An Efficient Dynamic Obstacle Perception and Avoidance Framework for Robust Real-Time UAV Trajectory Planning

The UAV trajectory planning is the foundation of various applications. In real-world environments, there are not only static obstacles but also dynamic obstacles, so the UAV needs to have dynamic obstacle avoidance capability. The existing dynamic obstacle avoidance frameworks have low perception efficiency and insufficient security. To address these issues, we propose an efficient dynamic obstacle perception and avoidance framework for robust real-time UAV trajectory planning. In the obstacle perception module, we use displacement vector to represent the motion of obstacles and propose a displacement vector correction method to improve the accuracy of point cloud cluster motion estimation. Then we establish Kalman filters for dynamic obstacles to predict their trajectories. A map including static obstacles and dynamic obstacle Kalman filters is provided for the trajectory planning. To improve trajectory safety, we propose the dynamic obstacle active perception yaw trajectory generation method, which pays sufficient attention to dynamic obstacles during flight. Finally, we compare our proposed framework with the SOTA methods in both simulation and the real-world environments. Our proposed method reduces the collision rate by 17.8-62.1<tex-math notation="LaTeX">$%$</tex-math> and exhibits high operational efficiency.