Waymo

[diplomat33]

Do we know, what’s Waymo’s plan? Are they going to stay in robotaxi service or are they going to sell their solution to OEM’s for regular cars for private use?

Waymo plans to do both, offer ride-hailing and also offer leases. But we don't have any specific timeline or road map yet.

Right now, Waymo is using the ride-hailing in Phoenix to work out the issues of a ride-hailing service. They are focused on improving the FSD capabilities, making the validation/deployment more efficient and improving the customer service/business side. They want to get to a point where they can efficiently just "copy and paste" the ride-hailing model in Phoenix to other cities. And they plan to use the 5th generation to scale up. But Krafcik has also mentioned that Waymo plans to offer leases for robotaxis in the areas where they are also offering ride-hailing. Customers can lease their own private Waymo robotaxi for a year. They can use it like their private car and Waymo will take care of all the maintenance and support. So the lease payment will be all inclusive. After a year, the robotaxi will go back into the ride-hailing network. This info comes from the Krafcik Autonocast interview from Jan 2020 and the recent Lex interview with Dolgov.

 

[ItsNotAboutTheMoney]

If there was truly no time to react, that's one thing. But AV's have sensors that can see hundreds of meters away to hopefully give the AV time to react in most cases. The AV should still try to avoid the collision if it is possible and safe to do so. It should detect the pedestrian's path with enough time to react, detect that the path is on a possible collision course, detect that they are distract and try to slow down if possible to avoid a collision. Waymo, Zoox, Cruise all have vision capable of detecting distracted pedestrians for this very purpose.

It doesn't need to care about distraction. A pedestrian can do anything they want, suddenly. All that matters is liability.

If AVs are passive around pedestrians, the pedestrians will give them no respect, and make them stop. They have to be aggressive and not care about hitting them when it's the pedestrian's fault.

 

[Lubestaff]

It doesn't need to care about distraction. A pedestrian can do anything they want, suddenly. All that matters is liability.

If AVs are passive around pedestrians, the pedestrians will give them no respect, and make them stop. They have to be aggressive and not care about hitting them when it's the pedestrian's fault.

Defeats the purpose of AVs, no? AVs should surpass humans in reaction time, thus safety, and avoid such accidents as possible, because even a person neglectfully stepping into your car's path doesn't deserve death.

And yes, I'm of the very steadfast view that AVs should avoid the accident even if pedestrians decide to assert dominance over Autonomous Vehicles by suppressing all fear and jumping onto their paths.

 

[Matias]

It doesn't need to care about distraction. A pedestrian can do anything they want, suddenly. All that matters is liability.

If AVs are passive around pedestrians, the pedestrians will give them no respect, and make them stop. They have to be aggressive and not care about hitting them when it's the pedestrian's fault.

Most human drivers try not to hit pedestrians even when it’s the pedestrians fault and yet pedestrians don’t jump in front of the car, because they care about their own lives.

 

[diplomat33]

Since we were discussing Waymo's 5th generation hardware, I thought I would share this. Waymo put out a little teaser video back in April 2020 that has some really nice close-ups of the different sensors.

And here is the graphic that shows what each sensor is:

I hope Waymo really starts mass deployment of the 5th generation platform soon. We know the 5th generation is more capable than the 4th generation so I'd love to see some videos of rides on the 5th generation platform to compare it to what we are seeing now in Phoenix, to note the improvements.

 

[ItsNotAboutTheMoney]

Defeats the purpose of AVs, no? AVs should surpass humans in reaction time, thus safety, and avoid such accidents as possible, because even a person neglectfully stepping into your car's path doesn't deserve death.

And yes, I'm of the very steadfast view that AVs should avoid the accident even if pedestrians decide to assert dominance over Autonomous Vehicles by suppressing all fear and jumping onto their paths.

Almost all of the deaths and injuries would be prevented just by obeying the law and not being distracted. No blowing through red lights and crosswalks, no failure to see motorcycles and bicycles, no driving too close and so on.

That doesn't require detailed tracking of pedestrian movement just in case a pedestrian is an idiot. If a liable pedestrian steps out in front of the car, the car can react, but you want pedestrians to know that the car is not actively trying to protect them from themselves so if they step out in front of the car, it could hurt, and they aren't going to get an insurance claim. Also, by not tracking potential idiots, reaction is enough. The more the car tracks the idiots, the earlier it _must_ react, which will give more confidence to the idiots.

