Autonomous Off-Road Driving in the DARPA Grand Challenge

Abstract

The Golem Group/UCLA team was finalist in the 2005 DARPA Grand Challenge, and traveled 22 miles on race day. The Golem Group was also one of the most successful teams in the 2004 Grand Challenge, traveling 5.2 miles on shoestring budget. We present the strategies, challenges, outcomes, and lessons learned from two years of autonomous vehicle develop- ment. Autonomous navigation in the off road environment is challenging problem, which requires the successful integration of many different sensors and systems. We discuss the integration of GPS, IMU, and odometry together with various models for vehicle dynamics. In particular, we find it useful to incorporate the non-holonomic property of the ground vehicle, but also point out situations in which this constraint is violated and how to compensate. We also explore our approach to selection of quantities to use for control, taking into account which pieces of the system are most sensitive to errors in position and pose estimates. Finally, we address the performance of our system, and future possibilities for improvement. I. INTRODUCTION In 2004, the U.S. Defense Advanced Research Projects Agency (DARPA) held its first Grand Challenge. A million- dollar prize was offered for the individual or team that could build an autonomous ground vehicle capable of traversing 150 mile off-road course in the desert in under 10 hours. The event was described as a field test to accelerate research and development in autonomous ground vehicles. No vehicle was able to travel more than seven miles during the first race, and second event was held (for twice the prize money) in 2005. We qualified for and competed in both races, completing five miles of the 2004 event, and 22 miles of the 2005 race course. A. Prior Work