October 09, 2010
Google Robotic Cars On California Highways

The science fiction future is going to become the real future in the lives of many people reading this. In an effort to build a practical robotic car Google hired a Stanford researcher who led a team that won the DARPA robotic car contest in 2005. A Google team led by this researcher has developed a robotic car system that has already logged many miles with cars under robotic control on real city streets and highways.

With someone behind the wheel to take control if something goes awry and a technician in the passenger seat to monitor the navigation system, seven test cars have driven 1,000 miles without human intervention and more than 140,000 miles with only occasional human control. One even drove itself down Lombard Street in San Francisco, one of the steepest and curviest streets in the nation. The only accident, engineers said, was when one Google car was rear-ended while stopped at a traffic light.

The article is worth reading in full. Truly amazing. Google researchers believe the cars are already legal because human drivers can take over control of the computer system starts to make wrong decisions.

The researchers think such a car is at least 8 years away from deployment. I see both legal (who is responsible in event of an accident) and engineering challengers for its release. Among the engineering challenges: How to test under a wide enough range of conditions that one can know the computers will always make the right decision? This is a far more difficult challenge than the verification and validation of airplane automatic pilots because the variations in ground conditions are much more numerous.

A robotic car could refuse to go somewhere off known roads under its own control. But conditions on known roads could change in ways that make them unknown. For example, floods could wash out a road section or construction workers could build in detour strips around road pieces being reworked. These sorts of conditions would need to be recognized and the car would need to start insisting that manual control be resumed.

Likely in 10 or 20 years robotic cars will take over much of the work of driving. When that happens accident rates and death rates will go down. After a much longer time of development and use a more mature robotic car capability will enable even non-drivers to be taken places. Among the occupations that will get automated out of existence: taxi drivers. This will lower the cost of taxis and shift more rides to taxis and away from cars owned by individuals. This probably will lead to a reduction in car ownership.

Automation will also reduce the demand for truck drivers and delivery vehicle drivers. The receivers of goods will have to do their own unloading. Do-it-yourself unloading delivery will compete by offering lower prices. I expect this to eventually make home grocery delivery commonplace. Robots in local warehouses (using lots of technologies and business practices already getting developed) will pack trucks with food and other products for a delivery route once enough people signal they are ready to accept deliveries. Therefore fewer trips will be made to stores.

Here's a blog post from Google engineer Sebastian Thrun (and former Stanford lead for their 2005 DARPA automated car win) about their robotic car technology.

To develop this technology, we gathered some of the very best engineers from the DARPA Challenges, a series of autonomous vehicle races organized by the U.S. Government. Chris Urmson was the technical team leader of the CMU team that won the 2007 Urban Challenge. Mike Montemerlo was the software lead for the Stanford team that won the 2005 Grand Challenge. Also on the team is Anthony Levandowski, who built the world’s first autonomous motorcycle that participated in a DARPA Grand Challenge, and who also built a modified Prius that delivered pizza without a person inside. The work of these and other engineers on the team is on display in the National Museum of American History.

Safety has been our first priority in this project. Our cars are never unmanned. We always have a trained safety driver behind the wheel who can take over as easily as one disengages cruise control. And we also have a trained software operator in the passenger seat to monitor the software. Any test begins by sending out a driver in a conventionally driven car to map the route and road conditions. By mapping features like lane markers and traffic signs, the software in the car becomes familiar with the environment and its characteristics in advance. And we’ve briefed local police on our work.

So Google, as usual, hired the best and is playing a key role in transforming another part of our lives.

Share |      Randall Parker, 2010 October 09 04:36 PM  Robotics Cars


Comments
John Moore said at October 9, 2010 8:20 PM:

I fear you are right about the legal obstacles, which is a shame.

Robotic automobiles would be a revolutionary development. The impact on society would be huge and hard to predict. Just for a start, consider all the person-hours wasted driving one's own vehicle - not to mention professional drivers.

Wolf-Dog said at October 10, 2010 12:28 AM:

I have read that the two Google founders also have a limited investment in Tesla Motors (for electric cars):
http://www.eweek.com/c/a/Green-IT/Tesla-Motors-Model-S-Backed-By-Google-Founders-Brin-Page-336717/
But I wish Google were doing a more serious investment directly to the heart of the matter: battery technology. Google can surely afford to donate a few billion dollars directly to the battery innovators in top schools.

Tim said at October 10, 2010 6:13 AM:

I think rolling out the technology in non disruptive ways will be key. Trucks, taxis and other delivery services make sense. Longer term I could see highways with robot car lanes where committed could zip along at 100 mph or better. I see high speed lanes working best in open rural areas at first, making city to city trips faster and easier. I see a big backlash against robot cars by people who see them as taking away freedom. I see this technology giving greatly improved mobility to the elderly. I see this technology making suburbs more attractive by sharply reducing rush hour traffic jams. Very interesting changes coming to our society.!

