backtop


Print 39 comment(s) - last by zebano.. on Oct 12 at 3:06 PM


  (Source: The New York Times)
Seven autonomous cars have driven more than 1,000 miles without human aid

New York Times report has outlined the details of a secret Google project to truly put the "auto" in automobile. The Mountainview, California-based tech company has tested seven cars that have driven without the aid of a human for 1,000 miles, and more than 140,000 miles with minimal human intervention.

The project was created by Google engineer and co-inventor of Google's Street View, Sebastian Thrun, director of the Stanford Artificial Intelligence Laboratory. In 2005, Thrun led a team from the university to win a $2 million Pentagon prize for designing the Stanley robot car, which drove autonomously for more than 132 miles.

For the current project, Google outfitted six Toyota Priuses and an Audi TT with advanced mapping technology and artificial intelligence software that can sense objects near the car and mimic human driver decisions. A passenger has been present to make minor adjustments, like when a bicyclist ran a red light during a recent test drive.

Google's motivation for the project, its engineers say, is to make the roads safer. "Robot drivers react faster than humans, have 360-degree perception and do not get distracted, sleepy or intoxicated, the engineers argue," NYT noted. Autonomous cars could double the capacity of our current roads by allowing them to be driven closer together, and, because of the decreased likelihood of a crash, could be made out of lighter materials, translating to better fuel efficiency.

The only reported crash, engineers said, was when one of the autonomous cars was rear-ended while stopped at a red light. Otherwise, one of the cars even drove itself down Lombard Street in San Francisco, what NYT calls "one of the steepest and curviest streets in the nation." The cars can be programmed with different driving personalities -- "cautious" is more defensive, while "aggressive" is, well, more aggressive.

Google has 15 engineers working on the project, as well as at least a dozen people with clean driving records hired to sit in the driver's seat as a precautionary measure.

But Google isn't the only party working on a self-piloted car. Researchers at Yale and NYU recently unveiled a human vision-based supercomputer called NeuFlow, which will aide in navigating cars in the future. A few years ago, DailyTech went for a ride in the 2007 DARPA Urban Challenge winner, a fully autonomous Chevy Tahoe.

Self-piloted autos are still years away from mass production, NYT notes, because computers have to become much more stable and less likely to crash, for one thing. Another obstacle beyond the technological aspect is the law. “The technology is ahead of the law in many areas,”  Bernard Lu, a senior attorney for the California DMV told NYT. “If you look at the vehicle code, there are dozens of laws pertaining to the driver of a vehicle, and they all presume to have a human being operating the vehicle.” Google has argued that, because there is a human being present to override any decisions of the automobiles, its experiments are legal. Lu agreed.

But there are those who believe that the technology could change society as profoundly as the Internet has. Google has not revealed a clear business plan for the new technology, but both Thrun and Google co-founder Larry Page have a shared commitment to increase the nation's highway safety and efficiency, sources say.

At least one thing is certain, autonomous cars, when perfected, would save more lives than any texting-while-driving bans.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

RE: Thanks NYT!
By tastyratz on 10/11/2010 10:16:14 AM , Rating: 1
Well said. Stability is in the hands of the programmer. It is usually not the computer that made an error but the programmer is still human...

One problem though? I cant see this working universally without manual control. You can control the lanes, you can control the lights, and you can detect pedestrians... but you cant control the weather. No matter what a patch of black ice needs the same amount of road and barrier control. No amount of computer control is going to fabricate traction so allowing tighter lanes etc. would only be dangerous.

Also don't forget special purpose vehicles and circumstances. A car towing a load or a truck with a full bed will drive differently and require special attention. I don't feel comfortable with a semi etc being on auto pilot completely as well.

I think it should be treated as and used only as a smart cruise control that's driver initiated based on conditions. If the driver feels the weather is good enough to autopilot and the roads are not changed because of this then fine. Otherwise no irobot or demolition man cars please.


