Waymo shows off how it obeys a cop redirecting traffic.


For years, the most common slide in a skeptical talk on robocars would show a cop directing traffic, and ask how the cars will ever handle that.

Here's a video from Waymo showing off that ability.

I have a New article on my Forbes.com page outlining how they did this and what it means.

Update: There is a report the car, while it did detect the hand signal, also waited for confirmation from the remote operations center. I am also waiting for confirmation -- on this tidbit.


This seems like a problem that might not be high priority if Waymo started with a more limited service. What do you think of the critique offered in the Ars Technical article "Google's Waymo risks repeating Silicon Valley's most famous blunder" (tried to post a link but was rejected as spam)

Nobody can be quite sure of the future of this, because it isn't out to the public yet. So it's not impossible that Waymo will bet wrong, and Tesla or BMW could bet right. I don't believe that's the case right now, and I know that Waymo is aware of the factors named in that article, and has made its decisions based on what it thinks is the best course. Many agree with them, some don't.

Now, it is true that this is a special fault at Google. Google definitely operates on a go big or go home strategy. They have shut down products with huge revenues that would be darling startups outside. So they might make the decision to launch at the wrong level. This is a conscious decision that they take. They would rather fail at doing something big than succeed at doing something smaller. So they start big.

"They would rather fail at doing something big than succeed at doing something smaller."
This is not unique to Google. History has shown large companies generally are not successful in new industries where the initial product is severely limited in some way and the market is small. This is not really a voluntary choice by Google as much as an economic reality.

Unfortunately, the fact that it's a scenario that's on every slide show makes it *less* impressive that the car knew how to handle it.

I wonder exactly what the car is doing there. There are actually three cues to go before the car goes. 1) the cop's hand signal; 2) the fact that the intersection is clear of cross traffic; and 3) the fact that the car in the right lane is already halfway through the intersection. Even without understanding the cop's hand signal, the car might very well have done the exact same thing - come to a complete stop and then waited for the intersection to be clear of cross traffic. It probably needed to understand that the person standing in the middle of the intersection was likely a cop, otherwise it probably would have yielded to them. But even on that point I'm not 100% sure.

I'm really curious if the hand signal mattered.

I'm also curious what the moving stop area was caused by. The stop area starts between the stop line and the crosswalk. Right before the car goes it turns off, then appears right in front of the cop, then moves to a place in front of the crosswalk, then moves to the crosswalk, then disappears again.

What motive would Waymo, already regarded as the undisputed leader of the field, who rarely releases a lot of claims, to lie about what their vehicle did here?

The person in the intersection could just be a pedestrian, or even a vandal trying to cause trouble.

But people would also be fooled as well. I suspect in that situation, if the humans were not fooled but the car was, the people behind would honk, and that would probably trigger the operations center to help.

Have you ever seen a person directing traffic trying to cause trouble? Heard of it?

That's part of my point about the fact that the hand signal doesn't really matter. Even if it was a vandal trying to cause trouble, the car still did the right thing.

(I didn't suggest that anyone was lying about anything.)

It seemed to me, since Waymo said "the car waits for the officer to signal" that you were suggesting that they might be lying or equivocating about that when you wondered if the car had been triggered by something else than the officer's signal as they claim.

It's not clear to me what that means. Is the officer's signal necessary? Is it sufficient?

I doubt it's sufficient. The car shouldn't crash into cross traffic even if it thinks a cop told it to. (It would be an interesting ethical question if it was sure that a cop told it to crash into cross traffic, but much more likely that it just misinterpreted the signal in that case.)

Maybe it's necessary. I'm not sure such a rule would make sense, though.

I think we'd have more information about how sophisticated the system is if the vehicle had seen a cop's signal to not even stop. What the car did in the video is exactly what the vehicle code says to do at a broken traffic light: Come to a complete stop and then proceed carefully when it's safe to do so. (This is sometimes slightly inaccurately phrased as treating the intersection as a four-way stop, but you actually have to use more caution than at a four-way stop, you have to also yield to cross traffic approaching the intersection at a moderate speed, as you don't know for sure that the cross traffic has a broken light as well.)

Yes, we don't know what other clues and conditions they use. But they do say that they wait for (and thus detect) the signal from the officer. In fact, I am sure if they are doing a good job they are looking at many other clues, and they would probably be reluctant (just like a human) if the officer told them to go in a way that would hit something.

You seem to be assuming that whoever sent that tweet had a clue what they're talking about.

Sorry, YouTube video, not tweet.

Just saw your update. That would explain why there's a delay between the hand signal and when the car actually starts going.

I was quoting Waymo's own description of their video.

"Sorry, YouTube video, not tweet."

"The video playback is increased by 3X for viewing convenience" haha. It also hides how slowly they drive until they pass the officer. Better safe than sorry, of course, but being super cautious sounds like a problem when it comes to their deployment, or lack thereof.

Add new comment