CyclemotorEngineer said at October 10, 2010 11:00 AM:

Using a robotic Prius to deliver a pizza is a great demonstration, even if the Prius is not the right tool for the job. Robotic vehicles will ultimately enable better matching of payload to vehicle capabilities and tare weight, with commensurate reductions in fuel use.

Engineer-Poet said at October 10, 2010 1:14 PM:
Among the occupations that will get automated out of existence: taxi drivers.
Without taxi drivers to enforce Sharia law against passengers with alcohol and dogs, how will they Islamize us?  And what will the robbers do when robo-taxis only take credit cards?  Oh, what a loss that will be for the scum of humanity!
Basil Ransom said at October 10, 2010 2:53 PM:

Reading the comments on this item at other sites is an education as to how stupid people are when it comes to technologies of the future. I thought it was just my father, but apparently the vast majority of people cannot conceive of technologies outside our current, limited circumstances. They interpret everything in current, narrow terms. They fixate on trivia. Some basic points:

Like autonomous cars getting into accidents - it's accident rate has to be lower than that of humans, but not necessarily zero, for it to be beneficial.
Most people most of the time are not happy to pilot a car. They wouldn't mind someone else driving their shitty commute.
Zip-car style sharing systems plus centrally organized carpooling will make getting from A to B cheaper and easier.
Goodbye to drunk or sleep deprived drivers.
Roads with exclusively automated cars could have much higher speed limits due to nearly nonexistent robot reaction times and perfect technique, and enable further suburban sprawl - bigger, more comfortable living spaces.
A diminished or nonexistent need for cars that can survive crashes would enable new form factors for cars.


Expect cars to become like horses, an expensive hobby appreciated for its panache and sensual pleasures. There's no pride of ownership in owning a Toyota econobox, come on now.

kenh said at October 10, 2010 2:55 PM:

Remember that it was likely only a 1-bit error that caused some of the recent Prius crashes. Now up the complexity of hardware and software by about 100, and calculate the spectacular crashes you'll get with fully robotic cars moving at 100 mph.

Basil Ransom said at October 10, 2010 8:01 PM:

kenh, 100*0 = 0. Hell, 0^100 = 0. In other words, that Prius crash was due to driver error. Look it up. Good job spreading BS though.

cancer_man said at October 10, 2010 8:18 PM:

I heard that a helicopter crashed near a mountain in Colorado, but nobody was seriously injured. The irony is that it was tracking an automatic test car.

Basil, do you know the final outcome on the Prius crashes? I thought the final report wasn't due out for a while.

em745 said at October 11, 2010 7:41 AM:

Any of you running a Windoze version on your PC? Ever had software bugs/crashes? A hardware bit giving up the ghost? Hardware and software, no matter how sophistimacated, are ultimately designed by humans. So "human error" can never be completely taken out of the equation.

Also, in order to approximate human "situational awareness," these robotic cars will have to be riddled with sensors, cameras and range finders (laser, radar, sonar). Even then, I doubt they'd be able to match the speed at which the human eye-brain "interface" can process a 3D visual field. It'll also be interesting to see how well these sensors will function in a snowstorm (i.e. when the car and a good portion of its sensor apertures are covered with sticky, wet snow), or after being splashed with muddy water.

And how will these things react when the unexpected happens? Flight 1549 is a perfect example. Computers are only as smart as their programming, and as sophisticated as modern autolpilots are (many can handle all facets of a flight, including landings), I doubt an autopilot would've managed to pull off what Captain Sully did that faithful day.

Will these cars be able to deal with construction zones? Pot holes? Will they be able to "understand" the hand signals of a policeman diverting traffic? Will they be able to recognize a school crossing guard? How will they react when coming up on a gang of kids playing street hockey? Etc, etc, etc. All of the aforementionned are easy to deal with for a human. Quite the conundrum trying to program that kind of awareness into a computer, however.

Computers crash, components fail... Combine that with a computer's inherent inability to quickly adapt and/or react to the unforeseen, and I for one and not so sure that a robotic car will be that much safer than a healthy, ATTENTIVE human driver.

I applaud Google for their effort, but the days when our roads are teeming with "Knight Riders" are still a _LONG_ ways off.

David A. Young said at October 11, 2010 9:10 AM:

Concerning automated delivery trucks that we have to unload ourselves.

Nah...that's what the ride-along robot's for.

