Tesla

[NomisR]

Source?

Obviously computing takes some amount of power, but I wouldn't expect it to be very significant when compared to what's needed to propel a 3500+ lb vehicle.

This isn't the exact article I read but still related

http://energyfuse.org/autonomous-driving-takes-power-will-that-matter-for-electrification/

[NomisR]

Your seriously saying you would rather trust your life to cloud computing then on board computing?

Did I say I trust it?  Like a lot of functions today, you can have both being done, split off part of the computing function to the "cloud" so to reduce computing usage.  And of course you still need some sort of computing power on board as well.  And i'm talking about fully autonomous driving that manufacturers are testing, not the pretend to be autonomous driving with cruise control + lane keep sensing assist + distance assist thing that  Tesla has. IOW, no steering wheel.

[BimmerM3]

This isn't the exact article I read but still related

http://energyfuse.org/autonomous-driving-takes-power-will-that-matter-for-electrification/

Thanks. I'm not too concerned about the 5-10% lower-end estimate, but the one-third estimate isn't great. I haven't had time to dive into the details of where those numbers came yet.

That said, by the time that real AV tech becomes available, I'd expect that both battery improvements and computing power improvements will make this a lot less of an issue.

Did I say I trust it?  Like a lot of functions today, you can have both being done, split off part of the computing function to the "cloud" so to reduce computing usage.  And of course you still need some sort of computing power on board as well.  And i'm talking about fully autonomous driving that manufacturers are testing, not the pretend to be autonomous driving with cruise control + lane keep sensing assist + distance assist thing that  Tesla has. IOW, no steering wheel.

Even with improved speed/latency from 5g, I don't think wireless service will be fast/reliable enough to handle full autonomy. Those vehicles will need to make near instantaneous calculations and decisions. Even if 5G is fast enough in ideal conditions, what happens if you lose signal in a car with no steering wheel?

[NomisR]

Thanks. I'm not too concerned about the 5-10% lower-end estimate, but the one-third estimate isn't great. I haven't had time to dive into the details of where those numbers came yet.

That said, by the time that real AV tech becomes available, I'd expect that both battery improvements and computing power improvements will make this a lot less of an issue.

Even with improved speed/latency from 5g, I don't think wireless service will be fast/reliable enough to handle full autonomy. Those vehicles will need to make near instantaneous calculations and decisions. Even if 5G is fast enough in ideal conditions, what happens if you lose signal in a car with no steering wheel?

Well theoretically, 5G speed and latency should be similar to that of land lines, we'll see if it holds up to the hype of course. You would of course need redundant backups with calculations also capable of being done on the vehicle when connection drops, based on current technology other than the actual 5G, it's not that far of a stretch.  The fully automated technology would likely be hitting luxury vehicles by the end of the next decade with the current autopilot type technology in regular cars at that time. 

[r0tor]

It's fairly obvious that small computers have 75,000W power supplies

[r0tor]

Google's cloud services went down last night locking people out of their houses and disabling cameras... Can't imagine if your car would suddenly be disabled because a server went down hundreds of miles away

[giant_mtb]

IoT FTW!

[12,000 RPM]

It's fairly obvious that small computers have 75,000W power supplies

Autonomous computing will be local and have a negligible impact on vehicle power consumption. If there is one place the computing industry is making huge progress... it's efficiency, thanks to mobile.

[MrH]

R0tor, can we get some hot takes on the stock price please?

[Laconian]

Cloud computing can be "fast" because you can scale out horizontally (as in, you can share workloads across clusters), you can do preprocessing, and you can cache data inside huge, cheap storage pools. A rackmount computer's single core performance will not be substantially faster than what's in a desktop PC, though.

An onboard computer with an OK GPU would be so much better than a cloudy computer interconnected by hundreds of miles of fail-prone networking links.

[BimmerM3]

Cloud computing can be "fast" because you can scale out horizontally (as in, you can share workloads across clusters), you can do preprocessing, and you can cache data inside huge, cheap storage pools. A rackmount computer's single core performance will not be substantially faster than what's in a desktop PC, though.

An onboard computer with an OK GPU would be so much better than a cloudy computer interconnected by hundreds of miles of fail-prone networking links.

+1.

Though I think Nomis was concerned with power usage more than processing speeds. Offload the heavy lifting to machines that aren't running off of a battery.

Even with on-board backups that can handle all of the processing but will just use more power, it seems like an over-complication to me. I think that by the time we have real, consumer-ready AVs, computer hardware and battery improvements will make that part of the equation pretty trivial.

[SJ_GTI]

I am curious to see if they can stay in business long enough to get the model to market.

[Soup DeVille]

We don't even trust critical industrial processes to a single computer, we run shit in parallel processors and on ring networks if losing control creates risk above a certain level.

