Login

Humans Cause Waymo Crash

A human driver swerved to avoid another car. The result was a crash with a Waymo vehicle.

On Friday, a Waymo (Google) van was involved in an Arizona crash. Police said the car was in automatic mode, but Google claims the car was in manual mode.

Tough Questions Not

The above video shows clearly what happened: Human error.

Seth Tyler, a spokesperson for the Chandler Police Department said “Waymo and the driver of the vehicle won’t get cited for anything because she didn't do anything wrong.”

Even if the Waymo car had been in autonomous mode at the time, self-driving tech can't make up for errors made by other human drivers.

"This crash is indeed pretty much unavoidable for AVs," says Raj Rajkumar, who researches autonomous driving at Carnegie Mellon University. "As can be seen in this video, when a vehicle out there goes out of control, no one really knows or controls what happens next—that vehicle can swerve in one of many different ways, roll over, etc. The AV in turn must make sure that, in an attempt to evade the situation, it does not make the situation worse given the speed at which events like this unfold."

“The images simplify the story and look like terrible accidents,” says Bart Selman, an artificial intelligence expert at Cornell University. “Lots of mistakes are made by human drivers. We have gotten used to that and don’t even report that anymore.”

Indeed, in 2016, human drivers in Arizona averaged nearly 350 crashes and two deaths a day. That’s why the promoters of autonomous technology harp on the facts that nearly 40,000 people die on US roads every year, and that human error causes more than 90 percent of crashes. Letting robots—which don’t get drunk, distracted, sleepy, or ragey—take the wheel could put a serious dent in those figures.

Sensational News in Perspective

These autonomous crash reports are sensational. No one reports on cars crashes of humans. There are simply too many of them. Here are some US Car Crash Statistics.

  • Over 37,000 people die in road crashes each year
  • An additional 2.35 million are injured or disabled
  • Over 1,600 children under 15 years of age die each year
  • Nearly 8,000 people are killed in crashes involving drivers ages 16-20
  • Road crashes cost the U.S. $230.6 billion per year, or an average of $820 per person
  • Road crashes are the single greatest annual cause of death of healthy U.S. citizens traveling abroad

Uber Retake

This crash is fresh on the heels of arguably negligent Uber testing. Some new facts have surfaced regarding the fatal Uber crash.

When Uber decided in 2016 to retire its fleet of self-driving Ford Fusion cars in favor of Volvo sport utility vehicles, it also chose to scale back on one notable piece of technology: the safety sensors used to detect objects in the road.

That decision resulted in a self-driving vehicle with more blind spots than its own earlier generation of autonomous cars, as well as those of its rivals, according to interviews with five former employees and four industry experts who spoke for the first time about Uber’s technology switch.

In scaling back to a single lidar on the Volvo, Uber introduced a blind zone around the perimeter of the SUV that cannot fully detect pedestrians, according to interviews with former employees and Raj Rajkumar, the head of Carnegie Mellon University’s transportation center who has been working on self-driving technology for over a decade.

Criminal Negligence?

Uber's decision to knowingly introduce blind spots was at best a piss poor decision. A case can be made criminal negligence on the part of Uber.

Technology Will Improve

The technology will improve. The skills, drunkenness, and sleep deprived behavior of humans won't.

Mike "Mish" Shedlock

I am looking forward to the day when everybody will be driving only at the government approved speed limit, i.e. in autonomous mode. You did realize that, didn't you?

My uncle has a country place, that no one knows about....

To Max. Yes. No more idiots speeding. Tens of thousands of lives saved. Millions of injuries avoided. And you’ll get places faster as the autonomous vehicles will work together to improve traffic flow. You did realize that? Right?

Seeing ain't believing.

I wonder whether it will become custom to release the real, raw videos from these incidents. Tripled up and scaled down for YouTube, this video seems almost as misleading as the Uber video.

The Waymo driver, human or not, would have seen a very different thing than what we see. For instance, the video doesn't make it clear whether it's the silver car that ran the light or the dark one. I gather it was the silver car - the one who cleverly swerved but didn't swerve back. And looked, in the video, to be doing about 120 mph without seeing the road in front. :)

It would be interesting to see the rear cameras' views. When you see a red-light-run coming up on the other side of the street, do you immediately bail to the right and hit the brakes as a precaution, letting the sap directly behind you shield you from whatever blowup comes your way?

Mish, you keep citing human auto crash data when there is almost NO comparable safety data of autonomous vehicles to compare it with. There is not even ONE SINGLE fully autonomous vehicle in commercial operation on the road yet. Not even ONE. I'm talking in real commercial use, not just a test vehicle with a safety driver. Most of these cars are being tested in "controlled environments" with safety drivers. They are not driving themselves around the cities of America by themselves. So you can make an assumption they will be safer. But you do not have the date to prove it yet.

Hmmmm, If you look at the car behind the black vehicle that pulled out from the left, it's moving quite slowly too. Makes me think they got the light and started moving forward. And, assuming it's a five second clip, and standard pavement painting, (10 feet of white line followed by 30 feet of space), the Waymo moved forward 5 white lines and about 5.7 30 foot spaces or about 225 feet in the 5 seconds, or about 31 mph. And the car that hit Waymo moved forward for roughly two seconds after the initial collision, covering about 75 feet in the two seconds, but very likely braking over the distance. So if we assume that he was goosing it some to beat the light, he's still not doing much over 50 feet/sec or 35 mph at collision. Steering wheel was likely jerked out of his hands by the collision and recovery time was minimal. Had it been engaged, AI could possibly have seen him accelerating through the intersection and moved to the right lane...