Basil Ransom said at October 11, 2010 10:15 AM:

cancer man, see http://mo.statesman.com/business/electric-fiat-planned-for-u-s-prius-crash-425671.html . Just google Prius crash error . All the links report police concluding it was driver error.

random said at October 11, 2010 11:26 AM:

It seems to me that robotic cars on the open road will only ever be as successful as voice recognition. There is simply too much fuzzy/contextual decision making for computers to handle.
Assisted driving seems possible, perhaps even some sort of temporary auto-pilot, but a vehicle that drives me from point A to point B is not going to be possible in our lifetimes. (Barring some major "SkyNet" type change in computing.)

Mike Anderson said at October 12, 2010 5:38 AM:

Cool technology, but with an atrophied ethical sense. Google seems to have conveniently ignored any idea that California drivers were not afforded anything resembling informed consent while doing over 140,000 miles of SECRET testing on public highways (can you say Public Hearings?). Another egregious example of socializing risk while privatizing reward. Don't be evil, just greedy. After all, It's For Your Own Good.

Randall Parker said at October 12, 2010 8:40 PM:

em745,

Agreed that Captain Sully is better than an autopilot. But Captain Sully is also more skilled and level-headed than most car drivers.

kenh,

Up the complexity to that of the human mind and with humans in control we get tens of thousands of deaths an a much larger number of injuries per year. Robotic cars do not have to be perfect in order to beat humans at driving.

em745 said at October 12, 2010 9:55 PM:

-"But Captain Sully is also more skilled and level-headed than most car drivers."

Commensurate with the need/skill to pilot an airliner. Driving a car is simpler, so there's no reason why a driver can't be just as attentive (on a relative scale).

IMO safer roards could be had simply by better (and mandatory) driver's ed., raising the min. driving age to 18 everywhere, and reforming current DUI laws (which are woefully impotent, IMO).

There are many tech heads out there who think that computers can be programmed to emulate (and exceed) ANY human ability, and THAT is both a ludicrous and dangerous mindset.

From your own article: "Google researchers believe the cars are already legal because human drivers can take over control of the computer system starts to make wrong decisions."

Right... So, if the damn thing makes a booboo, it'll still be up to the human to take control?? So much for relieving the human of all driving responsibility. What if that booboo is about to trigger an accident? Will the human be fast enough to "step up" and prevent it? There goes any "reaction-time" advantage. (Ambulance chasers will have a field day!)

When computers are able to design and improve themselves, carry out a conversation with humans (and be aware of what they're saying), postulate workable scientific theories (or solutions) based on what they're perceiving (and not on fuzzy logic "programming"), look at a detailed picture (or better yet, A MOVIE!) and be able to QUICKLY identify AND DESCRIBE everything in it, and compose an original piece of music, then we can talk about them being advanced anough to drive vehicles. Until then...

(BTW, this is not addressed to you specifically, Randall, just those who are all gung-ho about this nonsense.)

Randall Parker said at October 13, 2010 6:28 PM:

em745,

But a dangerous fraction of the population show themselves incapable of being as attentive as Captain Sully on a daily basis.

I know a guy whose wife just got a ticket for driving while talking on a cell phone. The CHP then pulled over another lady for the same thing right after he pulled her over. People are not paying attention to the road. The desire to gab is just too strong.

Then there's texting while driving. I am amazed that people do this. Yet they do.

Engineer-Poet said at October 13, 2010 9:33 PM:

It doesn't really matter if computers can't equal the perceptual power of humans.  The Google computers have vast superiority in attention, processing speed (it would be no difficulty to react to brake lights ahead in less than 100 msec), and sensory power.  Lidar can already do things that will be forever beyond the human eyeball.

Airplanes don't do what birds do.  Still, they're better.  We'll have robotic cars that will drive more safely than humans, you can be sure.  (And if we move from tire-on-road to wheel-on-rail as in Bladerunner, the problem is almost trivial.)

em745 said at October 14, 2010 6:00 AM:

-"But a dangerous fraction of the population show themselves incapable of being as attentive as Captain Sully on a daily basis."

And robotic cars are the answer?? Or perhaps, would MUCH stricter fines (in the 4 figures) be far more effective in curbing the problem? If fines were set very high, say $5,000 instead of the usual $150 or so, far fewer people would risk getting caught, and safety would improve.

Thing is, ticket revenue would very likely suffer... Municipal and state coffers would feel the pinch... And we can't have THAT, can we. The sad reality is that traffic law enforcement has always been more about revenue generation than safety. Local and state governments need bad drivers in order to stay solvent, and ticket fines (rates) always go through some kind of risk/return analysis when they're set.

But I digress.

