With human behind wheel, Google's self-driving car crashes

"Do you realize that we could change the capacity of highways by a factor of two or three if we didn't rely on human precision on staying in the lane but on robotic precision, and thereby drive a little bit closer together on a little bit narrower lanes and do away with all traffic jams on highways?"

Driving a little bit closer together is called tailgating and results in a traffic citation. We can do that now except for the legal constraints.

Can't narrow the lanes much unless you force the transfer trucks off the roads.

Google's has the resources to play at this and we will doubtless learn more about the limits of our technology by Google's experiments. But we are far from designing systems which can travel on standard roads with no human intervention.

You guys scraped the LA Times article, removed all links that pointed to the original story (which actually has pics of the crashed car), and then inserted your own keyword based hyperlinks back to internal tags that don't have anything to do with this? For some reason I had it in my head that this was a legitimate source of news, did this site change hands in the past couple of years?

The weakest link in Skynet.. are the humans.

Driving a little bit closer together is called tailgating and results in a traffic citation. We can do that now except for the legal constraints..

No, we (human) can't do that. Tailgating is dangerous at high speed and could cause a fatal crash. But if done properly it will reduce fuel consumption by 60% or more.

...we are far from designing systems which can travel on standard roads with no human intervention.

We don't need new system, the car can already drive by itself by using pre-existing traffic rule. It already have GPS, maps, and sensors and an AI that plotted the course and follow traffic rule. Its all already there...

No, we (human) can't do that. Tailgating is dangerous at high speed and could cause a fatal crash. But if done properly it will reduce fuel consumption by 60% or more.

Tailgating cannot reduce fuel consumption by 60%.

We don't need new system, the car can already drive by itself by using pre-existing traffic rule. It already have GPS, maps, and sensors and an AI that plotted the course and follow traffic rule. Its all already there...

No, we are not "already there".

But we are far from designing systems which can travel on standard roads with no human intervention.

That couldn't be farther from the truth! I encourage you to watch the TED conference video of the self-driving cars. They manage to avoid accidents in situations where I couldn't - and I'm a 24 yr old with a completely clean driving record who has professional racing experience.

You guys scraped the LA Times article, removed all links that pointed to the original story...

I have reasons to believe that the stories you see on PhysOrg are populated by a bot or some sort of aggregation algorithm, similar to Google News. Except PhysOrg's is geared toward science more than other such algorithms. I doubt there are actual humans posting these articles.

No, we are not "already there".

Why not? YOU can do that, given some directions, can't you? Why can't a car, that can plot its route with GPS and literally has 360 degree field of vision (compared to your two human eyes) - why shouldn't it be able to complete the same task? These cars have more sensors than humans do ;) And they're programmed by some really smart folks. 160,000 miles is proof enough. You also should watch the TED video.

Actually - even if the car was on autopilot it is unlikely that Google would admit this, isn't it? In any case the human driver/autopilot controller may be liable get a ticket for the accident. The argument,"It was the computer what did it!" won't cut much ice in court I would think. So no loss to say it was human driver error, except for some loss of face on the part of the driver course. Of course in future when computer driven cars are much more common and proven to be a lifesaver it will be on the onus of the driver not using the computer to prove the computer was at fault. I can imagine the glee with which this will be met in legal circles ;)

Were probably about 20 years out. Wont that be awesome to be able to sleep on a long road trip and get to vacation ready to have fun and not be exhausted form the drive.

No, we (human) can't do that. Tailgating is dangerous at high speed and could cause a fatal crash. But if done properly it will reduce fuel consumption by 60% or more.

That is a fascinating insight!

No, we (human) can't do that. Tailgating is dangerous at high speed and could cause a fatal crash. But if done properly it will reduce fuel consumption by 60% or more.

