Sensor fusion, situation awareness overcame Urban Challenge hurdles

The recent Defense Advanced Research Projects Agency (DARPA) Urban Challenge placed robot vehicles in an urban setting, letting them run with driver-driven vehicles and requiring adherence to California traffic laws. The competition took place at the former George Air Force Base in Victorville, CA.

No vehicles finished the first Grand Challenge in 2004, which was a drive down desert roads—where concern was mainly for what was in front of the autonomous vehicles. The second Challenge in 2005 saw five robots complete a desert course.

Before the recent race, Chris Urmson, director of technology for Tartan Racing out of Carnegie Mellon University, told me the Urban Challenge would be "an order of magnitude more complicated" than the Grand Challenges. With other vehicles, including robots, on the course, maintaining proper lanes and accomplishing parking tasks was now required—thus extensive sensing and processing data for situation awareness was added to the mix. Urmson's team's "Boss" Chevy Tahoe (below) took the top prize based on no traffic violations and the fastest time to complete three, 20 mile missions given to each vehicle on memory sticks supplied by DARPA.

Tartan Racing's Boss used a rotating 64-laser radar on its roof to project a 'wedge' of laser light sweeping a volume around the vehicle. This unit was augmented by other lidars on the corners of the front fenders and radar (four antennas in front of the grill).[PHOTO: Rick DeMeis]

Urmson went on to mention that technology from the team's previous Challenge entries was not directly used on the Boss, but rather "experience, know-how, and ideas were transferred." The team also developed software tools for quick analysis of vehicle behavior using graphic animations of the vehicle's sensor and data inputs during test runs. These graphics allowed the team to "see" what Boss saw as it drove so they could determine and how and why it responded to features in its environment.

Interestingly, the final Urban Challenge event almost ended for the CMU team before it started. While slotted to leave the "chutes" first, an RF interference problem, partially blamed on DARPA's Jumbotron monitor next to the start, damaged the Boss' GPS receiver. The screen was shutdown (to the consternation of the crowd), but team members were able to change out the GPS.

Knowing the way
Carnegie Mellon, as well as second-place Stanford and fourth-place MIT (Virginia Tech was third, University of Pennsylvania/Lehigh and Cornell took the last two slots but were not ranked) used inertial navigation, sensor, and encoder systems from Applanix to augment GPS data. (Five of the 11 final starters used the company's positioning technology, including fiber optic or laser ring gyros and accelerometers.)

Louis Nastro, land products director for Applanix, said the Challenge wasn't really an urban environment to them because of the lack of tall buildings to block GPS signals. "Only a GPS outage would cause a problem" for teams without any back-up for satellite navigation. Nastro added the company's mobile mapping GPS-based technology is even accurate enough to map roadway camber and other surface features. Tartan's Urmson pointed out to me the Applanix circular encoder on the rear wheel, which he said provided the accuracy needed at less than 1 mph to do low-speed and parking maneuvers (ABS wheel encoders were not accurate enough.)