Performance of BVC-based Obstacle Avoidance for a Quadrotor Relative to LiDAR Data Volume

Abstract

Among the various collision avoidance algorithms available, those based on "buffered Voronoi cells (BVC)" have been successful in terms of performance. In this approach, the control input can be determined easily using only the relative positions of the quadrotors. However, this approach, when applied to quadrotors in an environment with stationary obstacles, causes occasional "deadlocks", wherein the quadrotors can not reach the goal. Moreover, it was assumed that each quadrotor could use the relative positions of the obstacles at any time; however, how to obtain this information using the conventional method was not discussed. In this study, we propose an algorithm to tackle deadlocks. Furthermore, to tackle wall-like obstacles that can not be handled by conventional methods, we propose using all point-cloud data obtained by LiDAR as Voronoi seeds. Additionally, because the calculation cost increases as the number of Voronoi seeds increases, we verify the relationship between the calculation time of the control input and the behavior of the quadrotor with respect to the number of used LiDAR data points.