Predictive Artificial Potential Field for UAV Obstacle Avoidance

Abstract

A method to generate sequential obstacle avoidance trajectories for Unmanned Aerial Vehicles (UAV) is proposed. The proposed Predictive Artificial Potential Field (P-APF) method can generate smoother and less time-delayed avoidance trajectories in real time while maintaining the same high avoidance performance and low computational load as the conventional APF method. It is expected that the use of UAVs in transportation systems will increase, and the proposed P-APF method is highly effective in situations where multiple UAVs perform respective operations in the same flight envelope. The P-APF method generates a potential field and repulsion vectors based on current information and also on predicted future UAV trajectories and observed obstacles. Thus, since prediction is incorporated into the trajectory generation process, the proposed method can generate trajectories that start avoiding obstacles earlier than the conventional APF method. To evaluate the performance of the proposed method, simulations of a UAV avoiding static or dynamic obstacles were conducted using the proposed method and the conventional method. The results confirmed that the P-APF method can maintain safety and take evasive action with a smaller delay than the conventional method.