Munich, Germany -- The large red SUV turns around a corner and approaches an intersection. It stops, waits until another car has passed and then accelerates to cross the intersection.
The driver? None can be seen; the car is empty. Instead, an array of cameras, antennas and odd-looking sensors is mounted on the car's roof and on its bumpers. A stack of computers occupies the back seat. Carefully accelerating, the car disappears around the next corner.
This is a scene much like those that will take place during the Defense Advanced Research Projects Agency (Darpa) Urban Challenge on Nov. 3 at the former George Air Force Base in Victorville, Calif. There are 36 semifinalists in the running, though the number of teams will be whittled to 20 during the qualification stage in the last week of October.
The Urban Challenge, a research project of the U.S. Department of Defense, concerns itself with the design of autonomously driven vehicles. To make the driver dispensable, the vehicles are equipped with instruments that replicate the senses and brainpower of a human being.
This is far from a trivial task. In contrast to earlier Darpa contests, it will not be sufficient this year for the vehicles simply to find their way to a final destination. This time, contest rules demand that the vehicles navigate through an urban environment over a distance of 60 miles. That poses challenges such as merging into moving traffic, avoiding moving obstacles, finding the way through roundabouts and, finally, parking in a gap--all in accordance with California traffic laws.
More-sophisticated aspects of everyday driving--recognition of traffic signs, traffic lights and pedestrians--still are not on the list of required tasks, for the time being.
While the leader of each Urban Challenge team has to be a citizen and resident of the United States, many teams from Europe and Japan are participating. For instance, Volkswagen is contributing staff and technology to the Stanford Racing Team, which won the previous Darpa Grand Challenge.
Team AnnieWay is a spin-off of the Collaborative Research Center on Cognitive Automobiles, formed by the University of Karlsruhe, the Technical University of Munich, the University of the German Armed Forces and Fraunhofer-Gesellschaft. Team Berlin is a joint project of the Free University of Berlin, the Fraunhofer Institute for Intelligent Analysis and Information Systems, and Houston's Rice University. The CarOLO team includes students, academics and researchers from various institutes of the University of Braunschweig.
The vehicles will rely on a broad range of sensors to successfully navigate through the Victorville "urban jungle." Besides tapping GPS systems for navigation, the cars are equipped with optical cameras and a variety of lidar systems and scanners that jointly simulate a human driver's eyesight. The huge amount of input data is then fed into computers, which generate the signals that effectively steer the cars.
In practice, lidar--light detection and ranging--is the prevalent sensor type. Of the 36 teams, three-quarters use the infrared scanners of optical systems vendor Sick AG (Waldkirch, Germany).
But IR scanners alone are not enough. "A lidar system alone cannot distinguish among a car, a Coca-Cola can and a biker," quipped Ferdinand Dudenhoeffer, professor of automotive technology at the Gelsenkirchen University of Applied Sciences.
Most teams agree with that assessment and have rigged their vehicles accordingly. "We use two different types of laser systems, which, based on signal propagation time, compute a 3-D image of the surroundings," said Bernhard Rumpe, a professor at Braunschweig University and one of the managers of CarOLO. The group complements the lidar signals with radar systems. In addition, the CarOLO vehicle uses a camera for lane recognition.
"The laser system is not smart enough," Rumpe said. "The algorithms make the difference." The software, developed at Braunschweig University, very accurately discriminates between size and motion vectors of objects.
Despite all this, the CarOLO solution is not yet perfect, Rumpe admitted. "It still generates too many 'false positive' identifications, which have to be eliminated through plausibility checks," he said. The specific strength of his team is its multidisciplinary approach, he said, referring to the fact that the group includes experts from the university's mechanical engineering department and from two software institutes associated with the university.
In that regard, CarOLO resembles most of the other competing teams with strong university and research backgrounds. Team-LUX, however, has followed a completely different approach.
Robotic eyes take in the scene—with optical, infrared and radar sensors.
A joint project of Hamburg-based startup Ibeo Automobile Sensor GmbH and its parent company, Sick AG, Team-LUX relies solely on infrared sensors. It is testing a prototype of a new generation of laser scanners that uses four beams instead of one. Equipped with two scanners integrated into the front bumpers and one in the rear bumper, the vehicle stands out from the others in that no futuristic sensors are visible; the vehicle appears to be a normal car. Even the backseat--where all other participants' vehicles have installed an array of dedicated computers--is empty.
"No cameras, no radar, no ultrasound sensors,"said LUX marketing manager Tanja Müller. "We have managed to integrate the entire preprocessing into the sensors, reducing the amount of input data for the main computer. Thus, we get by with two small Pentium M-equipped Linux PCs in the trunk for the generation of the steering and speed control," she said. "The sensor system, derived from a brake assistant, is mature enough to detect motion vectors of obstacles and even the nature of the obstacle--if it has legs, it is human," she said.
Once the robot vehicles have done their "runoff," the organizers, as well as the participating teams, will evaluate the results. They will likely come to different conclusions, because their motivations differ. While Darpa sees the Urban Challenge as developing the technologies that will eventually enable vehicles to maneuver autonomously in high-risk areas, the automotive industry is more interested in how the technologies can improve traffic safety among cars driven by humans.
"I do not believe that we will see autonomous cars in public traffic for the next 20 years," said Dudenhoeffer. "But the cognition gained during the contest could be used for implementing features that increase traffic safety--for in- stance, it could do away with the 'blind area' for trucks. And, these technologies could provide safety backups in an aging society."
Automotive OEM Volkswagen, a major sponsor for several teams, has a similar point of view. "We want to find out which possibilities exist today. Our aim is not the vehicle that drives without a driver, but we want to develop more-intelligent driver assistance systems--for instance, adaptive cruise control systems with a 'follow to stop' function," said Volkswagen Technology spokesman Harthmuth Hoffmann.
"Our target is refining driver assistance systems," he said. "And we believe this is a trend throughout the entire automotive industry."
See related charts