The results from last year’s competition indicate that it is not about who has the biggest, most expensive, or most complicated design; the very simplest control systems did almost as well the most advanced. The team that wins will be the one that is ready to explore new possibilities and ignore existing preconceptions of how a self-driving vehicle should work. We are using a fresh approach to the Challenge, from the vehicle platform through the control algorithms.

Instead of modifying a sport utility vehicle intended for mild off-road outings, we are using a powerful and nearly indestructible military strike vehicle tailored precisely to the requirements of this challenge. Unlike many teams in this competition, we recognize the benefits a superior mechanical platform can provide. Even the best autonomous control systems will not be able to handle all situations, and inevitably the vehicle will stray off-road. The dirt roads of the Mojave may be traversable by a stock SUV or pickup, the terrain on either side of the path may be extremely rough - a situation a typical commercial vehicle is not equipped for. Having a durable and rugged base vehicle will allow us to overcome obstacles that would easily incapacitate our competitors.

To sense the environment surrounding the vehicle, we are using a versatile combination of sensors. We are integrating a GPS with dead reckoning, LIDAR, RADAR, vision, and tactile feedback are combined to provide the vehicle with information about its environment. The sensor systems are mounted on a stabilized gimbaled platform, allowing isolation from the pitch and roll of the vehicle. It will also give us the ability to aim the sensors appropriately to provide information about upcoming obstacles quickly enough so the vehicle can still react. These features will allow the vehicle to gather useful data while traveling at high speeds.

Our vision system will use variety of image recognition techniques which will allow us to develop a three-dimensional world map. Stereo image processing will be used for intermediate and short-range obstacle detection. New algorithms developed through research done here at Cornell will be implemented in hardware to provide dramatically improved speed and object resolution. In addition, motion parallax (a form of motion flow) will allow us to identify objects separate from the ground plane using sequential video frames from a single camera. By combining these two complementary techniques, the shortcomings of each will be compensated for. We will also employ texture recognition to classify obstacles and discern their properties. Distinguishing between penetrable and impenetrable obstacles will allow our vehicle to make more sophisticated path planning decisions.