You also want pedestrians to know that when they step out in front of the car forcing braking, the video will be saved and will be sent to the police as necessary. Facial recognition technology could be very useful for identifying and charging repeat offenders.

 

[Lubestaff]

Almost all of the deaths and injuries would be prevented just by obeying the law and not being distracted. No blowing through red lights and crosswalks, no failure to see motorcycles and bicycles, no driving too close and so on..

Almost. And humans don't necessarily choose to be distracted. You realize that, currently, the penalty of distraction in this context is often death or severe injury, right? People don't just decide to accept death; distraction happens, and safety features should account for that. In fact, this is should partly be a motivator for Autonomous Driving: to compensate for natural human distraction.

That doesn't require detailed tracking of pedestrian movement just in case a pedestrian is an idiot. If a liable pedestrian steps out in front of the car, the car can react to the appearance of the pedestrian, but you want pedestrians to know that the car is not actively trying to protect them from themselves so if they step out in front of the car, it could hurt, and they aren't going to get an insurance claim.

Don't worry, that point will remain clear nonetheless as long as the laws of physics stand. Even with a hypothetical perfect reaction time, a vehicle can't immediately brake to a halt, and were that possible, it wouldn't be without consequence to the passengers. With that accounted for, stepping in front of an AV could still very evidently hurt. Otherwise, preventable accidents should be prevented, as is a major point of autonomous vehicles.

 

[FloridaJohn]

I'm pretty sure you would have a hard time selling a car that will most likely hurt or kill pedestrians, but otherwise is the safest vehicle on the road.

Unfortunately, AV are going to be bullied on the road, and they will acquiesce. That will be the cost of following the rules of the road. As more vehicles become automated, this will become less of a problem.

 

[ItsNotAboutTheMoney]

Almost. And humans don't necessarily choose to be distracted. You realize that, currently, the penalty of distraction in this context is often death or severe injury, right? People don't just decide to accept death; distraction happens, and safety features should account for that. In fact, this is should partly be a motivator for Autonomous Driving: to compensate for natural human distraction.

People drive more dangerously when they feel safer. People will walk more dangerously when they feel safer.

The key safety benefit of AVs isn't to protect people from themselves, it's to protect other people from them. The same reason why drivers are required to pass a driving test.

If AVs just eliminates human error, the statistical improvement would be so great, the remaining injuries and deaths would be too small to care about.

Don't worry, that point will remain clear nonetheless as long as the laws of physics stand. Even with a hypothetical perfect reaction time, a vehicle can't immediately brake to a halt, and were that possible, it wouldn't be without consequence to the passengers. With that accounted for, stepping in front of an AV could still very evidently hurt. Otherwise, preventable accidents should be prevented, as is a major point of autonomous vehicles.

The AV doesn't necessarily have to stop to allow the idiots to walk out in front of it. It just needs to brake.

Earlier braking means it's safer to walk out in front of an AV, and if that's the case, then people will do it. If people do it, it will slow AVs down, which will make riding in AVs worse, and more expensive (time is money) which will make them less appealing.

If you want to maximize AV rideshare, don't program AVs to protect the idiot pedestrians, focus on giving a safe, smooth, fast ride to the riders.

 

[Lubestaff]

If AVs just eliminates human error, the statistical improvement would be so great, the remaining injuries and deaths would be too small to care about..

This'll be a hard sell. The point that humans get naturally distracted still stands, as does my view that safe driving, and thus the goal of AVs, must account for their safety as well. I do imagine AVs getting pushed around more than normal vehicles, but in favor of safety, so be it; this also should be possible to mitigate, as you proposed yourself a few comments ago.

If you want to maximize AV rideshare, don't program AVs to protect the idiot pedestrians, focus on giving a safe, smooth, fast ride to the riders.

Unwarranted aggressive driving and preventable pedestrian accidents will make it a harder sell than submission to a certain degree of pedestrian abuse, in my opinion. You realize that the hypothetical scenarios that you're proposing of disregarding negligent pedestrians, even when the accident might be preventable (without detriment to the passengers), imply a higher incidence of accidents, right? This goes strictly against safe, smooth, fast rides.

