Achievements of the 2005 Grand Challenge Team

Site Visit: To earn an invitation to the semifinals event of the GC, the 120 initial teams had to demonstrate basic GPS-following and obstacle avoidance capabilities during a site visit evaluation by DARPA officials. During the initial site visit in May 2005, Prospect Eleven showed successful GPS adherence, but failed to evade several trashcans and was not selected as a semifinalist. However, the performance was promising enough that DARPA granted “alternate-semifinalist” status and scheduled a second site-visit evaluation at the end of August 2005. The three ensuing months of development and testing by the team brought Prospect Eleven to the point where it performed flawlessly at the second site visit and earned an official invitation to the semifinals one of three additional teams.

National Qualifying Event (NQE): The NQE was the semifinals event of the GC. It consisted of 5 runs over a 2.2-mile course. The course included a 100ft tunnel under which GPS signal was lost, rumble strips, parked cars, tank traps, hay bales, a simulated mountain pass and tire stacks. Each run was judged on time, evasion of obstacles, and completion of course gates. Though Prospect Eleven performed admirably on 3 runs, its poor performance on the other 2 demonstrated serious software reliability issues. The fifth and final run was essentially perfect, within 2 minutes of the course record and secured Prospect Eleven a spot in the final race.

Grand Challenge: Twenty-three out of forty-three teams at the NQE were selected to participate in the final Grand Challenge Event on October 8, 2005, in Primm, Nevada. Prospect Eleven’s performance at the NQE earned it the tenth seed. The race started smoothly, and Prospect Eleven appeared on schedule at the eight mile mark. Shortly thereafter, at approximately 9.6 miles, steering control became unstable and Prospect Eleven was disabled. It was determined that this was caused by a bug in the obstacle tracking code, which neglected to delete all references to past obstacles. Tracking the position of thousands of irrelevant obstacles that had accumulated over the course ultimately overwhelmed the computers, causing the failure. However, 9.6 miles represented a personal record for continuous autonomous navigation by Prospect Eleven and was farther than any team had traveled in the first Grand Challenge. Our overall performance in the Grand Challenge is therefore an accomplishment we are proud of.

Post-Grand Challenge: After the Grand Challenge, the bug was fixed by changing one line of code in three places. Subsequent testing with profiling tools ensured that there was no additional problem with our code. We decided to explore Prospect Eleven’s full potential, so we returned to the desert for a second try at the Grand Challenge course. Prospect Eleven was able to drive the majority of the course autonomously, successfully navigating two tunnels, multiple gates, and descending winding Beer Bottle Pass at night without any intervention. The vehicle headlights provided sufficient illumination for robust obstacle detection. Certain conditions were encountered on the course that were not present during the Grand Challenge, such mud and ruts in the road; these were manually avoided. Prospect Eleven also navigated the entirety of the 2004 Grand Challenge course in a similar manner, driving through the treacherous Daggett Pass at night. The cumulative successes of our post-Grand Challenge tests confirmed that Prospect Eleven is a robust and reliable autonomous vehicle with refined capabilities.