The forward march of progress takes another step

Google announced a new phase of its self-driving car project Tuesday. The test vehicles, of which there are “about a dozen on the road at any given time,” have so far logged 300,000 miles of road testing without a single accident under computer control. In the next phase of testing, team members will start commuting to work solo, with the robot at the wheel.

The link is here, hat tip to @PaymonFarazi.


Here is the non-adfly link:

That's the big leap. For the marginal revolution in driving, how about automatic emergency braking becoming mandatory for five star crash ratings starting in 2014 in the EU? They will spread everywhere.

The autonomous car, in my view, will come in stages of "driver assist" features.

I like this prediction. It seems to follow what we've seen with the increasing automated safety of vehicles. Take automated traction control, for example. I was watching the documentary 'Senna' which at one point comments on the invention of traction control and it's controversial inclusion in Formula 1 racing. Now it's mandated in all cars in the US. Things moved slowly then (pre internet age and pre-computer controlled car age).

We cannot say robot cars are coming - they are already here. Expecting automatic emergency braking to go along side the automatic shifting, automatic wheel management, etc, seems completely reasonable.

+1. I am waiting for the algorithm that deals with interacting with bicyclists based on earlier discussions.

The mandatory automatic emergency braking ( AEB ) stuff sounds suspiciously like the high-end manufacturers lobbying to their benefit.

Also, the AEB presents an interesting game theoretic dilemma: Suppose I have AEB but the big truck behind me doesn't; won't it grossly increase my chance of being rear-ended into disaster?

Surely having AEB won't increase your chances of getting rear-ended any more than being attentive and having good reflexes would. Are you prepared to argue that those traits are not useful for a driver to have, even if they occasionally result in being rear-ended by another driver who doesn't have them?

It's the difference between good intentions and good results. "Being attentive and having good reflexes " is a commendable trait in normal drivers, yet the net-effect on safety of injecting a small fraction of "super-drivers" might very well be quite non-intuitive.

I'm not arguing against a large fraction of cars having almost equal-stopping-distance AEB's. That's surely a better safety-state point. OTOH a vehicle population with a very widely skewed distribution of stopping abilities might quite possibly be more dangerous.

I remember reading that adaptive cruise control systems, which use a radar to maintain following distance, have a surprising ameliorating effect on traffic. Since they can effectively and safely tailgate, they tend to push slow drivers out of the way, increasing throughput. Some small percentage of them on the roads had a significant impact. Alas, I can't find my source.

But I think you have a really good point. Traffic behavior is complicated, and it will be interesting to see what effect these devices have when there are a lot of them.

I think the argument is, if there is a semi behind you, putting the car in the ditch is better than stopping, but AED may prevent the ditch option.


Alternatively, trying to stop on a dime (commendable) but not being able to, and thus running into that kid / bike / dog / pedestrian that darted in front of you is a feature and not a bug. Given that the tailgater behind you has 10 times your momentum, and 1/10th of your reflexes and no chance of stopping in time.

Of course, in an ideal world your AEB controller would check for "behind-car-risks" and then decide. Don't know if any do.

I feel horrible writing this but.....3M jobs or 33,000 deaths per year? What is a job worth?

I'm not sure what the comparison is? Is 3 Million jobs the amount that will be replaced by autonomous cars? If so, there will be a rather large cost savings in addition to the massive amount of deaths prevented. However, if there were no savings the reductions in deaths alone would still be worth it. Assuming the average job replaced was on the order of $50,000 per year then that would value each life saved at about $4.5 million. Many government regulations are enacted at a higher cost than than.

If we don't constantly break windows, what will the glass-makers do for a living?

If it is indeed economical to replace manually driven trucks with fully automated ones, there is also a savings for the other 300 million Americans who no longer have to pay a trucker to get their goods delivered. This is on top of the other benefits they receive - greatly increased free time where they can be doing more productive things besides driving, more fuel efficient driving, and significantly reduced property damage. I don't see how there is any question this would be a great thing if it can be fully realized.

I wonder how many of those miles were in inclement weather. That will be the real test for robotic vehicles.


So far, Google has tested largely in California between 10am and 4pm on pre-scouted roads. Driving conditions don't get any easier.

I'm coming around to cautious optimism about this, largely because of the possibility of incremental feature adaptation as John Thacker mentioned above, but there's still a lot to be proven. I would strongly recommend a race series. go to 3:30. They have driven in the rain. It seems like the night is where they would really outperform humans, since the LIDAR can see things a lot better than a meatbag.

Snow, OTOH is a big problem. Imagine what a pack of snow falling off the top of a car that you are following looks like. From private talks I've also heard that it can also drift up on sensors.

"Pre-scouted" roads sounds like a cheat, but it was an awesome way of re-defining the problem. By pre-driving the road, you can make sure you never fail to see a stop sign or a traffic light. And since a primary market would be on a daily commute, it would be easy for each person to drive it once or twice themselves (assuming that someone else hasn't already scouted it for you).

Even if the problem could only be solved for pre-driven non-snowy roads, that would be a big win just by itself.

Thank you for the link. Having spent much of my youth driving through snowy and icy conditions may bias me. 24 percent of American crashes are weather related, killing 7,130 people a year:

The designated-driver problem is another one you'd want to solve robustly for weather and new roads. I believe alcohol related deaths amount to roughly 30 percent of traffic fatalities, but I'm not able to find such a good cite for that.

To your other point, Google is one of the few entities that could realistically pre-scout all roads yearly to make the cheat realistic.

I return to my post: they should run a Le Mans -style race series, or perhaps participate in the real thing.

Is the legwork for the pre-scout significantly different than just driving the route a few times? I imagine, perhaps incorrectly, that the roads would end up constantly re-mapped by every automated car. Google (or rather whoever is doing the back-end software when this hits the real world) will probably know faster than the public works department when a pothole appears or when a sign gets mangled / stolen. Even just a few dozen cars commuting daily would be able to feed an incredibly accurate back to home base. Google Street View cars would be obsolete.

I hadn't thought of it that way, and maybe it's possible.

My worry is that there would be a problem the first time some new feature was encountered. If a pothole is a problem, the first car to encounter the pothole would be in trouble. You'd need an actively driven scout car to see it first and a human to note it. I may be making a mountain out of a molehill.

How much longer is my job as a valet safe? I only need a few more years to finish my IT degree.

"The forward march of progress" is quite an elaborate oxymoron.

Oxymoron? It's a redundancy, no?

We march left, we march right, back and forth.

Ricardo: Outch, you're right, I got my terminology confused. It's a pleonasm. Thanks for pointing that out.

I read that as a "neoplasm". Now, would that be a Spoonerism , I wonder?

Geez, this car would drive my wife nuts! Who would she yell at and complain about?

So sad. I just want to drive and you're jumping on an analytic (quant) bandwagon. It's not about the outcome. it's about the trip. You can pry the steering wheel form my cold dead hands.

Comments for this post are closed