-"It doesn't really matter if computers can't equal the perceptual power of humans"

That's your opinion--one that I disagree with completely.

-"The Google computers have vast superiority in attention, processing speed (it would be no difficulty to react to brake lights ahead in less than 100 msec), and sensory power."

Then why the "human override" failsafe?... Why are there still flash-and-blood pilots in airliner cockpits if modern autopilots are already able to take-off, fly and land the plane on their own?

-"Lidar can already do things that will be forever beyond the human eyeball"

Lidar is a distance and speed measuring tool only. Can lidar selectively focus its attention on a particular target on its own? And how well will lidar function in adverse weather, like in a snow storm?

-"Airplanes don't do what birds do. Still, they're better."

Completely irrelevent analogy. Birds are animals, not transport vehicles. Might as well compare a cheetah with a motorbike.

-"We'll have robotic cars that will drive more safely than humans, you can be sure."

Maybe... some day... But not in our lifetime, you can be sure.

-"And if we move from tire-on-road to wheel-on-rail as in Bladerunner, the problem is almost trivial."

Yep, and matter/energy transporters (as in Star Trek) would be even better.

Engineer-Poet said at October 15, 2010 5:19 AM:
And robotic cars are the answer?? Or perhaps, would MUCH stricter fines (in the 4 figures) be far more effective in curbing the problem?
Humans aren't capable of extremely concentrated attention over long periods (unless drugged, which causes its own problems).  Extreme fines will just force people out of their cars, and ironically toward robotic cars.
Then why the "human override" failsafe?... Why are there still flash-and-blood [sic] pilots in airliner cockpits if modern autopilots are already able to take-off, fly and land the plane on their own?
Modern autopilots ("flight directors") largely do; much of the pilot's job is to tell the FD where to take the plane.  The pilot watches for situations the FD can't handle.

The Google car is superior to the FD, because it can operate autonomously using visual flight rules.  It also has a failsafe mode a plane in flight can only envy:  throw on the 4-way flashers and head for the shoulder.  At some point the errors in the robotic system will be fewer and less dangerous than human error, and the robotic car will be the safer alternative.  That point doesn't look to be very far off.

matter/energy transporters (as in Star Trek) would be even better.
The difference being the Bladerunner dual-mode has already been demonstrated.

em745 said at October 15, 2010 8:15 PM:

-"Extreme fines will just force people out of their cars"

Utter nonsense. Extreme fines would force people to drop their dangerous driving habits, like drunk driving, for instance.

-"The pilot watches for situations the FD can't handle."

MY point exactly. Thank you. Computers CAN'T do it all.

-"At some point the errors in the robotic system will be fewer and less dangerous than human error, and the robotic car will be the safer alternative. That point doesn't look to be very far off."

Again, YOUR opinion.

Robotic cars could theoretically be safer ONLY if you rid all roads of every human driver, bicycle rider, pedestrian, and hockey-playing kid.

Good luck with that.

Engineer-Poet said at October 17, 2010 7:30 AM:

Whoa, a troll.

Extreme fines would force people to drop their dangerous driving habits, like drunk driving, for instance.
License suspensions and jail don't do that, but you think fines will?  Whatever you're smoking, it's good.
Computers CAN'T do it all.
Your "reasoning" seems to be "computers aren't perfect, so they're no good even if they're better than humans".  Of course, you're a troll, stupidity is your raison d'etre.

It's going to be hard for a computer to match the skill of a Sullenberger.  Out-reacting your 75-yr-old grandma with sluggish reflexes, bad peripheral vision and hearing loss is another matter.

Robotic cars could theoretically be safer ONLY if you rid all roads of every human driver, bicycle rider, pedestrian, and hockey-playing kid.
As if humans deal with these things perfectly.  Let us know when that happens.  I won't hold my breath, but I encourage you to; it'll be one more improvement in the world.

em745 said at October 17, 2010 12:44 PM:

-"Whoa, a troll."

Poor, poor, POOR comeback... Worthy of a troll. :)

-"License suspensions and jail don't do that, but you think fines will?"

Four-figure fines? Having your car seized? Losing your license FOREVER? Yes, I do believe these would make even the most idiotic driver think twice about texting their honey bun while doing 70... or driving home after "only 2 beers."

-"Your 'reasoning' seems to be computers aren't perfect"

And your "reasoning" seems to be that computers ARE "perfect"... So my conclusion regarding your own "raison d'ętre" (with the "^" over the "e" BTW) is pretty much the same as yours.

-"Out-reacting your 75-yr-old grandma with sluggish reflexes, bad peripheral vision and hearing loss is another matter."

But these drivers (arguably) pose just as much of a threat as DUI's. There SHOULD be laws to address this. Unfortunately, the political implications involved with curtailing seniors' "rights" makes dealing with this problem all but impossible.