Not only can tailgating cause a fatal crash. But if you have a pack of 20 cars tailgating eachother all going 60 mph, and a single one of them crashes, you'll end up with an enormous pileup. Even in the most well designed systems are vulnerable to something. A robot can't predict when a deer will jump out into the road. If we ignore the risks and take shortcuts like tailgating rather than improving the system, then these deer could cause massive tragedies.

Hell, I don't drive, never have, what I don't understand is why the steering wheel is in the front seat. I would feel much more comfortable driving from the back seat. You hear me Google ?

Tailgating cannot reduce fuel consumption by 60%.

@Dogbert
While tailgating can't reduce by 60%, as much as 40% is feasible when all the cars are managed by computer. This wouldn't increase the size of crashes or anything like that, as each car would brake at exactly the same time, thus stopping any potential for collisions. Even in the event of an accident, this would not cause a pile-up at all.

Wearing my software architect hat now, with 29 years experience...

These self driving cars (which, I believed are derivatives of the Stanford winning vehicle in DARPA's X-Prize a few years ago) are incredibly impressive. But, I would not trust my life, my family's lives, the lives of anyone I cared about, nor the lives of complete strangers in them.

Why? As amazing as they are on a controlled track, and even impressive in Bay area traffic, apparently, there are far too many unpredictable circumstances that have not been thought of by the programmers and not coded into the system. They've got a great start, but before we all decide to buy one and take a nap behind the wheel, there's still considerably more work to be done. 90% of an app is done in the first 10% of the development time. The final 90% of the time is working out all the things you didn't think of, your bugs, etc...

Looks like they're 90% done after about 5-10 years of work. Maybe 50-100 years remains... 20 minimum.

I realize what I said isn't what anyone wants to hear, and certainly not what I want to believe, but the cold, hard reality is that these are good enough for impressive demonstrations, but to put many in the real world is not safe at this point in time. How many times have you had to adjust driving for the /predicted behavior/ of another driver? There are all sorts of characteristics of other cars and drivers that give us hints to their yet to be seen behavior, and we all adjust, subconsciously to avoid potential crashes.

How many times have we seen a demo of MS Windows that worked great during the demo install it, use it a while, and discover it wasn't all it was cracked up to be? Unfortunately, an impressive demo does NOT prove a hardened product.

bit.ly/rnVyLb

Again, I realize we've all been ready for this tech our whole lives and nobody wants to believe it's not ready yet, but we're going to have to wait just a little while longer. It WILL get there. Just not YET.

While I agree with the last 2 posts that we aren't there yet, I would like to suggest that the reality of self driving cars is closer than we think.

I would posit that ancilliary developments will move this technology along faster than we would otherwise naturally predict. My goto example for this is the Microsfot Kinect. The issues related to the development of that device were very interesting as they found a somewhat analog methodology to handle depth, which allowed significantly less processing requirements. It may take another 5 years, but I wouldn't be surprised if a computer driven car is safer than a human driven car within just a few years.

No, we (human) can't do that. Tailgating is dangerous at high speed and could cause a fatal crash. But if done properly it will reduce fuel consumption by 60% or more.

Not only can tailgating cause a fatal crash. But if you have a pack of 20 cars tailgating eachother all going 60 mph, and a single one of them crashes, you'll end up with an enormous pileup. Even in the most well designed systems are vulnerable to something. A robot can't predict when a deer will jump out into the road. If we ignore the risks and take shortcuts like tailgating rather than improving the system, then these deer could cause massive tragedies.

There is also the human element i.e. some dirtbag who hacks a car for fun. Or somebody wants to kill one person and has the rest die to cover up the murder. I suspect that by the time the cars are trusted to drive themselves unwatched that Minority Report will be seen as a joke because it didn't go far enough.

Yes! Hacking is going to be another HUGE problem. It's bad enough that someone can hack your PC or your cell phone, but your self-driving CAR?!?! That's a worst case scenario and it is all too real and we know from experience that it will most definitely happen. The manual override switch will have to be completely mechanical and not rely on the on-board computers to release control.