Autonomous cars are way above that level.

[CaminoRacer]

We don't even trust critical industrial processes to a single computer, we run shit in parallel processors and on ring networks if losing control creates risk above a certain level.

Autonomous cars are way above that level.

Could still have onboard backups though, right? No need to use the cloud

[12,000 RPM]

+1.

Though I think Nomis was concerned with power usage more than processing speeds. Offload the heavy lifting to machines that aren't running off of a battery.

Even with on-board backups that can handle all of the processing but will just use more power, it seems like an over-complication to me. I think that by the time we have real, consumer-ready AVs, computer hardware and battery improvements will make that part of the equation pretty trivial.

I'm feeling like a combo of local and cloud processing could optimize the loads on both sides as well. Let the cloud do the brunt of the machine learning and just make the car focus on applying that data to whatever situations it encounters.

[veeman]

We don't even trust critical industrial processes to a single computer, we run shit in parallel processors and on ring networks if losing control creates risk above a certain level.

Autonomous cars are way above that level.

Like in the Boeing 737 Max

[Soup DeVille]

Could still have onboard backups though, right? No need to use the cloud

Absolutely.

Of course, autonomous vehicles should absolutely communicate with each other, possibly on their own dedicated band; but vehicle decisions still need to be made by the onboard CPUs.

[Soup DeVille]

Like in the Boeing 737 Max

Well, if the programming is bad, it doesn't matter how redundant your hardware is.

[r0tor]

I'm feeling like a combo of local and cloud processing could optimize the loads on both sides as well. Let the cloud do the brunt of the machine learning and just make the car focus on applying that data to whatever situations it encounters.

I believe in Tesla's case exemption events (like drivers needing to intervene) are reported back to the cloud, analyzed, and then periodic updates are sent back to the fleet.

[GoCougs]

Well, if the programming is bad, it doesn't matter how redundant your hardware is.

Not true - it is common practice to build hardware to mitigate or prevent bad programming.

[BimmerM3]

I'm feeling like a combo of local and cloud processing could optimize the loads on both sides as well. Let the cloud do the brunt of the machine learning and just make the car focus on applying that data to whatever situations it encounters.

Oh, that's definitely true. But I think the semi-autonomous systems in the market today already do that. There's no need for the machine learning piece to be in real-time.

[Soup DeVille]

Not true - it is common practice to build hardware to mitigate or prevent bad programming.

Yes, of course. I was talking specifically about CPU redunancy.

[MrH]

I believe in Tesla's case exemption events (like drivers needing to intervene) are reported back to the cloud, analyzed, and then periodic updates are sent back to the fleet.

According to Tesla.  I don't believe that's happening at any sort of meaningful volume.

[AutobahnSHO]

ha a Tesla S pulled up next to us at a light the other night. When it went green that car was GONE and we were in a 35mph zone and a block later they were turning left. But my wife was shocked how quick they took off.

That would be fun!!!

[CaminoRacer]

ha a Tesla S pulled up next to us at a light the other night. When it went green that car was GONE and we were in a 35mph zone and a block later they were turning left. But my wife was shocked how quick they took off.

That would be fun!!!

I've been next to a Tesla a few times at stoplights and always avoid racing them. Don't want to be embarrassed

[Laconian]

It feels bad when you're rowing through the gears and your engine is making a lot of noise... then the yuppie in the Model S looks up from their phone, realizes there's room to accelerate, and their car just zaps forward silently annihilates you with hushed tire roar.

[Xer0]

Tesla's are very annoying when you're biking into work.  Give me V8 Stangs/Camaros any day cause you can hear those suckers a mile away but Tesla's just show up with no warning.  Priuses are similar but they are so slow it doesn't mater, I've literally out accelerated a few 

[NomisR]

I'm feeling like a combo of local and cloud processing could optimize the loads on both sides as well. Let the cloud do the brunt of the machine learning and just make the car focus on applying that data to whatever situations it encounters.

This is probably the most likely case along with inter-car communications that would be done for additional efficiency in decision making done with a network. 

[Galaxy]

http://www.youtube.com/watch?v=QlJ7lPdazyw

Yikes!

Basically a fender bender on a Model 3 takes 3 Month, and $10,000 to fix.

When they start talking about Tesla in general skip to 10:34 to get back to the repairs.

[Laconian]

This is probably the most likely case along with inter-car communications that would be done for additional efficiency in decision making done with a network. 

I think a mesh network would make a lot of sense. General purpose cellular data is easily overtaxed in areas facing high congestion. A mesh network, built for relaying automotive data, isn't going to be overloaded by adjacent bus riders streaming Netflix on their phones.