That said, *ANY* driver (regardless of age) with failing senses/reflexes shouldn't be behind the wheel, period... Just as I'm sure you wouldn't want to share the road with a robotic car with faulty/buggy hardware or software.

-"As if humans deal with these things perfectly"

(As if computers [could] deal with these things _AT ALL_.)

At least humans CAN deal with these things, "perfectly" or not.

A few more examples:

-How would a robotic car deal with a large animal (say, a moose) standing in the middle of a 2-lane road?

-Would a robotic car following a logging truck "notice" that dangling log that's an expansion joint away from falling off the truck?

-Will a robotic car be any better than a human at spotting and avoiding ice patches or water-filled truck ruts (i.e. hydroplaning hazard)?

-And suppose robotic cars do become commonplace... How long do you think it'll be before vandals looking for kicks start going around spray-painting the cars' sensor apertures?... Or covering them up with pieces of duct tape, or even chewing gum?


Finally, if your ultimate dream is to rid all roads of human drivers (seems to me like it is), how do you plan on convincing those (like me) who actually enjoy the act of driving to turn in their licenses?

Randall Parker said at October 17, 2010 7:08 PM:

em745,

Some people drive drunk after having their license suspended. I know a guy who went to jail for that. The mere threat of jail wasn't enough to stop him. He drove drunk many times before being pulled over. He had to get pulled over more than once to lose his license. Then he drove drunk some number of times without a license before being caught and sent to jail.

E-P isn't saying that computers are perfect. Computers only have to have a lower error rate than humans to save lives. They can certainly achieve that goal and already do that in production cars for smaller pieces of the total driving task.

I suspect you do not know how much existing car computers already do. Computers are already making better decisions than humans in anti-lock brake systems and in electronic stability control systems. Computers decide how much brake to apply and how much power to apply to each wheel.

A large animal in the road: They'd handle them the same way existing car computers do that with adaptive cruise control: with radar. In the Ford Taurus and other cars. Ford's going to bring it to their entire fleet.

All this incremental stuff keeps taking on more driving tasks. There are computer systems that park cars too. Lexus I think was first for that. Cars will soon warn about pedestrians and potential other causes of collisions. Then they'll apply braking to prevent various forms of collisions using radar and visual image processing to detect dangers. We'll get to fully automated cars thru many steps.

Engineer-Poet said at October 17, 2010 8:25 PM:

You'd have more humor potential if you weren't just plain stupid.

Four-figure fines? Having your car seized? Losing your license FOREVER?
So they don't pay (because they can't), they buy a $500 car and register it in a friend's name, and drive without a license (or insurance).  That's what they do today.  If you think you'd accomplish anything except getting the pols responsible voted out of office, you're even dumber than I thought you were (you're much too combative to be stoned).
And your "reasoning" seems to be that computers ARE "perfect"
Perfect?  Let's look at your reading ability here:

We'll have robotic cars that will drive more safely than humans, you can be sure.
At some point the errors in the robotic system will be fewer and less dangerous than human error

Not a word about perfection.  You're still rating room-temperature or below on the effective intelligence scale.

But these drivers (arguably) pose just as much of a threat as DUI's.
Do you propose to hit them with 4-figure fines, seize their cars, and take their licenses away?  What's sauce for the goose....
(As if computers [could] deal with these things _AT ALL_.)
The Google car is an existence proof that they can.
-How would a robotic car deal with a large animal (say, a moose) standing in the middle of a 2-lane road?
Um, by detecting the obstacle and stopping?
-Would a robotic car following a logging truck "notice" that dangling log that's an expansion joint away from falling off the truck?
So it doesn't "see" it until it falls, but the superior traction of tire on pavement means it can stop quicker than the log does.  This goes back to the above.
How long do you think it'll be before vandals looking for kicks start going around spray-painting the cars' sensor apertures?
How long do you think the car will take to detect the failure and refuse to move without a human in control?
Finally, if your ultimate dream is to rid all roads of human drivers (seems to me like it is), how do you plan on convincing those (like me) who actually enjoy the act of driving to turn in their licenses?
Sez the guy who wants to fine old drivers off the road instead of letting them use computer assistance.  Yeah, like we should listen to anything you say, hypocrite.

Randall Parker said at October 17, 2010 9:38 PM:

Obstacle detection: A computer can do this faster than a human. Computers can take in sensor data, process it, and turn around and send orders to actuators in milliseconds. Computers can act before human conscious awareness occurs.

As for dangling logs: I'm sure the more common sources of danger can be coded for. The really rare threats: Well, humans will be sitting in the control seat for some years to come. But gradually computers will become better at recognizing even the less common dangers.

