Robot car seeks to win traffic race
San Francisco (CA) - On the second floor of the West Moscone Center at Fall IDF 2007, escalator patrons were greeted with a specially modified 2006 Volkswagen Passat named "Junior". The Stanford Racing Team has hopes this car win the 2007 Grand Challenge sponsored by DARPA (Defense Advanced Research Projects Agency).
Junior has four doors, a hatchback, is diesel powered and, oh yeah, it's got an artificial intelligence (AI) controlling it. The throttle, brakes, steering, ignition, everything is controlled by the AI. This special vehicle is part of a DARPA research project with goals to develop the ultimate "smart car" for automated driving. While previous attempts have navigated similar vehicles around fixed obstacle courses without obstructions in motion, this new challenge take everything to a whole new level.
Mounted all around the car are all kinds of sensors and control input devices. These include a spinning, 360 degree range-finding laser which can provide an accurate topography of the surrounding environment. That data is augmented with specially mounted cameras which also provide a full 360 degree view. And there are stil more lasers and radar units mounted at various points around the vehicle. All of it is employed to make sure Junior can "see" everything necessary to complete the course.
When all of this data is processed through a custom developed computer system and logic algorithms, objects and motion are discerned. These are applied to the AI which then determines what is traffic, what is the road and what legal conditions there are around the car. This new legal stipulation exists because Junior must be able to complete a course with not only moving traffic, but it also must obey traffic lights.
To make Junior behave, a special team of twelve researchers, scientists, engineers comprised of faculty and students out of the Stanford Artificial Intelligence Lab (SAIL) and School of Engineering provide programming and hardware adaptation. The endeavor is comprehensive and the team has high hopes to build off their previous 2005 winning entry named "Stanley," which finished a similar race without moving obstacles or traffic laws.
The biggest challenges facing the team are translating the pixelated computer images and numerical range data into accurate 3D models which can be interpreted and responded to correctly. While this is something relatively easy for humans to do, teaching a car to do it is proving difficult.
The team will enter the Grand Challenge event on November 2, 2007, sponsored by DARPA. The team is funded by a host of private enterprises, including Intel, Google, Red Bull, MDV (Mohr Davidow Ventures), NPX Semiconductors, Applanix, and, of course Volkswagen.