EDIT:

If AVs just eliminates human error, the statistical improvement would be so great, the remaining injuries and deaths would be too small to care about..

Let me add the additional point that, alternatively, with proper avoidance of pedestrian accidents, the result of the occasional abuse (say, a slowdown or halt for a pedestrian who abused AV safety to improperly cross the road) could be too small to be significant. I doubt any preventable deaths would effectively neglected by the public or the riders. And evidently, I would personally detest lack of accounting for idiot pedestrians, among which abusers can likely not be distinguished from people who just happened to be distracted.

 

[mars_or_bust]

Since we were discussing Waymo's 5th generation hardware, I thought I would share this. Waymo put out a little teaser video back in April 2020 that has some really nice close-ups of the different sensors.

And here is the graphic that shows what each sensor is:

I hope Waymo really starts mass deployment of the 5th generation platform soon. We know the 5th generation is more capable than the 4th generation so I'd love to see some videos of rides on the 5th generation platform to compare it to what we are seeing now in Phoenix, to note the improvements.

Cool if you want to haunt ghosts, really ridiculous otherwise

 

[mars_or_bust]

This'll be a hard sell. The point that humans get naturally distracted still stands, as does my view that safe driving, and thus the goal of AVs, must account for their safety as well. I do imagine AVs getting pushed around more than normal vehicles, but in favor of safety, so be it; this also should be possible to mitigate, as you proposed yourself a few comments ago.



Unwarranted aggressive driving and preventable pedestrian accidents will make it a harder sell than submission to a certain degree of pedestrian abuse, in my opinion. You realize that the hypothetical scenarios that you're proposing of disregarding negligent pedestrians, even when the accident might be preventable (without detriment to the passengers), imply a higher incidence of accidents, right? This goes strictly against safe, smooth, fast rides.

EDIT:


Let me add the additional point that, alternatively, with proper avoidance of pedestrian accidents, the result of the occasional abuse (say, a slowdown or halt for a pedestrian who abused AV safety to improperly cross the road) could be too small to be significant. I doubt any preventable deaths would effectively neglected by the public or the riders. And evidently, I would personally detest lack of accounting for idiot pedestrians, among which abusers can likely not be distinguished from people who just happened to be distracted.

The abuse by pedestrians will be a nothing-burger, because the population will gradually become ware that cars film EVERYTHING. One can easily imagine penalties for abusive pedestrians, based on video recordings from the self-driving car.

 

[Lubestaff]

The abuse by pedestrians will be a nothing-burger, because the population will gradually become ware that cars film EVERYTHING. One can easily imagine penalties for abusive pedestrians, based on video recordings from the self-driving car.

Indeed. User @ItsNotAboutTheMoney admitted this himself, and I'm confident the issue of pedestrian abuse will be made insignificant without sacrificing safety; it might even be insignificant to begin with.

 

[EVNow]

If AVs are passive around pedestrians, the pedestrians will give them no respect, and make them stop. They have to be aggressive and not care about hitting them when it's the pedestrian's fault.

Wow - reminds me of a presidential debate when the audience screamed "let them die" when asked what to do with poor who are sick.

On the lighter side - just put Uber sign on the car. People will be afraid to cross the road even at the crosswalk, on green light.

 

[EVNow]

They can use it like their private car and Waymo will take care of all the maintenance and support. So the lease payment will be all inclusive.

Can't be cheap - given their hardware price.

But then, I don't think Tesla FSD will be any cheaper when it actually comes. Just that Tesla will be making a lot of margin than Waymo.

 

[mikes_fsd]

Yes. Based on a fact that there are paying customers on the backseats of Waymo’s driverless cars at this very moment.

By that example there are roughly 35% (take rate is from Troy's spreasheets) of all Tesla sales that buy FSD option anywhere from $5k to now $10k a pop.

We hit 1 millionth car in March 2020 (9 months ago) Tesla Builds Its One Millionth EV, and It's a Model Y
There are ~165k cars with AP1 or no AP hardware at all! (source Tesla Vehicle Deliveries and Autopilot Mileage Statistics - Lex Fridman )
Based on deliveries in Q2, Q3 and TroyTeslike projection for Q4 ... minus the 165k cars pre-Tesla in-house AP you have left about 1.3 million cars.
So, 1.3 million car x 0.35 take rate = 455k cars where Tesla made money (collected revenue for) from "paying customers" for the "Full Self Driving" at this very moment!
And remember, that ppl can buy the FSD option at ANY time after they take delivery.
So, the take rate (which is at the time of purchase) is on the low end.

