Robocars 2019 Year In Review


Here is a summary of the Robocar stories from 2019 that were the most significant. It was actually not a year of very big change. Waymo is still the distant leader, in spite of having slipped a bit on their goals. I talk about the trough of the hype cycle and the challenges going ahead for the 2020s. If you skipped most of my coverage in the year, these are the selected ones to read.

Read the year in review at Robocars 2019 in review


The '2019 year is review' might as well be the same as the '2018 year in review' or '2017 year in review'.

Autonomous vehicles have been among the most overhyped technologies ever. The Economist wrote about how the hucksters failed to deliver.

Actually, I think that while some companies hyped up things, the leading companies did not hype as much. The industry "failed to deliver" on the statements of those making the strongest hype. This is in fact the cause of the Garntner hype cycle. The peak of the cycle is driven by those who are overhyping, by definition almost, not the people who are hard at work building. (Though it can be executives of companies that are hard at work building.) So of course reality doesn't match that, which is disappointing. Then the hype scales back.

That is a lot of backpedaling and hedging.

Really the only important measure is the performance of the leading competitor.
In this case it's Waymo by a huge margin.
Brad is correct that they haven't been indulging in the hype.

Whether Waymo are hitting a wall where each additional increment in safety/capability is drastically harder to achieve, who can tell from the outside?

Everything a certain individual is associated with has turned out the be shady or worse. Do a google search on 'Singularity University scam', for starters.

It is no wonder that a variety of self-driving car startups are going to be sued for being snake oil.

Waymo seems to be the leader in the suburbs of Phoenix, Arizona. I'm not sure if they are the leader anywhere else, though.

Well, who else would you nominate for leader elsewhere? And what they can do in Arizona translates to the other places which are similar, but may take more work to move to harder places -- but nobody else is leader in those places.

I don't have much basis for knowing what Google can do outside of 20 square miles of Arizona, let alone what other companies who haven't yet released a product can do.

But as I understand it, Waymo's cars don't work autonomously at all in most places. They rely on significant infrastructure, including detailed HD maps and a constant connection to the mother ship. So any car that works at all is better than they are in most places.

If what they can do within their geofence translates to other places that are similar, why aren't they operating (driverless) in any other places? Apparently it's not as simple as you suggest.

They continue to operate extensively in Silicon Valley. You can't spit without hitting one in some parts of town. They say robotaxi service here is coming fairly soon.

Yes, they rely on maps, but that is a specialty at Google. (In fact it was people who did Streetview who started the project.) They do not need a constant connection to the mother ship. In fact, they can drive very far without intervention. They use connection to the ops center not for safety, but to resolve complex problems. How often they do that Waymo has not said. How many ops center workers per vehicle we don't know.

They are doing a pilot. They do that in the service area they decided to do. They could have a larger area. What makes you think a larger area is the right plan for their pilot?

What they're doing in Silicon Valley and elsewhere is not demonstrably better than every other company. It's only in a 20-square mile of Arizona that they're running "driverless" (without a driver in the car). In those 20-square miles, they're probably the leader.

I never said a larger area would be better for them. In fact, I assume they have good reasons for not having expanded the area. I assume those reasons are that they aren't ready to safely expand the area yet. Presumably either they're not comfortable enough with their software yet, or expanding the area is very expensive. Probably both.

Maybe Waymo's cars can drive very far without intervention (this could be said of several companies cars). But I highly doubt Waymo is letting its cars go "driverless" without a human monitoring them.

Hopefully they will start running a robotaxi service in Silicon Valley soon. And hopefully they'll quickly, if not initially, make it a "driverless" service. Then hopefully the next city after that will take a lot less than a year to expand to. At 20 square miles a year it will take several millenia to expand to most of the developed USA (let alone the world).

Actually, I say hopefully, but I guess I don't mean it. I kind of want the centralized Waymo model to fail. Google already controls and monitors enough of my life. I'm not sure how comfortable I would be with them also controlling and monitoring everywhere I travel by car.

(They gave me a free nest mini. It's still in the box. I'm probably going to try to sell it.)

I am sure the cars are "monitored" and I think the intention of Waymo and many others is to have unmanned (or "rider only" as Waymo calls it) be monitored for well into the future. It's reasonable if the vehicle calls for assist rarely enough that you can have one ops center employee for many cars. Remote monitoring never provides tactical urgent response assist and so only should be needed in a situation where a strategic decision is needed with some time to examine and think. Like a particularly complex intersection. But that's fine. Perhaps only 0.5% of your cars are going through those situations at any given time, so it's affordable.

As we know, getting a car to run unmanned is a long tail problem. Getting to 99.9% is easy, but getting to 99.99999% is much harder. 10,000 times as hard in some ways, though not 10,000 times as much work as you get better and improve faster with time.

But if you can say for the class of problems which can be solved with human oversight that you just punt them rarely to a human, you can be on the road much faster. Just as Tesla is on the road today with autopilot because it can punt all the responsibility to the driver. But this scales and provides great benefit to the users.

I've often heard the statement comparing getting to 99.9% with getting to 99.99999%. I'm still not sure what exactly it means. But moreover, I don't think it's necessarily true if you have a human available to "make strategic decisions." Yes, corner cases are abundant in driving. But there's almost always a simple choice that, while possibly annoying to the people in the car and possibly to people in other cars, is safe. Usually that choice is to stop, and in the 20 square mile area where Waymo is operating by remote control, that is almost always a safe choice (I assume there aren't any high speed highways in that area, but maybe there are).

So no, I don't think it's that's big of a problem to run unmanned in a 20-square mile radius. In fact, if that were your only goal, it could probably have been done a decade ago. What Waymo has done is impressive because we assume they are mostly trying to solve the general problem, not pull of a gimmicky demonstration.

I'm not faulting Waymo for using humans to monitor their vehicles and to "make strategic decisions". In fact, I think if anything that they should embrace that even more. Once they can get the cost operating Waymo One down to less than the revenue they can earn from it, they can start expanding widely, and can start to get the billions of miles they're going to need to create a true driverless car.

But they don't seem to be even there yet, so I don't think we can yet say that they are definitely the leader.

You say "this scales" and I'm not sure if you meant Tesla's model or the Waymo One model or something else. Tesla's approach is certainly scaling very well. Waymo One so far has not shown that it can scale. In fact, it's not even clear that they yet have a model that can compete on price against Uber and Lyft.

Add new comment