Silhouette-Informed Trajectory Generation Through a Wire Maze for Small UAS

[摘要] Current rapidly-exploring random tree (RRT) algorithms rely on proximity query packages that often include collision checkers, tolerance verification, and distance computation algorithms for the generation of safe paths. In this paper, we broaden the information available to the path-planning algorithm by incorporating silhouette information of nearby obstacles in conflict. A silhouette-informed tree (SIT) is generated through the flight-safe region of a wire maze for a single unmanned aerial system (UAS). The silhouette is used to extract local geometric information of nearby obstacles and provide path alternatives around these obstacles. Thus, focusing the search for the generation of new tree branches near these obstacles, and decreasing the number of samples required to explore the narrow corridors within the wire maze. The SIT is then processed to extract a path that connects the initial location of the UAS with the goal, reduce the number of line segments in this path if possible, and smooth the resulting path using Pythagorean Hodograph Bezier curves. To ensure that the smoothed path remains in the flight-safe region of the configuration space, a tolerance verification algorithm for Bezier curves and convex polytopes in three dimensions is proposed. Lastly, temporal specifications are imposed on the smoothed path in the shape of an arbitrary speed profile.

[发布日期] 2018-06-25

[学科分类] 人工智能

[关键词] SHAPES;TRAJECTORIES;TRAJECTORY PLANNING;MAZE LEARNING;ALGORITHMS;TREES (MATHEMATICS);SMALL UNMANNED AIRCRAFT SYSTEMS;CURVES (GEOMETRY);PROVING;MATHEMATICAL MODELS;POLYTOPES;HODOGRAPHS;POSITION (LOCATION)