at the same cost.
So if you have HW4 and new radar, it might also make sense to look at the cameras.