Waymo

[vickh]

they're live in AZ! But our roads/weather are mostly clear/grid based. Wonder how they'll do in the "real world"

On related note I'm thinking of also buying the Pacifica PHEV they use but scared by reliability issues, esp coming from Toyota/Tesla. I'm hoping it'll have consumer AP one day, like the Tesla

 

[Bitdepth]

I think I found a new Waymo video from a recent drive. It was posted just a few hours ago on reddit. It is a short drive but it is completely driverless and at night.

4 Minute Full Video in Waymo One - No Driver; Short Trip : waymo

That's fake news. Lol. On a serious note this is some impressive showing by Waymo.

People are starting to share their experiences.
https://twitter.com/jjricks_/status/1314532703937589248

 

[diplomat33]

That's fake news. Lol. On a serious note this is some impressive showing by Waymo.

People are starting to share their experiences.
https://twitter.com/jjricks_/status/1314532703937589248

Yeah, I think we will see more and more videos now that the NDA has been lifted.

 

[powertoold]

That's fake news. Lol.

Why is it fake news? Did I miss something?

 

[diplomat33]

Why is it fake news? Did I miss something?

Yeah, you missed the joke, hence the "lol". Bitdepth was trying to make a joke that the Waymo critics will call it fake news because they don't believe that Waymo has actually gone driverless.

 

[diplomat33]

Waymo Chief Operating Officer, Tekedra Mawakana, was interviewed by Tech Crunch on Oct 6.

Here is the full video interview:
Videos – TechCrunch

It is behind a pay wall. You need a membership to TC's extra videos section. I went ahead and paid to access the video so I can share the highlights with you.

Here are my biggest take-aways from the interview.

- Waymo is focused on commercialization of 3 applications of autonomous driving: ride-hailing (Waymo One), trucking and local deliveries (Waymo Via).

-State regulations (governor executive order) in AZ allow both testing and deployment of autonomous vehicles as a service which shaped Waymo's decision to launch ride-hailing in Phoenix, AZ.

-Waymo develops autonomous driving through closed course testing, simulation and then real world driving.

- Commercialization for autonomous ride-hailing is a step by step process, not a "flip of the switch". First, Waymo introduced the ride-hailing service to early riders (private beta) who signed up and provided feedback. Then, Waymo opened up the ride-hailing service to the general public. Waymo started rides for the general public in Dec 2018 with Waymo One. Waymo spend 2019 growing that service. Waymo One had 1,500 riders before covid hit.

- Out of abundance of caution, Waymo suspended Waymo One in March because of covid. In May, Waymo started bringing the service back up. Now starting the process of rider restart.

- Waymo will provide a premium driver experience. Waymo will handle everything from managing, cleaning and operating the robotaxi fleet. The customer just pays to use the robotaxi. Basically, autonomous driving as a service.

- Waymo mission is to build the world's first and leading autonomous vehicle operations platform.

- Waymo has option to buy 20,000 I-Pace's in order to have capacity to scale up to multiple cities to fulfill long term goals. Waymo does not build cars. Waymo only builds the "Waymo Driver" (Waymo's name for their FSD stack).

- Waymo uses 3rd party contractor, TransDev, to supply the safety drivers. Ultimately, the goal is driverless. Waymo will use TransDev globally for their testing needs but go driverless locally as they expand.

- Waymo intends to operate the fleet in-house. Waymo will not oursource operations of the fleet.

- Long term, Waymo plans to scale globally.

- In the app, you can have the car honk at pick-up to help you find it if you have vision problems. In terms of wheel chair accessibility that is dependent on automakers who make the base vehicles. Waymo encourages OEM's to provide wheelchair accessible vehicles.

 

[diplomat33]

This article from two days ago has some good info on the roll-out of driverless rides:

First, the company will allow the "thousands" of Waymo One riders, who also had to be preapproved to be in the program, to ride in a fully autonomous car. They can bring guests, too.

Once the existing Waymo One customers get the opportunity to ride in an autonomous vehicle, which could take weeks, Waymo offers its app to anyone from either Google Play or the Apple App Store.

Waymo has 600 vehicles in its autonomous fleet, and 300-400 of them are in Arizona, Krafcik said.

They operate in a “geofenced” area covering parts of Chandler, Mesa, Tempe, Gilbert and Ahwatukee.

