The coolest gadget at CES wasn’t at any of the exhibition halls. It was in the parking lot. There, a robotic Tahoe SUV named BOSS, took participants on a ride along an obstacle course. Who made this vehicle and how did it come to be? Carnegie-Mellon University, with its major sponsor GM and partners Continental and Caterpillar built the car for last year’s DARPA urban challenge contest. And, they won! In this contest, participants were to negotiate three 20 mile segments in a simulated urban environment in less than six hours. The prize was two million dollars. BOSS’s average speed was 14 mph, but it frequently gunned to 30 mph, the maximum allowable speed.

DARPA, the Defense Advanced Research Projects Agency, the group that funded the invention of the Internet, has obvious interests in encouraging fully robotic vehicle construction. In the CES parking lot, a human-occupied pursuit vehicle followed BOSS, or was otherwise in eye shot, equipped with a remote kill switch. The chase car would randomly approach intersections in the Convention parking lot, to give BOSS more of a challenge. There’s also a kill switch in BOSS, which was manned by a Continental engineer.

I had the opportunity to ride shotgun - next to no driver! This was an eerie experience. There are several small displays mounted on arms next to where the driver would be. The main display is a laptop that shows an integrated view of what the car sees - other cars, pedestrians, intersections, obstructions, even the cones marking the course.

How does BOSS detect and map these objects? It has three types of active sensors: lasers (LIDAR), radar, and cameras. GPS is used for position determination. Continental provided the adaptive sensors - cruise control and radar, cameras for early departure warning, and self-sealing tires. There are 11 sets of laser systems, and 11 sets of radar, four in front. There are three camera systems. The dome on the pole in the middle front of the vehicle rotates at 10 Hz and houses 64 point lasers. Some of the cameras, especially the big one near the center of the windshield, are for documentation purposes and serve no other function.

In addition to many RTOS (real time operating systems) for various sensors, there is an impressive computer stack in the back. There are ten blades in a compact PCI chassis, each blade housing two Core 2 Duo CPUs running Ubuntu 6.06. This stack runs the non-deterministic algorithms. I asked one engineer from Continental if he considered using any Windows flavor. He let out a big guffaw, and smiled.

The RTOS runs the deterministic systems, which comprise the safety and control systems. RTOS is on the embedded controllers, and includes steering and braking. It was the software that provided the winning edge over the next closest competitor, Stanford University’s "Junior", as the teams used similar sets of sensors. The Carnegie-Mellon team had several "profiles" they could use. For the challenge, they picked the most aggressive mode - more like a race car driver than a standard soccer mom driver.

Thinking about the problem, you realize that there are many unconscious decisions we make while driving in an urban environment. How do I maintain spacing with the car ahead? Is she slowing down? Is that pedestrian about to jaywalk? Who got to the intersection first? Those decisions all have to be translated into software. It’s like - well, driving a car! We just don’t realize the complexity of the decisions and constant adjustments we make.

To program these systems, the engineers on the Tartan racing team (Carnegie-Mellon) set up a three layer approach in customized software (TROCS): planning, behavior, and control. So for instance, on a simulated one way street in the Convention Center parking lot, with an implied stop sign, BOSS is programmed to stop. For testing purposes, the systems engineers used more of a black box approach, and did not distinguish between the three operating system layers. They were only interested in performance.

During my drive around, some idiot hopped the course barricades, and ignored the two security guards shouting at him. This happened behind BOSS, so we didn’t see it. At the same time we were set to negotiate a chicane simulated by garbage cans. BOSS came to a rather sudden stop - seen in the vehicle. Only after the ride did we find out what happened. BOSS saw an obstacle, and in its much less aggressive driving mode, stopped, even though we didn’t see the course pedestrian.

For this year’s DARPA challenge in Victorville California, entries were expected to find their way through an urban landscape simulated by a deserted base that included other robots, 50 added civilian vehicles, four way stop intersections, and even find a specific parking spot in a very large lot. Unbeknownst to the participants, a DARPA plane circled overhead, gathering information, in addition to the ground based DARPA crews. Not following the California Driver’s Handbook earned "tickets," or time penalties. The Lexus LS 460L - the self-parking car - has nothing on these robots!

Though the sensor technology used is available now, most of the features demonstrated by BOSS won’t be available for another ten years or so. No word on what Carnegie-Mellon and its sponsors will do with the two million dollar prize.