基于障碍凸化的改进环流APF路径规划

为改善人工势场法（APF）在复杂障碍环境下的路径规划能力，提出基于障碍凸化的改进环流APF路径规划方法。通过改变APF方法斥力势场的方向，使其形成绕行障碍边缘的环流，改善了APF方法在平台-目标位置连线与障碍边界垂直时可能陷入局部极小值的问题。针对多边形障碍（凹、凸）与圆形障碍，给出包括障碍初始化与相交障碍集合迭代凸化步骤的障碍凸化方法，将复杂障碍空间转换成为凸障碍的空间，避免平台进入凹多边形或相交障碍产生的凹区域而无解的问题。在仿真环境中，对比了传统APF方法、未使用障碍凸化的环流APF方法、使用障碍凸化的环流APF方法的路径规划结果，并分析了不同障碍数量、相交障碍数量条件下的方法计算耗时。根据结果，使用障碍凸化的环流APF方法能够明显改善APF方法在复杂障碍环境下的路径规划能力，方法实时性好，每个障碍所需的单步计算耗时约为0.02~0.03 ms，能够支持平台在复杂障碍环境下的在线路径规划。

Improved circulating APF route planning based on obstacle convexification

To improve the path planning ability of Artificial Potential Field (APF) in complex obstacle environments, an improved circulating APF path planning method based on obstacle convexification is proposed. By changing the direction of the repulsive potential field into a circulation around the edge of the obstacle of the APF method, this design solves the problem that the platform may fall into the local minimum when the platform-target line is perpendicular to the obstacle boundary. For the polygonal obstacles (both concave and convex) and circular obstacles, the obstacle convexification methods including the initial obstacle convexification and the iterative convexification for intersection set are given, which transforms the complex obstacle space into the convex obstacle space, avoiding the problem of no solution when the platform entering the concave polygon or concave zone caused by intersecting obstacles. In the simulation environment, the path planning results are compared when adopting the traditional APF method, the circulating APF method without the obstacle convexification, and the circulating APF method using the obstacle convexification. And the time consumptions of different methods are analyzed under different obstacle numbers and intersecting obstacle numbers. The results suggest that the circulating APF method using obstacle convexification can significantly improve the path planning performance under complex obstacle environments. In addition, this method has good real-time performance that the time consumption of single step calculation for every obstacle ranges from 0.02-0.03 ms, enabling the online path planning in complex obstacle environments.