GM Puts the First Robotic Vision System in a Production Car

Yesterday General Motors invited journalists to its Milford Proving Grounds in Michigan for what sounded like a mundane affair?show and tell with the new versions of standard safety features. But the tech behind those everyday safety features was what caught our eye: With minimal fanfare, GM has put the first camera-based robotic vision system into a production vehicle.

The new GM system uses a single forward-looking camera to aid lane departure and collision warnings?mainstream features already in most new cars. The system actively monitors traffic and road conditions within the camera?s 37-degree range of vision. The computer software sorts and then assigns geometric data to each vehicle within its line of sight and tracks it, checking 14 times per second for changes in position of nearby vehicles to decided whether they pose a threat. If the geometric data shows an object getting larger in the camera?s field of view, for instance, the robot system would interpret that the same way our brains would: The object is getting closer (and the faster it grows, the faster it and the car are moving toward each other). The system also watches for solid or broken lane markings as well as curbs and other obstacles.

Current systems use radar to give an approximate location of danger ahead, or use downward-facing cameras to track the lines on the road. This system represents the first baby step toward a car that watches the road in much the same way as the driver does.

This kind of robotic vision has been a long time coming. In 2002, the U.S. government?s Defense Advanced Research Projects Agency (DARPA) announced one of its typically aggressive goals: to build ground vehicles capable of fully autonomous driving by 2015. The DARPA Grand Challenge offered a $1 million prize to any team able to meet the mark, and at the first Grand Challenge event in 2004, none of the 15 roboticized vehicles finished the 7.36-mile course. A year later, a Stanford team took home the grand prize for finishing the course in less than 7 minutes.

These teams had to overcome the daunting obstacle of creating a computerized vision system that could identify and track obstacles quickly enough to navigate a car. Even six years ago, the limitations of computing power meant that the Grand Challenge teams could barely meet that goal. Today, the state of the art has advanced enough that the Pentagon has deployed Lockheed Martin?s SMSS unmanned ground vehicle to Afghanistan. At the military level, at least, DARPA met its goal well ahead of schedule.

What was Defense Department?grade technology just a few years ago will be available next year in the 2013 GMC Terrain. Expect to find that technology making its way into other GM vehicles as well. And while this tech could point to a new era, perhaps the best part is that the single-camera system is between two and five times cheaper than competitive radar systems.

Right now the GMC Terrain's system will warn drivers if the vehicle crosses a lane without signaling or following another car too closely, and a flashing and audible alert will sound it the computer thinks an impact is imminent. ?With these features we?re really just scratching the surface of what this system can do,? says Eric Raphael, the active safety manager on the program. ?In the future I can imagine active crash avoidance, street sign identification, and much more.?

Source: http://www.popularmechanics.com/cars/news/industry/gm-puts-the-first-robotic-vision-system-in-a-production-car?src=rss

autumnal equinox rob bell jaycee dugard meg whitman f8 f8 catherine the great