A Data-Driven Dynamic Obstacle Avoidance Method for Liquid-Carrying Plant Protection UAVs

Abstract

Autonomous sprayer UAVs are one of the most used aerial machines in modern agriculture. During flight missions, some common narrow obstacles appear in the flying zone. These are non-detectable from satellite images and one of the biggest challenges for autonomous sprayer UAVs in farmland. This work introduces an obstacle avoidance architecture specifically for sprayer UAVs. This architecture has generality in the spraying UAV problem, and it reduces the reliance on the global mapping of farmland. This approach computes the avoiding path based on the onboard sensor fusion system in real-time. Moreover, it autonomously determines the transition of several maneuver states using the current spraying liquid data and the UAV dynamics data obtained by offline system identification. This approach accurately tracks the avoidance path for the nonlinear time-variant spraying UAV systems. To verify the performance of the approach, we performed multiple simulations with different spraying missions, and the method demonstrated a high spraying coverage of more than 98% while successfully avoiding all vertical obstacles. We also demonstrated the adaptability of our control architecture; the safe distance between the UAV and obstacles can be changed by specifying the value of a high-level parameter on the controller. The proposed method adds value to precision agriculture, reduces mission time, and maximizes the spraying area coverage.