Path Planner for Dynamic Environments

Many modern approaches to autonomy use a layered planning technique where a conventional path planner plans a nominal desired path, and a reactive planner plans trajectories over a local region. In relatively structured environments with good localization, this layered approach has been proven to work well, as shown by TORC’s successful 3rd place entry in the 2007 DARPA Urban Challenge. The overall goal of this Phase I SBIR program is to develop new planning algorithms that allow operation in environments with dynamic obstacles, and do not depend on idealized sensors or accurate global localization. TORC proposes to evaluate two distinct approaches during this Phase I feasibility study. The first approach will use new Rapidly Exploring Dense Tree (RDT/RRT)-based techniques to plan among possible actions. The second approach extends our previous success in the Urban Challenge by creating time indexed costs based on forward simulations of object behavior. This augmented cost map will leverage the ease of representation and searching trajectories, and provide a method of including object motion.