Motion Planning in Urban Environments to Achieve Sensor Quality Metrics
Ryan Hurley, Rick Lind, Joseph Kehoe · 2010
Advancements in trajectory generation for autonomous flight through urban environments have made autonomous remote sensing within these obstacle rich environments a possibility. This paper illustrates the use of a 3-dimensional trajectory planning algorithm utilizing a random dense tree and motion primitives from a 3-dimensional version of the Dubins Car. Additionally, a remote sensing quality function is formulated that is based on the range and orientation of the vehicle with respect to known targets in the environment. This quality function is used to influence the trajectories generated for remote sensing to be conducted. An example demonstrates the trajectory planner can avoid obstacles and view targets using feasible paths that are sub-optimal solutions to minimize the cost of flight time while maximizing metrics of sensor quality. I. Introduction Advancements in micro air vehicle technology have produced a class of aircraft of appropriate size and airspeed to enable flight within urban environments. Traversing these obstacle rich environments will require fully 3-dimensional flight and trajectory planning will be critical in taking advantage of the significant maneuvering capabilities of these aircraft. However, traversing these environments is only one aspect of operation. Close range remote sensing using a micro air vehicle could produce invaluable information for operators. Close-proximity flight amongst obstacles presents challenges for traditional path planning techniques such as the implementation of waypoints. This method may not be suitable unless some guarantee of feasible maneuvering is provided between those waypoints. In trajectory planning, the incorporation of dynamically-feasible motions is typically treated in either a direct or a decoupled fashion. 1 In direct planning methods, such as optimal control, a representation of the vehicle dynamics is considered in the formulation of the planning problem and the optimal system inputs are resolved. While optimal trajectories are produced, this method is often unmanageable for realistic problems. Alternatively, decoupled methods implement a simple vehicle motion model to plan a reference path and then smooth the path to satisfy vehicle dynamics using methods such as feedback control. This method often exhibits tractable complexity properties but with a lack of optimality.