DARPA Urban Challenge: Boss navigates city traffic

On 3 November 2007 eleven finalist teams ran autonomous vehicles through a mock urban course on the former George Air Force Base near Victorville, California. CMU's Boss (Tartan Racing) finished first and won $2 million, with Stanford's Junior in second place.

Event date: Published:
HistoryResearch

On 3 November 2007, eleven finalist teams ran autonomous vehicles through a mock urban course on the grounds of the former George Air Force Base near Victorville, California, in the DARPA Urban Challenge. Unlike the desert Grand Challenge, robots had to obey California traffic rules: merging, parking, stopping at intersections, and negotiating mixed traffic that included both robot and human-driven vehicles. Organizers announced the winner on 4 November 2007.

Boss takes the $2 million prize

Boss, a Chevrolet Tahoe entered by Tartan Racing (Carnegie Mellon University and General Motors), completed the route in a little over four hours and won the $2 million first prize. The vehicle fused LIDAR, radar, and cameras through sensor fusion and planned maneuvers with onboard motion planning software. William Red Whittaker led the CMU effort; Chris Urmson was a key technical lead on Boss. CMU’s published race summary puts the course at roughly 90 km.

Stanford placed second with Junior, a Volkswagen Passat built by the same group that had won the 2005 desert race with Stanley under Sebastian Thrun, earning $1 million. Virginia Tech finished third with Odin ($500,000). Six of the eleven finalists completed the full route within the time limit.

From desert to downtown

The Urban Challenge followed Stanley’s 2005 desert win and proved that autonomous driving could handle structured roads, not only open terrain. Teams combined mapping, prediction of other vehicles, and conservative planners tuned for safety. The event kept military and automotive sponsors investing in perception and planning years before deep learning dominated street vision.

Why it matters

Boss showed that layered sensors plus explicit planning could satisfy a strict driving rulebook in public view. The same CMU and Stanford lineages later fed commercial self-driving programs. The Urban Challenge sits in the timeline between classical machine learning robotics and the GPU-accelerated neural network revival that AlexNet would highlight in 2012.