GM Volt Forum banner
1 - 15 of 15 Posts

·
Administrator
Joined
·
22,266 Posts
Discussion Starter · #1 ·
Waymo has turned to counterfactuals, or “what if?” scenarios, meant to showcase how its robot vehicles would react in real-world situations.

In a bid to prove that its robot drivers are safer than humans, Waymo simulated dozens of real-world fatal crashes that took place in Arizona over nearly a decade. The Google spinoff discovered that replacing either vehicle in a two-car crash with its robot-guided minivans would nearly eliminate all deaths, according to data it publicized today.

Waymo identified 72 crashes to reconstruct in simulation in order to determine how its autonomous system would respond in similar situations.

For crashes with two vehicles, Waymo ran separate experiments simulating its autonomous vehicles in the role of each vehicle — first replacing the vehicle that initiated the crash and then replacing the vehicle that responded to the other vehicle’s actions.

The results show that Waymo’s autonomous vehicles would have “avoided or mitigated” 88 out of 91 total simulations, said Trent Victor, director of safety research and best practices at Waymo. Moreover, for the crashes that were mitigated, Waymo’s vehicles would have reduced the likelihood of serious injury by a factor of 1.3 to 15 times, Victor said.

Being rear-ended was the hardest to mitigate.

 

·
Registered
Joined
·
3,872 Posts
Simulations are NOT real world. One of the huge problems in the world today is that everyone, including the alleged scientists who run them, takes simulations as gospel.

Bottom line is that automation has made our roads safer, but until fully autonomous vehicles are on the roads we won't know how much safer.
 

·
Registered
Joined
·
3,244 Posts
Simulations are NOT real world. One of the huge problems in the world today is that everyone, including the alleged scientists who run them, takes simulations as gospel.

Bottom line is that automation has made our roads safer, but until fully autonomous vehicles are on the roads we won't know how much safer.
These weren't simulations. These were real world accidents and they know how the self driving cars would have reacted in the same situations. No surprise that they can't do much for rear enders as where the car stops for something they have no control over the car behind that is following too close or the driver is texting and driving. They will know how much safer as the accident and deaths drop as the percentage of self driving cars increase.
 

·
Read Only
Joined
·
0 Posts
These weren't simulations. These were real world accidents and they know how the self driving cars would have reacted in the same situations. No surprise that they can't do much for rear enders as where the car stops for something they have no control over the car behind that is following too close or the driver is texting and driving. They will know how much safer as the accident and deaths drop as the percentage of self driving cars increase.
From the OP:

"...Waymo ran separate experiments simulating..."
 

·
Registered
Joined
·
3,872 Posts
No surprise that they can't do much for rear enders as where the car stops for something they have no control over the car behind that is following too close or the driver is texting and driving.
Many, not all, but a large percentage of rear end collisions can be avoided by the car being rear ended by simply not following or stopping too close to the car in front of them. This allows the "middle" car that gets hit some room to move forward to avoid being hit from behind.
 

·
Administrator
Joined
·
22,266 Posts
Discussion Starter · #8 ·
Of course they are a simulation. You can't roll back the actual accident and replace one car with a self driving car. These were more like the NHSI crash testing but recreating actual crashes with two cars where the crashes happened.

What you can do is recreate the accident using real cars under the same conditions on the same road and see what if anything changes. That's what they did. The results are impressive but not surprising given the 360° sensor view, continuous monitoring, and faster reaction times.

However if by "simulation" the reference is to computer modeling in a lab, not these were actual field tests.
 
  • Like
Reactions: scottf200

·
Read Only
Joined
·
0 Posts
Of course they are a simulation. You can't roll back the actual accident and replace one car with a self driving car. These were more like the NHSI crash testing but recreating actual crashes with two cars where the crashes happened.

What you can do is recreate the accident using real cars under the same conditions on the same road and see what if anything changes. That's what they did. The results are impressive but not surprising given the 360° sensor view, continuous monitoring, and faster reaction times.