Engineer-Poet said at October 18, 2010 3:55 AM:

Objects or holes in the roadway are fairly common, so it's certain that Google has already dealt with the issue.

em745 said at October 18, 2010 9:05 AM:

Randall:

-"Some people drive drunk after having their license suspended."

Which is why I added the 4-figure fines and car seizures. If THAT still doesn't work, throw them in a max security jail for a couple years.

Some people will argue that the punishment should fit the crime. I maintain that the punishment should fit its value as a deterrent.

-"Computers only have to have a lower error rate than humans to save lives"

And as I said before, this will only be possible--and of benefit--if the whole of "humanity" is removed from our roads.

As long as there are humans on our roads, be it drivers, cyclists, pedestrians, etc., driving will continue to be an UNPREDICTABLE endeavor... not unlike weather forecasting (and we all know how "accurate" computers are at predicting a hurricane's track).

-"They can certainly achieve that goal and already do that in production cars for smaller pieces of the total driving task"

With ABS, torque-sensing LSD and/or AWD, stability and brake assist, the on-board computers are only concerned with what the car's individual components (wheels, steering wheel, suspension, throttle and brakes) are doing. All of it involves basic, linear number crunching. The computers are still totally deaf, blind and dumb as to where the car is going in relation to its immediate environment. That stability assist program still hasn't a clue as to what lies ahead of that curve.

-"A large animal in the road: They'd handle them the same way existing car computers do that with adaptive cruise control: with radar"

Well of course it would stop. But what if the animal is standing there and won't budge? What if it's a fallen tree or a bale of hay blocking the road? Would the car simply stop there and wait? Or would it be able to scan for a safe (improvised) way through? Would it have the "presence of mind" to safely do a U-turn and search for an alternate route?

BTW, you haven't addressed my other points (ice patches, etc.).

-"There are computer systems that park cars too. Lexus I think was first for that"

If only parallel parking were a suitable yardstick for driving proficiency. BTW, Car & Driver tested the Lexus system a couple years back and they said is was SLOW and required driver input to operate correctly.

Ford's system (~2009) is reportedly a whole lot better (it's sonar based rather than video based, as with the Lexus). But again, parallel parking--that is, maneuvering at crawl speeds in a limited space, is night-and-day compared to navigating through traffic on a 4-lane street while doing 40+ mph.

-"Cars will soon warn about pedestrians and potential other causes of collisions. Then they'll apply braking to prevent various forms of collisions using radar and visual image processing to detect dangers."

But with these kinds of assisted braking systems (as with your "adaptive cruise control"), you're only altering the car's forward momentum. In other words, the computer(s) is processing data for one dimension of movement. It isn't doing (or processing) anything having to do with modifying the direction/vector of travel (2-D). Big difference.

-"Obstacle detection: A computer can do this faster than a human. Computers can take in sensor data, process it, and turn around and send orders to actuators in milliseconds."

Driving safety isn't only about "reaction time." Reaction time isn't what saved Flight 1549. Experience and improvisation did.

em745 said at October 18, 2010 9:22 AM:

The poet:

-"So they don't pay (because they can't)"

Can't pay the $5,000 fine? Fine. Go to jail for six months, do not collect $200. Simple, eh? Deterrents need to hit HARD (way harder than they do now) in order to be effective.

-"they buy a $500 car and register it in a friend's name"

Right... That's some friend to be risking an aiding and abetting charge. And there aren't that many clapped out $500 clunkers with valid tags on our roads nowadays. Odds are he'd get pulled over sooner rather than later.

(Who's smoking what now?)

-"That's what they do today."

Sure, criminals find all sorts of clever ways to circumvent laws. It's been that way for centuries. Is that a reason to abolish the whole criminal justice system??

-"Do you propose to hit them with 4-figure fines, seize their cars, and take their licenses away?"

Look up the legal definition of INTENT.

(What was that about room temp. and intelligence?)

-"The Google car is an existence proof that they can."

Better read that article again:

"WITH SOMEONE BEHIND THE WHEEL TO TAKE CONTROL IF SOMETHING GOES AWRY"

"more than 140,000 miles WITH ONLY OCCASIONAL HUMAN CONTROL"

"Google researchers believe the cars are already legal BECAUSE HUMAN DRIVERS CAN TAKE OVER CONTROL OF THE COMPUTER SYSTEM [if it] STARTS TO MAKE WRONG DECISIONS"

(Now, what was that about reading ability?)

When Google comes back saying that this über-car drove itself--THE WHOLE TIME, WITHOUT ANY "INCIDENT"--from the Bronx to Brighton Beach during a snowstorm, THEN I'll be impressed.

