Search This Blog


Friday, May 30, 2014

Police, Pedestrians and the Social Ballet of Merging: The Real Challenges for Self-Driving Cars

Google (Photo credit: warrantedarrest)
Recently at a press event held to showcase Google’s research in self-driving vehicles, project leader Christopher Urmson said that the problems posed by driving on city streets are between 10 and 100 times more difficult than freeway driving. Robot vehicles confronted with other vehicles, pedestrians and bicyclists do seemingly random things, and the roadway can change at a moment’s notice.
By dramatically slowing the speed of its robot car – limiting it to 25 miles per hour – and by removing the human driver entirely, Google is attempting to simplify the problem as well as mitigate any damage that the machines might cause should they fail.
Mr. Umson said that when a car brakes at 25 miles per hour, “you have half the kinetic energy you have at 35 m.p.h.”
Even at more languid speeds, one person who believes that Google has undertaken a tremendous challenge with self-driving cars on city streets is John J. Leonard, a veteran Massachusetts Institute of Technology roboticist, who developed one of the basic navigation techniques being widely used in autonomous vehicles. Dr. Leonard was a key member of the MIT team entered inDARPA’s 2007 Urban Vehicle challenge, a contest for robotic vehicles sponsored by the Defense Advanced Research Projects Agency.
He has taken his camera to the streets in Cambridge, Mass. and Boston to hunt for situations that might be challenging for robot vehicles. These “edge” cases – unusual events that might be unexpected by the car’s sensors and navigational equipment – are potentially a huge stumbling block to safe driving, even if they are extremely rare.
He has not yet compared notes with Google’s researchers to see which of his challenging situations the Google car can already solve, but some of them are clearly driving hurdles that would be tough for the best human driver.
My personal favorite of Dr. Leonard’s videos features a driver who comes to a busy intersection with traffic coming by in both directions. The challenge is not only watching the partially obscured traffic coming at high-speed from the left, but the continuous line of traffic coming from the right which requires a social as well as visual ballet to merge. The driver must use his car as a wedge and hope that the oncoming driver will give way gracefully. (I wonder how programmers will learn to deal with computer “road rage.”)
In a second case, even though lights are green at an intersection, a uniformed police officer motions with one arm and then steps out into a crosswalk to stop traffic and make way for pedestrians.
In the third video, the driver must carefully keep an eye out for a double yellow freeway separator that has been obscured either by weather or roadwork so as to avoid oncoming traffic.
The challenge in the fourth video, in which pedestrians run out into the intersection after a light has turned green, is one that I believe Google’s software can already handle with ease. In the demonstrations the company has given, the software can efficiently track individual pedestrians and bicyclists and make allowances for erratic behavior.
Finally, there is a still photograph of a snow-covered avenue in which lane markings are entirely obliterated. This is a challenge Google has said it has not yet solved.
Google has said it has not yet solved the ability for one of its autonomous vehicles to be able to detect lane markers covered by snow.
John J. LeonardGoogle has said it has not yet solved the ability for one of its autonomous vehicles to be able to detect lane markers covered by snow.
Despite enumerating the remaining challenges, Dr. Leonard said he is impressed with the progress that Google has made so far, both in advancing existing navigational techniques and doing so while they have reduced the amount of computing resources necessary to navigate safely.
“I have mixed emotions,” he said. “I have amazing respect for Google, but I do worry about public misunderstanding of what has been accomplished.”
The problem, he suggested is the public may come to believe that the problem is closer to being solved than it actually is.
Enhanced by Zemanta

No comments:

Post a Comment

wibiya widget

Disqus for Surut Shah

Web Analytics