The fully driverless rides only operate in a portion of the Waymo service area, which the company has previously said is about 50 square miles, or about the size of San Francisco. The entire area where Waymos operate in Arizona is about 100 square miles, Krafcik said.

After giving people the opportunity to experience the fully autonomous rides, Waymo will begin to reintroduce safety drivers in the vehicles, which allows the vehicles to operate in the larger area than those with no safety driver.

"But for, I don't know, we'll see, the next several weeks, perhaps a month or more, every ride, 100% of rides with Waymo One, will be fully driverless," Krafcik said.

Ready to ride in a driverless car? Waymo vans going public

Interesting tidbit from the video: Waymo cars can honk the horn when it detects another car doing something bad but the software is apparently programmed to use the honk sparingly so as not to be annoying.

 

[Bitdepth]

Why is it fake news? Did I miss something?

Just a joke. But here are some more cool waymo videos.

Pick up

Boring driving

 

[NHK X]

Hopefully this will end this ridiculous beef you two have.
It sounds like we're finally going to get a large enough sample size to see how well their vehicles really work. It's too bad Arizona doesn't seem to have public reporting of autonomous miles and accidents.

With ghost drivers watching remotely and able to take control when needed, how will one know how well it’s doing? How can you know whether or not they had to step in?

 

[diplomat33]

With ghost drivers watching remotely and able to take control when needed, how will one know how well it’s doing? How can you know whether or not they had to step in?

I suspect we would see the FSD system disengage (indicated on the screen) if the remote operator had to intervene. And if customers write reports about the car "getting stuck" that would probably also be a good indication that a remote operator had to intervene.

 

[diplomat33]

Good podcast interview with Waymo's CEO John Krafick from Jan 2020:

John Krafcik of Waymo — Autonocast

Lots of good info.

Waymo is planning to expand ride-hailing to LA or SF next.

 

[zecar]

I think de congesting and optimizing public transportation in cities is where this tech will really shine. Private cars freely doing FSD in cities will never work. Free market FSD would lead to mass congestion in urban areas.

Where I see this headed is banning privately controlled cars from city centers during busy time. Private cars with FSD could bid-in to use excess street capacity. But the system would always be optimized towards low cost shared transportation. So the CEO may still get her private car to work, but it will be very expensive.

Going to work on robo-uber might look like this:

Lowest Cost - shared ride (van) pickup and dropoff at the corner bus stop. Time given on phone app.
Medium Cost - Door to door shared
High Cost - Door to door private

Medium and high cost rides are price dynamic depending on demand. Vehicles are specialized. Vehicles range in size from Model Y to smaller bus.

 

[NHK X]

I suspect we would see the FSD system disengage (indicated on the screen) if the remote operator had to intervene. And if customers write reports about the car "getting stuck" that would probably also be a good indication that a remote operator had to intervene.

It’s more a question of will they be transparent.

 

[diplomat33]

It’s more a question of will they be transparent.

Now that it is open to the general public, they can't really stop someone from reporting a problem.

 

[NHK X]

Now that it is open to the general public, they can't really stop someone from reporting a problem.

You aren’t really getting the point.

With a ghost driver they could take over at any point, have nothing show up on the display, and the public tester would have no idea that anyone ever had to take control. Of course if the car obviously got stuck and stayed there for a couple minutes they would know. But a momentary pause, and takeover from the ghost driver the rider would have no idea. To them it would still seem that a completely “driverless” car made the trip, when in reality it’s still using a safety driver.

 

[Daniel in SD]

With ghost drivers watching remotely and able to take control when needed, how will one know how well it’s doing? How can you know whether or not they had to step in?

Hopefully there won't be a way to tell but in practice I bet riders will be able to tell. It seems like it will take some time for a human to evaluate a situation and give the car instructions. Hopefully their claimed 10k+ mile disengagement rate indicates that this happens quite rarely!

But computers can’t always come up with every strange real-world scenario or react to real-time construction detours. This is where Waymo’s fleet response team come into play. If the vehicle encounters a complex driving scenario that it struggles to interpret, it automatically calls in the problem to the response team to weigh in with a solution, which is then shared with the rest of the fleet so Waymo’s vehicles can avoid the area if necessary. Those remote operators are based both in Phoenix and Austin, Texas, but they have no direct control over the vehicle’s operations, Perez said; they just serve as an extra set of eyes for difficult-to-navigate scenarios. “The car might see cones up ahead and could ask for context,” she said. “Should I move to another lane? Should I turn ahead? Should I reroute myself?”