-"So it doesn't 'see' it until it falls"

A human will notice the danger ahead of time and put distance between him and the potential danger, making any "reaction time" argument secondary.

-"but the superior traction of tire on pavement"

You OBVIOUSLY don't live in the snow belt... Or will these self-drive cars only be operable in picture-perfect weather and/or road conditions?

-"How long do you think the car will take to detect the failure and refuse to move without a human in control?"

"Refuse to move?"... At a (green) traffic light, in the middle of traffic? LOL! End result: It's still screwing around with a technology that will be laughably easy for vandals to exploit. And we haven't even delved into the prospect of "zapping" (jamming/spoofing) a moving robot car's sensors with lasers, radar guns or all manners of ultrasonic devices.

-"Sez the guy who wants to fine old drivers off the road instead of letting them use computer assistance. Yeah, like we should listen to anything you say, hypocrite."

No, I'd rather see mandatory, yearly testing for all drivers beyond 60... And yes, I'm all for taking away an elderly's driving privileges if that means safer roads for all.

And again, self-drive cars capable of RELIABLY and SAFELY sharing the road with human drivers are still _DECADES_ away. So "computer assistance" won't be a viable option for a long time.

BTW, the puerile ad hominems are becoming quite boring... I'd expect a lot more from a "poet." ;)

Engineer-Poet said at October 18, 2010 10:30 AM:

The king of non-sequiturs hasn't taken the hint.

And as I said before, this will only be possible--and of benefit--if the whole of "humanity" is removed from our roads.
Let's use an all-too-common example to see why this claim is bovine effluent:
  1. A child runs into a one-lane street from between two parked cars.  A human-controlled car is approaching at 25 MPH from 35 feet away.  The driver takes 750 msec to recognize the obstacle and hit the brakes.  The car travels 27.5 feet before the brakes activate, and goes another 23 feet at 0.9 G deceleration until it stops.  Result:  the car over-runs the child by 15 feet.
  2. A child runs into a one-lane street from between two parked cars.  A computer-controlled car is approaching at 25 MPH from 35 feet away.  The computer takes 100 msec to recognize the obstacle and hit the brakes.  The car travels 3.7 feet before the brakes activate, and goes another 23 feet at 0.9 G deceleration until it stops.  Result:  the car stops 8.3 feet short of the child.  (Optional:  driver blares the horn to scare the pants off the stupid kid, perhaps teaching a lesson and preventing a recurrence in which s/he may not be so lucky.)
The computer makes humanity safer in such cases even when they both use the roads, QED.
Reaction time isn't what saved Flight 1549. Experience and improvisation did.
The next time a car loses all engines due to bird strikes at a couple thousand feet over New York and has to make an emergency landing, I'll personally pay you $10,000.  Situations such as a moose in the road can be handled by a 6-yr-old picking a menu option like "road impassable, turn around" after the computer performs the stop.

If you have anything else stupid to say, keep it to yourself.  You were already tiresome and your stream of psychobabble isn't getting any better.

Engineer-Poet said at October 18, 2010 10:34 AM:

Oh, I see you were at it while I was writing.  I see nothing worth a response; the exceptions needing human intervention are exactly the kind of things used in software development to provide test cases.  As soon as the error rate is substantially lower than humanity's, they'll be a big asset to public safety.

em745 said at October 18, 2010 12:51 PM:

-"A child runs into a one-lane street from between two parked cars. Yada yada yada"

Yet another one-dimensional example (i.e. no steering, just braking)... And again avoiding the all too real possibility of poor braking conditions (snow, sleet...), which would require STEERING out of the way, as well as braking.

Yes, such a system would make roads safer, and could actually be implemented in current cars. But until such a system can reliably STEER as well as brake a car, in ANY kind of weather, I'd hold off on the "QED."

-"I see nothing worth a response"

Translation: I'm stumped... You got me. (QED!)

THAT, and your constant barrage of infantile ad-hom attacks = textbook trolling. :)

Happy trails, my trolling poet. See ya when Skynet has taken over the world! ;)

Randall Parker said at October 18, 2010 10:03 PM:

em745,

With electronic stability control systems what happens is the computer decides where it thinks you are trying to go and overrides your choices to get you there. It does this by looking at the orientation of the car, your steering wheel, and brakes and it decides it can apply adjustments to get the car going in a direction you haven't succeeded in getting it to go.

That's not yet looking around at what is there. But adaptive cruise control is looking at what is approaching and responding appropriately.