Just to make sure this argument is not used again.
455k cars with FSD --- lets take an averaged price for FSD of $6000 / car
That is $2,730,000,000 ($2.73 billion) in revenue from FSD sales alone.

I would love to know how long it would take Waymo to reach $2.73 billion in revenue with taxi business model with their tiny little fleet.

 

[Doggydogworld]

People drive more dangerously when they feel safer. People will walk more dangerously when they feel safer.

The key safety benefit of AVs isn't to protect people from themselves, it's to protect other people from them. The same reason why drivers are required to pass a driving test.

If AVs just eliminates human error, the statistical improvement would be so great, the remaining injuries and deaths would be too small to care about.



The AV doesn't necessarily have to stop to allow the idiots to walk out in front of it. It just needs to brake.

Earlier braking means it's safer to walk out in front of an AV, and if that's the case, then people will do it. If people do it, it will slow AVs down, which will make riding in AVs worse, and more expensive (time is money) which will make them less appealing.

If you want to maximize AV rideshare, don't program AVs to protect the idiot pedestrians, focus on giving a safe, smooth, fast ride to the riders.

Pedestrians and self-driving cars will co-exist in perfect harmony:

 

[diplomat33]

Can't be cheap - given their hardware price.

We don't know what the hardware actually costs. I know Tesla fans like to say that the hardware costs around $100k but that is pure speculation and based on outdated numbers. Waymo themselves say that 5th gen hardware is half the cost of 4th gen hardware. So I doubt that the 5th gen hardware will be as expensive as Tesla fans think.

Having said that, I don't expect the robotaxi leases to be cheap since Krafcik says the leases will be a "premium service". But again it will be a real driverless robotaxi in your driveway, with all maintenance costs covered. So you are definitely getting a lot.

Tesla is charging $10k for much cheaper hardware and for something that is not real FSD yet.

I would love to know how long it would take Waymo to reach $2.73 billion in revenue with taxi business model with their tiny little fleet.

Waymo is not planning to stay with a "tiny little fleet" forever. They plan to scale up to a much larger fleet! With a larger fleet of robotaxis, Waymo could make that revenue per year!

In fact, doing a quick calculation. From a Waymo ride video, a 14 minute ride costs $6.76. If a robotaxi drove non stop 24/7, 365 days a year, that would be a maximum revenue of $262,800 per robotaxi per year. A fleet of 100k Waymo robotaxis could make up to $26.2B a year. Obviously, not every robotaxi will be taking rides non stop 24/7 so the actual revenue would be less than that. Waymo would just need a fleet of 100k robotaxis operating only 10% of the time to make your revenue of $2.73B per year.

 

[ItsNotAboutTheMoney]

Let me add the additional point that, alternatively, with proper avoidance of pedestrian accidents, the result of the occasional abuse (say, a slowdown or halt for a pedestrian who abused AV safety to improperly cross the road) could be too small to be significant. I doubt any preventable deaths would effectively neglected by the public or the riders. And evidently, I would personally detest lack of accounting for idiot pedestrians, among which abusers can likely not be distinguished from people who just happened to be distracted.

There's around 40,000 preventable deaths per year and many more injuries in the USA because so many people don't care about themselves or other road users while driving. I doubt they're going to start caring much now about people who are at fault, when the number will be so much smaller.

 

[Doggydogworld]

Can't be cheap - given their hardware price.

Their h/w doesn't cost much. Cameras are dirt cheap, radars a little more. Waymo said it was 150k extra in 2012, but that included a 75k Velodyne HDL-64e plus perimeter lidars that cost much less but were still expensive. They said they cut costs 90% with Gen 4 and another 50% with Gen 5. So they're in the 5-10k range. That's 1-2 cents per operating mile assuming a 500k mile life.

Waymo famously "ordered" 62k Pacificas and 20k iPaces a few years ago. At some point they'll go to a cost-optimized custom vehicle, like the Cruise and Zoox designs. When we discussed this earlier I forgot that Magna owns part of Waymo and also builds the iPace. Interesting....