Slashdot Mirror
Google Releases Report On Autonomous Vehicle Accidents
An anonymous reader writes: Back in May, a report from the Associated Press pieced together information on car accidents that involved autonomous vehicles. Google, the company testing the most self-driving cars on public roads right now, said the automation technology was not at fault in any of the accidents. However, they took criticism for declining to provide any detail. Now, they've changed that stance, releasing specifics on all of the accidents involving their autonomous cars. They set up a new website for releasing monthly reports. According to their first report (PDF), there have been 12 accidents since 2010. The vast majority of them involved another car rear-ending the Google car while waiting at a stop sign/light. There was one incident where another car rolled a stop sign, one in which another car veered into the AV's lane, and one incident where a Google employee driving the car in manual mode rear-ended another car. None of the accidents resulted in an injury.
I just had my first car accident ever. Got rear-ended at a stop light soon after moving to San Francisco!
Well least there was no google car that got hit by someone texting while driving on the 101...
All the accidents... Were causing by HUMANS and not by the machine.
Can't wait until we get rid of the stupid monkeys behind the wheel...
NEVER your fault. it's the other guy.
I do a great deal of driving. About once a week, someone tries an unsafe lane-change with me opposite. Yesterday, someone attempted to change lanes with me directly beside them. No turn signals or anything. As far as I could figure out, the lady had no clue she had even done a lane change.
It is really hard to detect, react, and prevent someone trying to lane change on top of you, or to prevent someone from rear-ending you. I really hope someone figures out something better than what we have right now.
2011 August
A Google Prius model AV operating in manual mode was involved in an accident on Charleston Road in Mountain View, CA. An employee operating the Google AV to run an errand (i.e., he was not using the vehicle to test our autonomous technology) rear-ended a vehicle that was stopped in traffic. No injuries were reported at the scene. The Google AV sustained some damage.
Yeah, right!
I read this somewhere else at least a week ago.
The pdf is either based in part on psychic predictions or mis-dated. http://static.googleusercontent.com/media/www.google.com/en/us/selfdrivingcar/files/reports/report-0515.pdf
2015 February
A Google Lexus model AV was travelling northbound on El Camino Real in autonomous mode when another vehicle travelling westbound on View Street failed to come to a stop at the stop sign at the intersection of El Camino and View Street. The other vehicle rolled through the stop sign and struck the right rear quarter panel and right rear wheel of the Google AV. Prior to the collision, the Google AV’s autonomous technology began applying the brakes in response to its detection of the other vehicle’s speed and trajectory. Just before the collision, the driver of the Google AV disengaged autonomous mode and took manual control of the vehicle in response to the application of the brakes by the Google AV’s autonomous technology. The Google AV was in manual mode. No injuries were reported at the scene. The Google AV sustained some damage.
Would the accident have happened if Google AV just continued through the intersection? Other drivers are also anticipating that trajectories and speeds of cars on the road do not change suddenly.
our new accident free Diving Miss Daisy autonomous vehicle Google overlords. I would also add I'd be more than willing to serve as a passenger seated in my new autonomous vehicle Overlord while I get driven to my work.
by TheSpoom (715771) Uncaring Linux user here. I have nothing to add to this but please continue. *munches popcorn*
At some point, there will be a significant fraction of autonomous vehicles along with human drivers. The AVs will drive the speed limit, stop at stop signs, and in general perform as the model driver would. The problem is few of the human drivers perform this way, most of us cruise at 5-10 mph over the speed limit, roll through stops, and change lanes without signalling (well, I always signal at least).
I can only imagine the uptick in accidents because of the frustrated drivers waiting on the model AVs. Those will be interesting times.
"Who are you?" "No one of consequence." "I must know." "Get used to disappointment."
Yeah, that strikes me. The human thought the use of brakes was wrong and tried to correct, too late, but just in time for Gogle to say human was in control.
Would the accident have happened if Google AV just continued through the intersection? Other drivers are also anticipating that trajectories and speeds of cars on the road do not change suddenly.
If the computer had continued to apply the brakes (not been overridden by the human) would it have allowed the crossing vehicle to pass in front of it?
If the computer had continued to apply the brakes (not been overridden by the human) would it have allowed the crossing vehicle to pass in front of it?
And, if it had, would the car behind have crashed into it?
But a slower moving car is less likely to cause injury in a crash and a rear collision is less dangerous than a side crash. Given a choice of those two, I'd risk being rear ended. But it sounds here to me that either approach alone may have worked, the switch in the middle failed miserably.
I still have more fans than freaks. WTF is wrong with you people?
I think they just fucked up the description of what happened there. Consider that it takes about 150 ms for a human to react even when they know what they should do, and much longer if they have to think about it, and try to imagine what happened in this accident taking that into account.
According to Google's account of the accident, the self-driving car noticed the approaching vehicle on its right and decided to stop. We aren't told where it decided to stop, whether that was before entering the lane that the approaching vehicle was in, right in the middle of it, or after it. Then the human noticed the car was stopping, thought about what was happening, and decided to do something we presume they rarely decide to do, which is take over control of the car. Given the rarity of such action, I can't imagine the decision to do it being a very fast one. So I have a hard time imagining that, if as the description says, the human took over control of the car in response to its decision to brake, that the car hadn't already stopped at that point, with the approaching vehicle passing in front of it, or if it had decided to stop right in the middle of the lane being used by the approaching vehicle, that it wouldn't have already been struck by the time the human had time to react.
Thus, I can only conclude that the human's action wasn't in response to the car's decision to brake, but rather, the human had already seen the approaching car and was acting independantly of the self-driving car. Thus what we actually have is the self-driving car deciding that it can stop before entering the lane that the approaching vehicle is traveling in, and the human simultaneously deciding that it can accelerate and be out of that lane before the approaching vehicle collides with the car. Obviously both approaches might work depending on the exact situation, but one thing you can't do is decide to take one approach, then half-way through it, decide to take the other approach.
I think that's what happened here. The car decided that braking was the best coarse of action, and applied the brakes hard for about 200 ms before the driver, thinking independantly, began to implement his own corrective action. The result was that with the previous braking, the car was now traveling too slowly to fully miss the approaching vehicle using acceleration, but by the time the human's 200 ms delay allowed him to realize the car had decided to brake, it was now too late for either action to avoid the accident. So it's quite possible that, had the self-driving car not been subject to being overridden by its back-seat driver, it could have avoided the accident.
This is certainly the one accident I'd love for Google to go into much more detail about.
I went to a talk by a guy working on the project. Those cars have cameras pointing in all directions and recording everything. He said in 95% of the accidents you can clearly see the other driver looking at their cell phone.
Since I've had three very minor prangs or praning into, none of which were reported because there wasn't even a scratch on the cars involved, none of those would have been reported unless my car was being specifically monitored.
Additionally, since the vast majority (more than 75%) were other drivers, this indicates, if anything about google cars, that other drivers will deliberately try to ram a google car. Or it's just that reporting thing and it's not non-google drivers being especially dangerous around google cars.
we used to call it the granny lane.
I guess the way to check this is to analyze the liklihood of a human accident versus an automated accident.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Look from the rear mirror, and move more if needed.
"There was one incident where another car rolled a stop sign, one in which another car veered into the AV's lane, ..."
I wonder to what degree it is programmed to not assume that other drivers on the road abide the law, or even drive reasonably?
J