Testing driver skill for high-speed autonomous vehicles

Abstract

The 2005 DARPA Grand Challenge, a 212-kilometer race through the Mojave Desert, showcased the state of the art in high-speed, autonomous navigation of trails and roads. To win the challenge, a team's robot had to complete the course faster than any other robot, and it had to do so within 10 hours. Carnegie Mellon University's Red Team developed two robots, which used a combination of autonomous and human preplanning to become two of only four robots to complete the Grand Challenge. The robots used onboard sensors to adjust a preplanned route to avoid obstacles and correct for position-estimation errors. To be successful, teams had to develop innovative algorithms and systems - and rigorously test them to verify performance. The Red Team used the tests regressively to evaluate how unit changes in hardware and software affected the robots' overall driving ability