(edited)

@abend237-04 : Nice calculations. I think there was only 1 dark car. The 2nd looks like an artifact of the two cameras' views.

Right about the artifact; I meant the silver/white approaching the intersection behind the black.

abend237-04 : Looking frame by frame (of 82) it appears Waymo goes from 36mph to 26. Slowing for the red. The red shows clearly in the last few frames but is apparent (along with a No Walk) by frame 51.

Which, 51 is the frame the gray car realizes he's done bad. By then he's in the intersection and a few feet from the black car. He cranks his wheel in the next couple frames and is past the black car a hair on frame 54. (YouTube has the video at 16 fps.) He hits Waymo at the half-way dot this side of the last dashed line.

My bet is his foot mashed the gas when he whipped his wheel. He sure flew over the median. Probably hit Waymo wide-eyed, frozen in his seat, accelerating like an old lady in an Audi.

Yeah, black should not have gone in to the intersection. Lights don't stop cars. Yeah, Waymo should have dived right rather that playing spectator saying, "Look at me, everybody. I'm safe in the middle lane."

But after Mr. Civic's fixes that $$$Waymo$$$ van, he's gonna be paying for an old car every year in insurance rates. Doofus managed to hit at two out of three cars inside a hundred yards. Some kind of world record, there.

MntGoat is partially correct. There is not yet much data on autonomous vehicle crashes, but it is building rapidly (many millions of miles, which will grow exponentially to billions in the next decade). So far, so good. Not a single accident “caused” by the autonomous vehicle. All accidents are still the result of humans. And every accident involving an autonomous vehicle will be carefully scrutinized, to make sure that the AV was not part of the problem. And if the AV was part of the problem, it will result in further improvement to AV technology. Sadly, we accept the millions of accidents caused by humans, as “normal”. We accept more than 1 million auto deaths annually worldwide (40,000 in US), and the tens of millions injured. Yet we criticize the scientist’s and engineers who are developing this technology which will undoubtedly save so many lives.

@realist Of those supposed "many miles" of autonomous vehicles, there is not one single autonomous vehicle on the road operating by itself. Not ONE. So "technically" there are “no miles” of data. I do agree of course there will be far less accidents once the technology gets worked out and human error gets taken out of driving. I think there is a application of self driving vehicles very soon in “fixed” environments like say a large mining operation where it’s a lot less complex for a truck to go from point A to point B, or in Agriculture where a tractor drives itself through fields where there is nothing to run into or people to run over. Or say a airport shuttle going from terminal A to terminal B. But as far as self driving cars navigating our complex city streets level 4 and 5 with no drivers in them at all, that may take more time. There are a lot of hurdles there. People may be jumping the gun and hyping that side of self driving cars a bit. But who knows, I am no expert. I just know that there is not even one single self driving car driving itself on the road yet.

“Uber's decision to knowingly introduce blind spots was at best a piss poor decision.”

...and the reason we should expect intelligent decisions from Uber is what, exactly?

Incidentally, this accident dramatizes the opportunity to transform red light camera systems from revenue-grabbers into life-saving accident prevention systems. Deploying AV code and hardware at intersections, where 45% of accidents occur, would accomplish this. If such a system had been deployed here it could have, for instance, switched every single red light in the intersection to rapidly blinking mode thereby warning every driver and pedestrian that the intersection had been instantly transformed into a killing zone by a red light runner. Conditioned by such systems, the black car's driver would have instinctively slammed on the brakes and looked left. Oh, and yeah, snap a 5 second video for the light runner and send them a link to it, along with a $1,000 fine when the AV system has likely prevented a collision.

(edited)

Having worked at tech start ups myself. People have to understand the PR “hype machine” that is part of start up tech companies. We used to call it “marketecture” at software start ups. You goal is to get more VC funding and build “hype” and “buzz” to get bought out or go public, etc…. You are never going to come out and say things like, “this is harder then we thought”, “this might take longer”, “we really can’t do this yet”. You are going to hype how it will “change the world”, how its “right around the corner”, “invent” phantom features that don’t exist, etc… So you can’t read stuff about SD cars from people at Google, VC’s, etc… who have a biased self interest. You have to listen to 3rd party non-biased folks who can look at something objectively and have no dog in the fight. And I’m not saying SD cars don’t merit the hype and they may be here soon. I haven’t done enough research. I just know to be careful as there can be a massive hype machine around new technologies that some times go into years of “troughs of disillusionment” before they become mainstream.

Agree totally. Spent several decades in close-coupled high performance networking and computer systems. Truth is, the best people to move the ball forward on complex new technology tend to be smart, intense, and tightly goal-focused and the over-looming start-up goal is to launch before the cash burn rate kills the project. Cost cutting in AV, however, is a whole different animal; lives are at stake BEFORE the launch. The whole AV thing feels at times like watching a thoroughbred break a leg trying to get into the starting gate...

He says it used to be a farm before the motor law. On Sundays I elude the eye and hop the turbine freight, to far outside the wire where my white-haired uncle waits...

Very fitting!

When you see insane stupid shit like that, how does any artificial intelligence compensate. The people are the bugs in the ecosystem

" No more idiots speeding. Tens of thousands of lives saved." A friend of mine who is a retired EMT/ambulance driver told me he hardly ever went to accidents that were caused by speeding. They were pretty much always driver errors, like the 16 year old in Mish's example. Seems like speeders are actually paying more attention and being more careful, because they are speeding. Speed limits are set to the speed the most incompetent drivers can handle; you did realize that, didn't you?

The clear and obvious solution, and the humane one I might add, is to ban human driving immediately. I'm not kidding. Humans are goddamn incompetent and murderous.

Stories