A day in the life of a Waymo self-driving taxi

 

[zecar]

Human intervention is not a binary event. The drive plan is probability based. So a human operator simply joining the car and looking at the sensor raises the confidence factor of the driving plan. The car may continue its driving plan because a human is watching. That event is human intervention of a sort, but it's not a human implementing a plan or operating the car.

Humans were always going to be necessary, regardless of the pure implementation usually assumed for Tesla. The only issue with humans is cost. The need for humans will steadily decrease. So at some future point 100 humans in a call center in Mumbai can support the worldwide fleet of autonomous cars. A more generalized AI would then theoretically replace humans completely.

 

[diplomat33]

Krafcik reveals something, that I find interesting, in the autonocast interview around the 38 mn mark:
John Krafcik of Waymo — Autonocast

He says Waymo is focused on improving FSD right now but that Waymo is planning to do "personal car ownership" with the Waymo Driver, using a subscription model. He prefers subscription model over outright buying because with technology becoming obsolete so quickly, he does not think that owning and maintaining a 25 year old robotaxi will make sense. The automakers will build the cars. It might be the current Chrysler Pacifica minican, the Jaguar I-Pace or a new car. Waymo will retrofit them with the FSD hardware and software. The customer will then be able to lease the robotaxi on a yearly subscription model. Once they cancel the subscription, the robotaxi is returned and added to the ride-hailing fleet. But this would allow people to have their own driverless robotaxi at their house to use as they need. Of course, the cars might have an option for manual driving too if owners wanted to manually drive in some instances.

I did not know that Waymo was interested in personal car ownership for FSD. So, they are not going to do just ride-hailing. I suspect the ride-hailing is a good way to test the FSD before they do personal car ownership. My guess is that eventually the personal car ownership with the subscription model will be the main money maker for Waymo and the ride-hailing will be the secondary source of income.

I find this subscription model fascinating and thought provoking. I love the idea of having your own truly driverless robotaxi that could go off to pick your kids from school or take you to work while you sit back as a passenger. It will be interesting to see what the cost will be. It would need to be competitive with other options of owning a car. I think that single people living a city where car ownership is difficult, might still prefer to use a ride-hailing service. But I suspect leasing a robotaxi might make a lot of sense for families. They will be able to replace 2-3 cars with a single robotaxi. The subscription cost for one robotaxi will probably be much less than the cost of owning 2-3 vehicles. And of course, folks who have trouble driving will also benefit from owning their own robotaxi. Entrepreneurs could also lease robotaxis to start their own mini ride-hailing service to make money.

Krafick also explains in the interview why Waymo discontinued the firefly even though it was such a cool driverless robotaxi.

He mentions two automotive industry standards that regulate how cars are built: the Federal Motor Vehicle Safety Standard 208 (FMVSS 208) and Federal Motor Vehicle Safety Standard (FMVSS 500). FMVSS 208 regulates current cars on the road and does not allow you to remove the steering wheel or pedals. FMVSS 500 regulates "neighborhood electric vehicles" and does allow you to remove the steering wheel and pedals but limits the speed to 25 mph. So firefly fell under FMVSS 500 thus they were able to remove the steering wheel and pedals. He also says that Waymo would love to remove the steering wheel and pedals completely.

 

[Bladerskb]

You aren’t really getting the point.

With a ghost driver they could take over at any point, have nothing show up on the display, and the public tester would have no idea that anyone ever had to take control. Of course if the car obviously got stuck and stayed there for a couple minutes they would know. But a momentary pause, and takeover from the ghost driver the rider would have no idea. To them it would still seem that a completely “driverless” car made the trip, when in reality it’s still using a safety driver.

there are no ghost riders. There are no remote safety drivers. You can’t joystick the car, steer, accelerate or brake remotely.

you can only plot tracjectory, reroutes or confirm an uncertainty the car has.

 

[NHK X]

there are no ghost riders. There are no remote safety drivers. You can’t joystick the car, steer, accelerate or brake remotely.

you can only plot tracjectory, reroutes or confirm an uncertainty the car has.

How sure are you? And state how you are so sure.

What happens when the “driverless” car gets stuck in a situation? Does it just stop and tell the rider to get out?