There's still another step or two up from there. That's what Google's already doing on real roads and it was what DARPA autonomous vehicle competitors were doing 5+ years ago. But Google's taking cars into much more complex environments. It really is quite impressive that they are doing as well as they are with 2010 technology.

I see Google's biggest problems as:

- developing a big enough data set of successful miles driven to prove the cars are safer.
- coming up with a legal model for how to apportion blame when an accident happens.

Are you telling me that in 10 years that automated cars will not be safer than most human drivers? If so, then do you also believe that about 20 years, 30, 40, 50, and 100 years?

em745 said at October 19, 2010 10:00 AM:

-"the computer decides where it thinks you are trying to go"

That's a large overestimation of what really happens. The VSC keeps tabs on steering wheel angle, vehicle speed, and (thru accelerometers), lateral G and yaw rate. It's still basic number crunching. When the VSC's fuzzy logic does "decide" it needs to kick in, all it does is apply brakes to the appropriate wheel to vary understeer or oversteer, and in some cases reduce power to the wheels. There's still no direct action on the steering. In terms of sophistication, VSC is merely an incremental evolution of systems that have existed for years, namely ABS and TCS.

-"But adaptive cruise control is looking at what is approaching and responding appropriately."

Currently by braking only. Still no steering involved. When you apply the brakes on slippery, snowy roads, ABS will actually lengthen stopping distances, not shorten them (making any "reaction time" argument an academic one at best). ABS preserves SOME steering ability, which is why I keep saying that if you want these automated systems to be failsafe, they're going to have to allow for steering inputs as well as braking. (Again, this will require some SERIOUS situational awareness and super-fast number crunching, since you're no longer working in "1-D".)

-"But Google's taking cars into much more complex environments."

Someone will have to point me to Google's DETAILED test results. "More than 140,000 miles with only occasional human control" may sound impressive at first read, but it really doesn't say much. What percentage of those miles were on the highway? What were the traffic and weather conditions? And what exactly does "occasional human control" mean in terms of frequency and severity of the corrections required? Etc.

And that "Lombard Street" test? Again, sounds impressive. But is it really? Keep in mind that Lombard's "curvy" section is one-way and has a speed limit of 5 mph (wow!). So taken in that context, Lombard Street seems only mildly more complexe a task than Ford's parallel parking assist.

-"I see Google's biggest problems as:
- developing a big enough data set of successful miles driven to prove the cars are safer.
- coming up with a legal model for how to apportion blame when an accident happens."

Before anything else, they'll have to prove that their car can handle REAL-WORLD driving conditions, like driving hundreds of miles through dense city traffic during inclement weather... all WITHOUT "occasional human control." Drivers living in the snow belt have been doing that for decades, and most manage to do it without incident.

-"Are you telling me that in 10 years that automated cars will not be safer than most human drivers?"

I'm telling you that:

1. there won't BE any (production) automated cars on our streets in 10 years

2. for the 2nd/last time, being "safer" has more to do with merely cutting braking reaction times

3. for the third/last time, as long as humans are walking, cycling and driving on our streets, throwing a few automated cars into the mix won't do much to make them safer.

You want to increase safety on our streets? Pass legislation forcing _ALL_ new cars to have ABS, VSC, brake assist, and traction control. This could be done right now, and would be far more effective at reducing accidents than waiting decades for some self-drive car that in the end may or may not be RELIABLY operable.

em745 said at October 19, 2010 10:09 AM:

"2. for the 2nd/last time, being 'safer' has more to do with merely cutting braking reaction times"

Sorry, that should read "...being 'safer' has more to do THAN..."

robotic car? said at April 11, 2011 2:02 AM:

Robotically controlled cars are here. Sounds great. Or, maybe not. Frankly this is too much...too fast. Google is starting to remind me of The Matrix. I've often wondered what rush-hour traffic would be like without an entire city full of idiots driving their vehicles while yapping on cell phones, doing makeup, eating, driving drunk, asleep, etc. Can you imagine, as the light turns green, a line of thirty cars all rolling forward at the same time? Sounds great to me, but in the back of my mind is this little voice going...I'm sorry Dave...

Ok, here's my main objection...while it's possible that a well-designed computer system could drive better than me...like that would be hard to do...what if the software is compromised? There's just too many computer hackers in this world to trust myself on a highway full of massive steel, plastic, and carbon fiber robots driving at seventy miles an hour or more. Don't even get me started on the number of recalls that car manufacturers perform every year. If this happens, if robot controlled cars take over the road, I'm moving out to the country where all they have are dirt roads and a shop-n-go.

Post a comment
Comments:
Name (not anon or anonymous):
Email Address:
URL:
Remember info?

                       
Go Read More Posts On FuturePundit
Site Traffic Info
The contents of this site are copyright ©