A new research paper from the USC Center for Robotics and Embedded Systems offers a 3D navigation system for Unmanned Aerial Vehicles (UAVs). The paper, titled Vision-Based 3D Navigation for an Autonomous Helicopter (PDF format), describes a vision system that uses several image processing algorithms including optical flow and stereo vision. The vision system is combined with a Probabilistic Roadmap path-planner to implement a 3D wandering mode. The UAV is able to plan a path through a typical urban environment, taking into account an existing 3D model as well us unexpected obstacles detected by the vision system. The researchers investigated such details as the optimum camera angle for optical flow based centering, egomotion compenstation, and control strategies. A Bergen Industrial Twin RC helicopter was used as the base for the UAV.