Research On Key Technologies Of UAV Flocking Formation And Obstacle Avoidance Control

In recent years,with the development and change of Unmanned Aerial Vehicle(UAV)and battlefield environment,the UAV flocking has attracted extensive research attention.As a complex system,the UAV formation flocking system has a large number of individuals,a complex environment,and a rapidly changing situation.Because the calculation dimension of the UAV flocking is too large,it is difficult to use graph theory and other methods to solve problems such as large-scale flocking control.Distributed artificial intelligence technology provides effective means for large-scale flocking air combat research,including the use of multi-agent systems and bionic intelligence-based cluster control methods.By designing the behavior and control strategies of individual drones,a large number of drones with limited capabilities can be brought together and interact to complete complex flocking tasks.This article takes the multi-UAV flocking formation control system as the research object,with a view to improving the obstacle avoidance intelligence and the environment adaptability of the multi-UAV flocking formation.The research work of this paper is mainly carried out from the following aspects:(1)For the UAV flocking and obstacle avoidance,the consistency between UAVs in the flocking is not good.Based on the study of the multi-agent flocking and its obstacle avoidance motion characteristics and motion model,based on the simplified UAV dynamic model,a multi-UAV flocking and its obstacle avoidance motion algorithm were constructed.Considering the fact that when a UAV flocking avoids obstacles,individual UAVs may fall into a local minimum and cannot avoid obstacles,and collisions between UAVs within the flocking may occur,an algorithm for collaborative obstacle avoidance for the UAV flocking is proposed.The algorithm improves the ability of collaborative decision-making and consistency when the UAV flocking avoids obstacles.The paper also analyzes the rationality of the algorithm,and verifies the effectiveness of the algorithm by building a MATLAB/Simulink simulation platform,and achieves the expected results.(2)Aiming at the various obstacles in obstacle avoidance and navigation in unknown and complex environments,the UAV flocking controlled by the flocking algorithm derived based on traditional rigorous mathematical theory.In this paper,a control method for intelligent obstacle avoidance and navigation of a UAV flocking is studied by combining the flocking algorithm and a single agent reinforcement learning algorithm.The flocking algorithm controls the UAV to perform flocking movements,and the reinforcement learning algorithm controls the virtual pilots in the flocking to complete intelligent obstacle avoidance and navigation.The environment detection information of the UAV flocking is used as the status information of the virtual leader,and dynamic feedback is formed between the virtual leader and the flocking.The simulation results achieved the expected results.(3)Due to the poor performance of traditional control algorithms in the field of collaborative decision control,and the research of multi-UAV collaborative systems mostly separates the decision-making and control layers of UAVs,this paper studies the decision control of multi-agent deep reinforcement learning of multi-UAV.Aiming at the problems of instability of reinforcement learning in a multi-agent environment,a multi-agent deep reinforcement learning algorithm with centralized training and distributed execution capabilities,a multi-agent joint proximal policy optimization algorithm,is proposed.The algorithm uses the weighted average of the state value functions of different agents in a coordinated environment to obtain a centralized state value function to enable multi-UAV to achieve better collaboration.The algorithm directly outputs the control instructions of the UAV,thus connecting the decisionmaking layer and the control layer.Based on the 6-DOF 12-state dynamic model of the fixed-wing UAV,the paper uses Tensorflow and Python to build a parallel network structure and algorithm system.This paper uses this algorithm to train multiple UAVs formations and obstacle avoidance in a dense random obstacle environment.The test results verify the advancedness of the algorithm in the obstacle environment,and the algorithm improves the intelligence and environmental adaptability of multi-UAV control.(4)Build a software-in-the-loop simulation platform for multiple UAVs,and verify flocking and formation algorithms on the platform.The paper initially builds a multiUAV software-in-the-loop simulation platform with ROS robot operating system as the distributed control system communication platform,PX4 as the flight control core,and Gazebo as the simulator.The speed,altitude,and heading angle commands of the UAVs in the flocking were obtained initially through the flocking algorithm.In the softwarein-the-loop simulation of the PX4 flight control,the PI control loop needs to be designed to make these commands work on the flight control.In the paper,a UAV model with attitude control loop is used for training to obtain a network model of dual aircraft formation,and then this model is directly transplanted into the software-in-theloop simulation environment.The software-in-the-loop simulation tests of both algorithms have preliminary verified the effectiveness and practicability of the algorithms.