The Waymo Study Says Their Cars Can Prevent Accidents

In the world of Robocars, from the very beginning people have claimed that cars will do better than humans when it comes to safety. Because more than 90% of accidents are caused by human mistakes, robots do not make those mistakes and we have fewer accidents. This has always been true by definition – largely no one wants to release a car to the world until they reach that level of safety, and perhaps more.

Waymo threw a gauntlet last fall when he released a document showing that they were, at least, and better, at least in the easy-going atmosphere of suburban Phoenix. Today they released a study stating that their car is better at avoiding an “other” car in an accident, which is not at fault. They believe that the Waymo car will do better to get out of the way when someone else becomes the cause of an accident.

To perform this test, they extracted a database of every fatal accident in recent history inside the area near Chandler Arizona where they currently operate. They wanted an area where they were ready to emulate things in a realistic way, and that could be compared to their current records.

He received 72 accidents since 2008 and read all the details in police reports to try to recreate the accident in his simulator. Then they had the Waymo driving system in which the car had a role that originally led to the accident. Each time, it went better and there were no accidents. This is not too surprising – indeed it would be surprising if it is not capable of doing so. The Waymo driver should not be at all likely to make the same mistake to some stupid person made on the roads in the same situation.

More interestingly, they repeat each two-car crash with the Waymo system, causing the other car to run. He did well. They prevented 82% of accidents. 10% of the time they reduced the severity of the accident. Things stayed the same 8% of the time – but in all cases where 2 cars were stopped and hit by a bad car, that can be a very difficult situation to improve.

They are particularly proud that two-vehicle collisions that resulted in fatalities for pedestrians or cyclists were always able to prevent collisions.

There are several reasons that these simulations may not be correct, and Waymo goes into several of them in his paper. For one, you only have a police report and there can be many factors that the police did not write or which witnesses did not see. And simulation is still simulation. But we find some good evidence that robots can be capable of being superhuman in quick response times to accidents, and make us safe even when they are dangerous.

Furthermore, as we know, Waymo cars have been involved in many real and fake accidents, not as a fault party, so they are certainly not a good record as the simulations suggest. Simulations focus on fatal accidents and convert them to non-fatal or prevent them, which is easier than fixing all accidents. The video shows an example of a Waymo that was badly hit by a car that jumped the median and was going the wrong way, which could not stop it:

I am waiting for Wyomo to go even further. In his fall report, he outlined 29 accidents where his safety driver intervened and prevented an accident. In the simulation, they found the Waymo driving software, it did not disintegrate, continued unabated in an accident. The safety driver was better, it was able to avoid an accident that the Vaymo system could not. Waymo has declined to comment, but it is my expectation that when something like this happens – and each one is a big deal on any self-driving team – attempting to get the car to stop that particular accident Will be attempted, which was recorded by the car, as it can be simulated with better accuracy. Since humans avoided it, it is possible to avoid it, even if they are specially trained professional drivers.

As a result, one hopes that if a similar situation arises again, this time the Waymo software will succeed where it previously failed. Broadly speaking, if a robber was involved in an accident that could have been avoided, it would never make the same mistake. It is not at all like humans. If a human makes a mistake, it does not prevent the next human from making the same mistakes. Robots work differently.

In time, it is probably worth codifying the millions of fatal and non-fatal accidents on which simulators have good reconstruction data. Every car should learn if it can survive the mistakes of that accident – and never make them for the first time in the real world. This is the way to true automotive safety with hoods. When they report that more than 90% of crashes are due to human error, they do not talk about drivers in other cars. They can help prevent accidents even when it is not their fault, and thus Roboker can reduce accidents simply by not causing them.

Leave a Comment