RE: Thanks NYT!
By ArcliteHawaii on 10/11/2010 10:01:41 PM , Rating: 2
Your concern about the weather can easily be handled by a combination of road sensors and accelerometers that are patched into the main driving control system. The precursor (both hardware and software) to such a system already exists in anti-lock brakes to ensure the proper amount/frequency of application to prevent sliding, so actually this would be one of the easier pieces to implement.

While initially such an automatic driving system may be implemented/marketed as a smart cruise control, once the data comes back showing the system performs better than human drivers, insurance companies will jack the rates on vehicles that have manual control and people will gravitate to vehicles that have no manual control.

We don't hand fill boilers with coal to get hot water, despite the fact that automated gas boilers do blow up on occasion, although decades of design have made them extremely safe. Self driving cars will be the same, it's only a matter of time, and it seems that time will be sooner rather than later.


RE: Thanks NYT!
By Divineburner on 10/12/2010 10:23:31 AM , Rating: 2
In contrary to your view point, I can definitely see this working universally without manual control.

As of now, the cars are on autopilot for themselves only. Car A has the system, the system only cares for car A. Imagine this system working on a large scale and actually able to communicate to each other. The swarm intelligence alone (if mimic-ed correctly) should make it a terrific AI.

Now that this system only works on cars. Let's expand it to traffic lights that send signals to the cars, weather stations that broadcasts weather conditions to the cars real-time, GPS systems that are able to precisely locate each car and along with the dimension of the cars, create a detailed traffic map. Not to mention that the cars can make lightning-fast decisions and are able to pass that decision down to any car in the same lane or nearby lane almost instantaneously. Eg. Someone goes across the road. One car detects it via advanced shape and pattern recognition and brakes the car. At the same time, the decision is passed down and all the cars following immediately brakes, while the rest slows, and so on.

The technology for the recognition already exists and are used extensively in the movie industry. Look how far facial recognition had come. Imagine the time where such a system is implemented, such technology would be improved by leap and bounds.

Personally speaking though, I wouldn't jump into a closely-packed, high speed lane of traffic. If someone do so and gets killed, well, it's Darwin at work.


RE: Thanks NYT!
By tastyratz on 10/12/2010 12:17:22 PM , Rating: 2
To counter - facial recognition has come a long way but is also easily fooled with a simple printed photograph (or 2 offsets for dual cameras)

Pedestrian accidents aren't just suicide runs. It could be anything from people trying to make a crosswalk as its changing, falling off a bike on the side of the road, being pulled by a dog, etc.

Maybe someday it would be more acceptable but its not as simple as getting the cars out there, it would be a massive infrastructure change within the travel system.

There are a million unprogrammable unaccountable scenarios that would require manual intervention.

A computer might not be able to distinguish a tree swaying from a branch falling and a reasonable rate of speed would not be one that left much room for waiting till it across the road. The AI might also not be able to tell if a telephone pole is going down, or if power wires are across the road.
It has no prioritized sense of urgency when traveling to the hospital or family doctor for those in rural areas etc.

ABS/Traction control and similar systems detect when its too late. Antilock brakes detect sliding tires during decelleration, traction control detects power delivery greater to that of a given traction. Neither system detects a slide at cruising speed nor compensates for current or future conditions. By the time the system picks up on a problem you have already... LOST traction. If it was cruising at 60mph on an interstate for example and a patch of snow melted then refroze ahead you might see the ice but the car would not. Manual intervention would be what saved your life.

Also consider target markets - a vehicle complex enough to perform these duties with heavily invested AI as well as many additional sensors has a higher purchase/ maintenance cost. Roads with appropriate feedback are likely expensive and would greatly increase dpw overhead especially in smaller towns. Consider dirt roads and private roads which are not marked on a gps.

Don't get me wrong I think its an excellent idea, but I think as long as computing power does not equal that of the human brain we can not assume it can understand an infinite number of situations and interpret those with which it was not programmed for. In my lifetime? maybe the end. Till then its a smarter cruise control.


"I modded down, down, down, and the flames went higher." -- Sven Olsen














botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki