Path Planning for Unmanned Aerial Vehicles in Hilly Orchards Based on Deep Reinforcement Learning

Abstract

When drones operate in hilly terrain, multiple factors such as flight energy consumption, operation time, mechanical wear and flight safety restrict each other. Traditional path planning methods find it difficult to effectively balance these goals. This paper models the path planning of unmanned aerial vehicles (UAVs) in hilly terrain as a sequential decision-making problem and proposes a solution model based on deep reinforcement learning. The covering path planning problem is transformed into the generalized traveling salesman problem through discretization processing. The state features are jointly encoded by the graph attention mechanism and the long short-term memory network, enabling the unmanned aerial vehicle to learn the path selection strategy with the lowest comprehensive cost in three-dimensional terrain. The experimental results show that this model achieves better path planning effects than multiple comparison algorithms in three different map experiments, verifying its effectiveness in multi-objective optimization problems.