Session 2: Autonomous Driving

Abstract

This session had four talks on autonomous driving, ranging in topic from path planning, detection and tracking of moving vehicles in the scene and classification of the terrain using a supervised learning method.

The first paper from Stanford describes path planning used by vehicles driving around in an urban environment such as was used in the DARPA Urban Challenge. The authors use a hybrid A* search on an evolving map that is built dynamically as it is created by the moving vehicle. Two heuristics are used: The cost to go to the goal is produced with using two cost functions. The A* trees are expanded minimally and from the best leaf node, a single Reed-Shepp curve is used to get to the goal. In addition, a multi-element cost function is used to smooth the path generated by the search. The method can run fast in a few milliseconds even in a complicated environment. The authors show results from the DARPA Urban Challenge.