Research On Unmanned Aerial Vehicle Three-Dimensional Track Planning Algorithm

Unmanned aerial vehicle(UAV)have been widely used in civilian and military fields because they can replace people to complete complex tasks under certain circumstances.With the continuous expansion of the scope of use of UAV,the issue of autonomous flight of UAV has become a research hotspot in the field of UAV in recent years.The key to realizing autonomous UAV and improving flight safety lies in the selection of flight paths.Therefore,UAV trajectory planning has become the research focus of UAV autonomous flight.This paper takes unmanned aerial vehicles as the object,and conducts in-depth research on the problem of UAV trajectory planning.The main research contents of the article are as follows.Firstly,this paper describes the UAV track planning problem and sets the constraint conditions based on the UAV own performance parameters.The paper then constructs a UAV track planning simulation environment and establishes a terrain threat model,a static obstacle model,and a dynamic obstacle model.For global track planning,an objective function composed of four indicators is established according to the requirements of the UAV to perform flight tasks.To address the slow convergence speed of traditional particle swarm algorithms and the issue of falling into local optima,the paper proposes a spherical coordinate particle swarm optimization algorithm combined with the beetle antennae search algorithm.This improved algorithm directly uses the spherical coordinate system to constrain the UAV heading and pitch angles,and utilizes the beetle antennae search algorithm to avoid falling into local optimum values.Then,due to the many sudden threats that UAV may encounter during actual flight,relying solely on global reference tracks can lead to mission failure.Therefore,a local track planning algorithm is studied to enable UAV to urgently avoid obstacles.The problem of local track planning is solved using the deep reinforcement learning algorithm.The reinforcement learning model and classical reinforcement learning algorithms are analyzed,and a state space,action space,and reward function combined with artificial potential field ideas are designed for the local track planning of UAV.The local track planning algorithm based on the deep deterministic policy gradient algorithm is proposed.Simulation results demonstrate that the algorithm of deep reinforcement learning can effectively and quickly plan local tracks in dynamic and static obstacle environments,and the reward function of artificial potential field can guide the UAV to the target point at the early stages of algorithm training,thus improving the convergence speed of the algorithm training.Furthermore,the local tracks planned using this method are far smoother than those planned using only distance as the reward function.Finally,this paper proposes a hybrid track planning algorithm based on the improved particle swarm optimization and deep deterministic policy gradient algorithm.The improved particle swarm optimization algorithm performs well in global static environments,while the deep deterministic policy gradient algorithm is effective in dealing with sudden dynamic obstacles.The hybrid track planning method combines the strengths of both algorithms and is advantageous for rapid track planning of UAV in complex three-dimensional environments.