I’ve been following the online AI class from Stanford.  Included in that was a video with descriptive commentary of the robot that won the DARPA challenge.  The video is also on youtube.  The ‘trick’ was to overlay very accurate but short range laser sensor data, with much less accurate but long range camera vision, to ‘infer’ what was further ahead than it could see with the lasers.

http://www.youtube.com/watch?feature=player_embedded&v=Q1xFdQfq5Fk