Is the commercialization of fully autonomous driving near? Ability seen from Waymo’s unmanned taxi experiment | WIRED.jp

Waymo, the developer of autonomous vehicles, has long kept secrets about its industry-leading self-driving technology. Waymo has conducted millions of miles of test runs in Arizona and California, including thousands of miles without humans in the driver’s seat.

However, until November 2020, almost everyone who had experience riding this autonomous vehicle was bound by a strict non-disclosure agreement. Under these circumstances, Waymo began disclosing information on its autonomous driving technology in October 2008.

People around Chandler, a suburb of Phoenix, Arizona, are now able to take a fully unmanned taxi without a driver. At this time, it was also allowed to record while riding, publish videos, and tell reporters about their experiences.

Joel Johnson, who lives in Arizona, jumped at the opportunity to record the actual performance of Waymo’s unmanned taxi. Johnson, a student at Arizona State University, is on leave due to a pandemic of a new coronavirus infection.

Johnson lives near Waymo’s service area and has used his spare time to test the performance of Waymo’s unmanned taxis.He says he has boarded the service more than 60 times in the two months since it opened the service to the public, and posted dozens of videos on YouTube.Posting

Driving that was “boring” in a good way

Perhaps the most amazing thing about the video posted was that it was very “boring”. In the five-hour video, Waymo’s vehicle never made a big mistake.

This is in contrast to Tesla’s October 2008 release of “fully autonomous driving” software. I watched a video of a Tesla user trying fully autonomous driving for about 3 hours, but there were more than 10 scenes where the driver manually intervened, two of which could be a dangerous collision. It was.

Johnson’s experience is quite different. “It’s been a solid drive so far,” Johnson told a telephone interview.

You still have to wait in line to use the unmanned taxi service “Waymo One” provided by Waymo. For this reason, Johnson has provided a ride experience to a large number of people who do not yet have access to this service. In addition to friends and family, he has brought in industry officials and YouTubers who have come all the way to Phoenix to actually experience Waymo’s autonomous vehicle as his guests.

“Everyone in the unmanned taxi I called trusts Waymo’s self-driving car,” Johnson says. “It’s so smooth that you forget you’re driving on your computer alone.”

“The way I brake and accelerate is so good that I don’t even notice it,” he said while driving.One of the passengers says.. “It’s getting really smooth.”

Progress in dealing with pedestrians

Johnson has been in Waymo’s self-driving car since joining Waymo’s private Early Rider program in mid-2019. And he has realized great progress so far.

“Waymo has solved issues such as turning left in the presence of oncoming vehicles (on the right side),” Johnson said.Talking in one of the videos.. “It’s definitely improving over time.”

“It was awesome,” a passenger told Johnson at the end of the ride.Said.. “You’re getting smarter. It’s much better than in March.”

See Also:  New State Program to Provide Millions in Rental Assistance | Univision Phoenix KTVW

Johnson also points out that coping with pedestrians is improving. He has a video of Waymo’s self-driving car driving in a pedestrianized Costco parking lot.Posted in October 20.. The vehicle waits patiently until there are no pedestrians around and starts with confidence.

“With all these pedestrians, by March 2008, I would have used the brakes so many times that I could be whiplashed,” Johnson commented in a subtitle in the video. “And as of July 19th, I would have given up completely. That’s not the case now.”

Accident occurrence is equivalent to “once every 10 years or more”

Waymo’s autonomous vehicles still seem a bit too cautious when there are pedestrians around. In a recent video of Johnson calling an unmanned taxi into a crowded parking lot at a retail store, he had to wait nearly three minutes after pressing the “start driving” button to travel a reasonable distance. There were clearly so many pedestrians and other vehicles around that Waymo’s vehicle determined that moving forward was unsafe.

If humans were driving in this situation, they would almost certainly have been moving earlier. However, it is difficult to complain to Waymo about this. It’s much better to be a little slower than to run the risk of catching people.

Of course, Waymo’s self-driving car, even if it’s been driving impeccable for four hours, and even 40 or 400 hours, isn’t enough to prove its safety. A lot of data is needed to properly assess safety.

And Waymo has actual driving data of over 20 million miles (about 32 million km). However, most of them are public road tests with a safety driver in the driver’s seat, and the total unmanned driving is only 65,000 miles (about 105,000 km) by September 2008, which is very small.

Until recently, Waymo kept the data private, making it generally difficult to evaluate its technology. Then, in October 2008, Waymo released data on the actual driving performance of its own vehicle and took a major step toward increasing transparency.

This data includes a record of 6.1 million miles (about 9.8 million km) of driving in the Phoenix metropolitan area with a safety driver in the driver’s seat in 2019, as well as unmanned driving from the beginning of 2019 to September 2008. Includes records for 65,000 miles (about 105,000 km). In the course of more than 6 million miles (about 9.6 million km), there were 18 collisions involving Waymo vehicles.