However if by "simulation" the reference is to computer modeling in a lab, not these were actual field tests.
Yes, they recreated actual crashes. With hindsight it's easy to avoid an accident as you "simulate" away all the variables.
 

·
Administrator
Joined
·
22,266 Posts
Discussion Starter · #10 ·
Yes, they recreated actual crashes. With hindsight it's easy to avoid an accident as you "simulate" away all the variables.
Your claim is the tests were faked. OK.
 

·
Super Moderator
Joined
·
6,511 Posts
Aside:
... given the 360° sensor view, continuous monitoring, and faster reaction times.
I'm very happy to have computers and cameras assisting even when paying attention ---

I was driving back from hiking in Starved Rock (IL) yesterday late-morning and was changing lane just by first looking at all my mirrors and then tapping the turn signal to let the Tesla make the lane change.

... I was probably 1/5th - 1/4th into the other lane when my car abruptly moved back to my original lane. I looked in my passenger mirror and a saw from 2 lanes over a med-sized truck decided to move into the lane between us. He was not there when I first looked. After I was moved back into my lane he was just barely behind it seemed as I rechecked all my mirrors. Perhaps we both looked before changing lanes and with clear but we didn't know that each of us were going to move to the same lane.

Both of us would not fit in that one lane and we were driving ~speedlimit at 65-70 MPH. Not a lot of time for human reaction at that speed!
 

·
Registered
Joined
·
3,244 Posts
My claim is these were simulations where all the "errors" were known. As such it's possible to avoid them in a simulation.
The accidents they were looking at were real. What the self driving car would do in similar situation was simulated. The exercise (simulation) wasn't to get the self driving car to avoid the accident, it was to see what would happen if the self driving car was in the same situation with a given set of instructions and sensors. They weren't trying to find a way to avoid the accident, they were trying to see if self driven car would avoid the accident if it were in the same situation. Two different situations. Clearer now?
 

·
Read Only
Joined
·
0 Posts
The accidents they were looking at were real. What the self driving car would do in similar situation was simulated. The exercise (simulation) wasn't to get the self driving car to avoid the accident, it was to see what would happen if the self driving car was in the same situation with a given set of instructions and sensors. They weren't trying to find a way to avoid the accident, they were trying to see if self driven car would avoid the accident if it were in the same situation. Two different situations. Clearer now?
I know this, hence my comment. It's easier to build to the known than the unknown.
 

·
Administrator
Joined
·
22,266 Posts
Discussion Starter · #15 ·
Yes, their aim was to see what would happen if they replaced one of the cars with a self driving car. How would it react? Would it be any better or worse in how it handled the situation, all things being equal? This has been one of those ongoing debates, would the AV prevent the accident? Would still be involved in the accident but reduce death or injury? Or would it be no better than the human driver it replaced?

In recreating actual crashes, I think of what they did as a more sophisticated crash test than you get slamming new cars cars into a barrier while high speed cameras and data sensors record the results. It's the AV version but looking at situational testing. Perhaps a version of this will become a new testing standard for AVs. We know how drivers are handling these accidents, now let's see how an AV does. An updated DARPA challenge course, but now focused on improved safety during accident scenarios.

In the early 2000's Congress decided to invest in driverless vehicles, not so that we could commute to work while checking Facebook but to keep soldiers safer on the battlefield and they tapped the Defense Advanced Research Projects Agency a.k.a DARPA to make it happen.

That's the same agency that helped create the predator drone and Agent Orange, along with GPS and the internet. So in 2004, the US Government held a new kind of competition to jumpstart the industry. The DARPA Grand Challenge. It was the first of three races that would help make self-driving cars a reality.
See How a Bunch of Geeks and Dreamers Jump-Started the Self-Driving Car


16 years later we have driverless Bolts spinning around congested city streets. This video show what the car is seeing in the top screen.

 
1 - 15 of 15 Posts
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top