Development of an Autonomous Vehicle for the DARPA Urban Challenge

摘要

This paper describes the approaches taken by the authors in developing an autonomous vehicle (Trobo) to compete in the Defense Advanced Research Projects Agency (DARPA) Urban Challenge. The overall technical goal was to develop new technical solutions for situations where either a commercial solution is not readily available or prohibitively costly and integrate with affordable existing commercial solutions in order to produce a winning entry that exceeds expectations. This process resulted in the development of several new technical methodologies as well as significant improvements to existing solutions in the following areas: (1) A new integrated brake and gas controller with an independent emergency brake mechanism, (2) a new split-frame formulation for vehicle navigation, (3) new control strategies for vehicle turn maneuvers, (4) improved dead reckoning technique for operating without GPS, (5) a new computer vision algorithm that produces robust and reliable dense optical flow fields several times in a second, (6) a new image segmentation algorithm, and (7) user-friendly and intuitive vehicle operating procedures for non-technical users. This is an ongoing effort and this paper gives an overview of the above approaches and presents preliminary results from field tests and model tests.