the earliest work dating back in the 1980s. Until recently, however,
these vehicles could only operate in few environment conditions with
frequent human interventions. For these reasons, DARPA chose to hold a
competition called the "Grand Challenge", whose rules required the
competitors to travel 150 miles completely autonomous through rough
desert terrain. While no vehicle finished the 2004 race, the 2005
challenge was won by Stanford University.
This talk will introduce Stanford's vehicle "Stanley", focussing on
the software system. Stanley utilizes a number of novel ideas on
interpreting laser-based 3D-scans with a probabilistic method, using
computer vision for long-range road finding, on real-time path
planning and adaptive speed selection. The talk will show how machine
learning was able to aid in solving the problems encountered.
Finally, the vision of autonomous driving in traffic will be discussed
and initial results on that topic will be shown.