Team Cornell : Technical Review of the DARPA Urban Challenge Vehicle

Executive Summary Team Cornell’s 2007 DARPA Urban Challenge blends the team and technologies of a largely successful 2005 DARPA Grand Challenge with internationally recognized experts in probabilistic analysis, GPS, control systems and artificial intelligence to develop a vehicle capable of executing autonomous urban operations. Key state of the art technologies developed and integrated in Team Cornell’s entry include: 1) reliable, high performance vehicle automation using brake, steering, transmission and throttle control; 2) tightly coupled vehicle position and attitude estimation with time synchronized antennas, square root information filtering, and integrity monitoring; 3) an integrated sensor suite of laser range finders, cameras, and radars to monitor the environment; 4) a novel probabilistic approach which fuses the many sensors in order to track moving vehicles relative to Cornell’s vehicle; 5) a unique scene estimation algorithm, which condenses local probabilistic information to produce a lane-specific estimates of the environment used by all levels of planning; and 6) a hierarchical planning algorithm, with operational, tactical, and behavioral layers to achieve smooth local control despite switching between an expandable set of road behaviors and higher level mission goals. This document gives an overview of these key technologies, along with a sample of the experimental results.