Research On Key Technologies Of UAV Formation Based On Environment Perception

As a new form of robot,the emergence of drones has promoted technological changes in traditional industries such as agriculture and forestry,broadened the technical means of the film and television industry and enriched the daily life of human beings.As a new t ype of intelligent carrier,it has been integrated into various fields such as military,civilian,and industrial,and has promoted the development of social economy,humanities and other aspects.However,most of the current UAV applications are simple applications of a single individual,often used to replace simple automation tasks,without the intelligence emphasized by a single agent,and rarely see the synergy of multiple agents.Therefore,current UAVs do not function well in the face of complex scenes and complex tasks.Therefore,current UAVs do not function well in the face of complex scenes and complex tasks.How to give full play to the advantages of multi-UAV formation collaborate to perform tasks will become an important research direction of UAV collaborative intelligence technology,including: three-dimensional space environment sensing ability,multi-UAV formation technology,drone autonomy recycling and integrated control.In order to study the pain points,this paper mainly studies a formation technology based on environment perception,and presents a complete solution for the formation of drone.In this paper,the whole UAV formation plan is divided into perception module,formation module and self-recovery module,and corresponding algorithms and implementations are proposed for each module,specifically: real-time three-dimensional perception of drones in low-altitude scenes,research and implementation of self-organizing network UAV formation fuzzy control method,multi-sensor fusion UAV autonomous dynamic landing recycling system.Firstly,the three-dimensional sensing capability of the UAV guarantees the safe flight of the UAV formation,and provides auxiliary environmental information for the intelligent decision-making and planning control of the UAV formation,and then uses the UAV formation technology to achieve high efficiency and stability.Finally the autonomous dynamic landing recovery technology of the drone is used to improve the autonomous completeness of the drone formation,so that the entire flight process from take-off to landing does not require human participation.The main work and contributions of this thesis are as follows:1.The three-dimensional sensing ability of the surrounding environment is the premise of safe flight and intelligent control of the drone.However,the practical three-dimensional sensing technology of the UAV requires simultaneous real-time,high-precision and stability requirements.Aiming at the above requirements,a real-time 3D point cloud segmentation method based on Squeeze Net and cyclic CRF is proposed.The algorithm firstly transforms the collected radar point cloud data into a standard data format similar to an image by using a spherical pre-processing method.Then,for real-time requirements,the basic network selects the Squeeze Net network architecture,and optimizes the basic network into a residual connection mode.The network outputs the point-wise segmentation labels,and then processes the CRF into a cyclic structure to achieve further point-wise refined classification.Finally get the label type of each point to realize the segmentation of the 3D point cloud.Finally,deploy the model in the ROS environment to achieve engineering applications.The experimental results show that the proposed network model can achieve high-precision point cloud segmentation,and meet the real-time requirements.The average frame processing of 85 ms can satisfy the stability requirement,the standard deviation of the running time of each frame is within 5ms in environment of ROS.The sensing method based on Li DAR point cloud data effectively avoids the lighting,weather and other factors,and can comprehensively cover various application scenarios of the drone.2.In order to solve the inter-cluster communication problem and realize the high-precision formation flying of the UAV cluster,a UAV formation using self-organizing network and fuzzy control technology is proposed.Firstly,Zigbee communication technology is used to build a self-organizing network communication system capable of automatic networking,automatic routing and dynamic maintenance.The customized data encoding format and the specific complete code and decode procession ensure efficient and reliable self-organized communication networks.The relative position information between the unmanned aerial vehicles in the cluster is solved by processing the GPS information of the cluster.Then the mathematical model of the "long machine-smashing machine" formation is modeled.The designed formation navigation algorithm is used to convert the relative position information between the drones into the three-dimensional position information under the mathematical model.Finally,the fuzzy PID controller uses the obtained three-dimensional position information to calculate the control amount on the three-dimensional coordinates of the rigid body,and then realizes the high-precision formation control of the UAV cluster formation.The experimental results show that the UAV cluster has the ability to automatically network and realize communication.It can fly in accordance with the specified formation.The fuzzy PID controller improves the dynamic response and steady-state accuracy of the system.The formation error is within 0.3m.The problem of networking communication of the UAV cluster is basically solved,and the requirements for high-precision formation flight of the UAV cluster are met.3.In order to improve the efficiency and completeness of the execution of the UAV cluster formation task,to make the UAV can complete the autonomous landing recovery and reduce human intervention in the all-weather and multi-scene scenario,this paper propose a multi-sensor fusion UAV dynamic landing recycling system.The drone can perform landing recovery independently and efficiently even in the harsh scene where GPS information is missing.First of all,the system will use UWB equipment to realize the three-dimensional positioning of the drone,and use the three-dimensional spatial positioning data to guide it roughly to the vicinity of the visual sign of the auxiliary landing.The visual processing algorithm is then designed to quickly and accurately solve the three-dimensional information of the drone relative to the visual tag.After a complete analysis of the dynamic uniform velocity model of the landing action,kalman filter is used to estimate and simultaneously correct the spatial information in the horizontal direction to improve the landing accuracy.Finally,a PID controller based on position control is designed.The filtered relative three-dimensional information is used to realize the high-precision dynamic autonomous landing control of the drone.The experimental results show that the drone can be correctly guided to the vicinity of the landing visual label,and the final landing accuracy is within 5 cm.The way UWB and camera sensors are combined is better than the way only GPS information is used,and the final landing accuracy is improved.At the same time,it can also meet the landing recycling requirements in the special scene without GPS information.The proposed recycling system basically covers all the scenes under the weather,and can complete the requirements of high-precision autonomous landing recovery of the drone.At present,application development based on drone-related scenarios is in a stage of steady and continuous development.The research work of this paper focuses on the three-dimensional environment-aware technology,cluster formation technology and self-recovery technology in the field of UAV formation technology,and gives the basic solution for UAV formation,which initially solves the relevant problems in the field of UAV formation.The problem lays the foundation for the application of multi-machine cooperation in drones.