As explained earlier, most of the mileage recorded in the data had a safety driver on board to allow manual intervention in the event of a potential collision. For this reason, Waymo is also conducting simulations to estimate how well the vehicle would have performed without intervention in all situations where there was manual driving intervention. These simulations predict that 29 more collisions would have occurred without manual driving intervention.

See Also:  "Alaskan Bush People" -Bear mourns his father Billy (†)

It may sound like a total of 47 accidents, but it’s important not to overlook the denominator. The frequency of actual accidents involving Waymo vehicles and accidents that would have occurred without intervention is about once every 130,000 miles (about 210,000 km). With a typical American mileage of about 1,000 miles per month, accidents occur more than once every 10 years or more.

Most of the accidents are “minor” or “negligence of the other party”

However, it is surprisingly difficult to derive numbers that are comparable to typical human driving. The 47 collisions reported by Waymo include some extremely minor ones.

For example, there are reports of pedestrians hitting the side of a stopped Waymo vehicle at 2.7 mph (about 4.3 km). Two of the simulated collisions were that a bicycle and a skateboard would have hit a stopped Waymo vehicle at 2.2 mph and 5.9 mph, respectively. It was.

If it is such a minor low-speed collision, it is unlikely that it will actually be reported to the police. For this reason, the frequency with which a typical human driver experiences such a “collision” is unknown.

More importantly, most of these 47 accidents appear to be the negligence of the other driver. For example, one-third of actual and simulated accidents were rear-end collisions.

In addition, of the 14 rear-end collisions that actually occurred and one that was simulated, all but one was a rear-end collision with Waymo’s vehicle. The remaining two rear-end collisions were simulated, but it was that Waymo’s vehicle would have collided with the rear of another vehicle at 1 mph (about 1.6 km).

Most of the side-rubbing accidents (8 out of 10) occurred when the other vehicle changed lanes to the lane in which the Waymo vehicle was. In addition, regarding one of the accidents that occurred when the Waymo vehicle changed lanes, the company explained that the other vehicle was traveling over the speed limit of 30 miles (about 48 km) or more.

In addition, over 6 million miles (about 9.6 million km), there were only three serious collisions that actuated the airbag (five more in the simulation). None of these did (or would have) led to serious or life-threatening injuries, according to Waymo.

Reasons for the very time-consuming deployment

In summary, the incidence of Waymo vehicle collisions is low and no life-threatening collisions have occurred over 6 million miles (approximately 9.6 million km). Most of the further collisions were the negligence of the other driver. From these results, it is reasonable to consider that driving a waymo vehicle is safer than driving an average human driver in the overwhelming majority of situations.

Still, there are still big unsolved questions. Is Waymo’s self-driving system less likely to cause a fatal accident than humans?

The problem here is that in the United States, there is only one fatal accident on the highway for every 100 million miles (about 160 million km) of travel. In other words, to prove that Waymo’s self-driving system is less likely to cause a fatal accident than a human driver, even with almost impeccable driving records of 6 million miles. , It’s not enough.

See Also:  Former Chicago Bears star Michael Richardson charged with Arizona murder

There are some nasty questions about this, such as “chicken or the egg first”. After all, even a well-funded company like Waymo probably can’t afford to test their technology for hundreds of millions of miles before commercialization. Still, it’s dangerous to commercialize an unmanned autonomous vehicle before proving it to be safer than humans.

This is why Waymo spends a great deal of time tackling the problem. The maximum speed of the company’s vehicle is 45 mph (about 72 km). This makes sense given that highway collisions are likely to be fatal.

Waymo is also rolling out a fully unmanned driving service at a very slow pace. Even now, two years after the planned commercial service for unmanned taxis was scheduled to begin, the service is still provided only about 100 times a week. Converting this to a human taxi driver, it is only about 2 to 3 people full-time.

Is it profitable with a careful strategy?

Perhaps behind the scenes, a large number of people are involved, monitoring and analyzing each run to perfection. If the result of the evaluation is positive, it is presumed that Waymo will increase the number of vehicles on public roads. Eventually, they should be fully confident in their strategy and expand the target area to other metropolitan areas, starting with the entire Phoenix area.

This is very different from Tesla’s approach. In the case of Tesla, there are relatively few tests, and there is a tendency to respond by updating software. And it relies on its customers to monitor system mistakes. In the United States, at least three Tesla users have died since 2016 without being able to correct the misjudgment of the semi-automatic driving technology “Autopilot.”

Waymo, on the other hand, seems to be more determined to maintain a near-perfect driving record. The question, however, is whether it is possible to scale up at a profitable speed in the business while taking these cautious strategies.

* Articles related to autonomous vehicles by “WIRED”Here.

RELATED ARTICLES

.

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.