Google Self-Driving Car Might Have Caused First Crash In Autonomous Mode (roboticstrends.com)

An anonymous reader writes: While driving in autonomous mode, a Google self-driving car was involved in an accident with a public bus in California on Valentine's Day, according to an accident report filed with the California DMV.The accident report, signed by Chris Urmson, says the Google self-driving car was trying to get around some sandbags on a street when its left front struck the bus' right side. The car was going 2 mph, while the bus was going 15 mph.Google said its car's safety driver thought the bus would yield. No injuries were reported. If it's determined the Google self-driving car was at fault, it would be the first time one of its cars caused an accident while in autonomous mode.

Comments

How many autonomous crashes were overridden?

[Your.Master]

I'm a big believer in autonomous cars, but when I see

Google said its car's safety driver thought the bus would yield.

it makes me wonder how many crashes we would have had in autonomous mode, if there weren't an attentive driver who was fully aware he was sitting in an experimental vehicle.

Even if the first rounds of autonomous cars still require a driver for override (for legal reasons if nothing else), it seems like the number of autonomous crashes that likely would have happened is the number has to be driven way down to be comparable to, or less than, the ones with human drivers*; it's not really the amount of autonomous crashes overall that is important.

Also makes me wonder whether any of the manual mode crashes were initiated in autonomous mode and the manual override driver just couldn't recover the situation.

*whether average human drivers or above-average human drivers or even below-average human drivers are the standard is up for debate.

Talented Slashdotters

[avandesande]

Could somebody please come up with a fitting car analogy?

Re: The approach is too hard...

[Anonymous Coward]

For the number of miles driven it is significantly safer when compared against an average person driving the same. I would also suspect that we would see even greater safety in areas other than California, have you ever driven here?

Re:Buses have right of way

[Obfuscant]

And if the google car was going 2mph then the correct action is for the google car to stop for the sandbag rather than jump in front of a bus.

From the description in the summary, it sounds like the car ran into the side of the bus. It didn't jump in front of it, it sideswiped it as it tried to go around sandbags in its lane. Assuming that the bus was in its own lane, the car had to leave the lane it was in to do that.

Every discussion about safe driving I've seen in this forum has had the "safe" drivers claiming that the only safe thing to do is stop when faced with an impediment to traffic, not to try swerving around it. And the autonomous discussions have all claimed that AV will be safer than humans. Considering that the car was going just 2 MPH, I cannot imagine that it could not have stopped for the sandbags and waited until the next lane was clear, but I'm sure someone can explain why it didn't.

And the car "safety driver" assumed the bus would yield to him. I for one welcome our new robotic overlords.

This touches on what I said before.

[morethanapapercert]

In previous posts about autonomous cars, I raised the question of how these vehicles handle the highly variable and difficult to anticipate changes in the routing caused by construction. I worked for several years in road construction and can tell you that an appalling number of humans get confused by having to change lanes in response to a flagman or pylons/barrels, ignoring any existing lane, curb and signed markings.

In this case; having read the article (I know, I know...) it seems that the car programming is overly optimistic about predicting the behaviour of vehicles overtaking it. It seems possible that the programming includes implicit assumptions of the likely stopping distance and reaction times it should expect from other vehicles as well. In other words; it "thought" it had sufficient space and time to perform the manoeuvre because it "assumed" a bus would behave and react the way a car might.

I have two thoughts, each in defence of one of the vehicles in this collision:

1) Even the safety driver expected the bus to yield and from I can glean from the article, legally the bus should have yielded. So this was a mistake that even the majority of human drivers might have made in the same situation.

2) Others in this thread have posted criticisms of bus drivers in their city or in general. Much of the annoying behaviours they mention though are pretty understandable from the bus drivers POV. You can't just suddenly hit the brakes if a smaller vehicle or pedestrian darts in front of you. Not only do you have a hell of a lot of momentum (highly variable, depending on passenger load) you also have to make as gradual velocity changes as you can. Your passengers aren't buckled up, you might have a fair number of them standing, with any number of knapsacks, briefcases, skateboards etc etc that become flying hazards when you come stop too suddenly. You have to ease to the left a fair bit when making a right turn because you have a much larger turning radius than most other vehicles. You have to drive straddling lines sometimes because if you stayed tight to the right, you are going to crunch someone, hop the curb or both. On the other hand, if you do stick to the left as much as you can, lots of people are going to pull what Torontonians call a "cabby pass" where the cab illegally passes a bus or streetcar on the right so as to get out from behind it. If they don't use their rear end to block the traffic lane, quite often they'll never get back out because no one wants to stop at the buses back corner and let the bus back in. (I have a relative who is a TTC bus driver and he has passed along some training and daily work anecdotes)