Research On UAV Swarm Path Planning Based On Electromagnetic Environment Perception

As a key technology for the autonomous flight of UAV swarms,path planning aims to plan a safe trajectory from the starting point to the target point.With the electromagnetic environment of UAV swarms becomes increasingly complex,the path planning of UAV swarms should not only avoid collisions,but also avoid areas of strong electromagnetic interference.However,due to the fact that the global electromagnetic environment information is difficult to obtain in advance in an unknown environment,the trajectory planning algorithm based on the known electromagnetic environment map is difficult to apply.In order to adapt to the unknown electromagnetic environment,how to design an electromagnetic environment perception method that infers global information from local observations,and conduct track planning and control based on the perception results is the key to successfully realize the autonomous navigation of UAV swarms in unknown electromagnetic environments.Therefore,thesis studies the UAV swarm trajectory planning problem in an unknown electromagnetic environment,considering two aspects of electromagnetic environment perception and trajectory planning control respectively.The main research work is described as follows.1.Most of the existing methods for constructing electromagnetic environment maps are suitable for data defects caused by uniform sampling,and the application scenarios are limited.Therefore,thesis further considers the more general data defect situation:Overall data defects(block defects)caused by the existence of restricted areas(no-fly zones,etc.),and a generative adversarial network algorithm based on weakly supervised learning is designed for electromagnetic map construction.In order to improve the construction accuracy,a U-shaped Markov discriminator is used,and a generator is designed in combination with U-Net and residual network.The simulation experiment shows that,compared with the existing methods,the root mean square error of the construction results of the algorithm in the two kinds of defective data is reduced by 0.526 and 2.024 respectively,which improves the range and accuracy of the stand-alone perception environment.2.In order to avoid the area of strong electromagnetic interference during the cruising process of the UAV swarm and ensure the collision avoidance between the drones,thesis proposes a multi-agent depth strategy gradient algorithm based on electromagnetic environment perception,and uses multi-agent reinforcement learning to solve the problem of the swarm.optimal track.On the one hand,the electromagnetic environment perception method proposed in thesis is added to the model training,and the electromagnetic map is periodically constructed and used as model input to assist the agent in planning the flight path,thereby improving the learning efficiency;On the other hand,the collision avoidance within the UAV swarm is used as a security constraint,and the corresponding protection mechanism is designed based on the control obstacle function to correct the unsafe strategy.Compared with the traditional model-free reinforcement learning method,the method in thesis accelerates the convergence of the algorithm through the agent’s perception of the environment,and can realize track planning with a better path.At the same time,it avoids violating security constraints in the process of individual trial-and-error learning,and ensures the security of the swarm in unknown environments.Finally,the performance of the algorithm is verified by simulation experiments.3.In this paper,a hardware-in-the-loop simulation platform consisting of a quadrotor UAV,an optical motion capture system,and a universal software radio peripheral is built.The signal distribution modulated by the general-purpose software radio peripheral is emitted by a directional antenna as an electromagnetic environment map.Through the hardware-in-the-loop experiment The effectiveness of the algorithm is verified.