Efficient Online Trajectory Planning for Fast Flight in Dynamic and Cluttered Environment

Abstract

This article proposes an efficient online trajectory planning system for autonomous aerial vehicles (AAVs) to navigate dynamic and cluttered environments. The system encompasses three components: trajectory prediction, path searching and trajectory optimization. In the trajectory prediction part, an adaptive multimode trajectory prediction method is proposed, which accurately predicts moving objects by superimposing multiple motion modes. In the path searching part, the obstacle-aware hybrid-state A* is proposed to improve the path searching efficiency. It uses the obstacle information in the map to reduce the number of extended nodes of the graph search algorithm. In the trajectory optimization part, dynamic obstacle avoidance is achieved by extracting the convex hull from the trajectory of the AAV and the predicted trajectory of dynamic obstacles, and using the optimal separation plane as a constraint. Finally, we generate safe and smooth trajectories by solving a soft-constrained trajectory optimization problem. Extensive experiments confirm the proposed trajectory planning system achieves the shortest computation time and flight time compared with state of the art methods in dynamic and cluttered environments.