Research On OODA Control Of UAV Based On Hardware And Software Cooperation

In the UAV scene,the fusion of sensor and big data will cause the exponential growth of data volume,and increase the pressure of UAV flight control hardware.At present,the computing power of UAV end-side hardware is insufficient,and the widely used method of offloading tasks to the UAV ground control station through image and data transmission equipment is also greatly restricted by the communication bandwidth and delay.In the future,it is necessary to explore new end-side UAVs’ strong real-time capability and high computing power.Therefore,this paper designs and implements an intelligent UAV system and introduces AI technology to improve the real-time processing capabilities of UAV end-side tasks.The research work of this paper is mainly divided into the following three parts:(1)Hardware design part: PHYTIUM FT-2000/4 CPU and Cambrian edge computing Si Yuan MLU220 accelerator card are selected to build an physical edge intelligent computer by the connection of PCIe Gen3 to provide intelligent computing power for the UAV end side.According to the UAV application scenarios,the OODA hardware channel connection is completed by using SDI,Ethernet,UART serial port and other data transmission methods.The portable power supply system is designed through interface conversion and voltage conversion technology.(2)Software design part: The OODA software layer is designed,and the functions of environment awareness layer and data abstraction layer are realized.This paper configures the end side computer network to realize the function of real-time receiving and sending each frame picture and file address by RTSP data stream.According to the experimental results of execution time,power consumption and stability of YOLO＿v3 and Mobile Net-SSD of MLU220,this paper chooses to deploy Mobile Net-SSD network,and completes 120000 training on GPU server.The final test accuracy is up to 90%.Finally,the target model is deployed in MLU220 and the offline operation function is realized.Considering the factors of security and efficiency,the data format is customized,and the OODA data stream transmission is completed by UART serial port.(3)System optimization part: This paper proposes the re-customization technology,and uses the region interpolation technology to offload the preprocessing task to the FT-2000/4 CPU,which greatly reduces the work delay of the intelligent computer.In addition,this paper also proposes the streaming technology,which divides the task into sub tasks and assigns them to each computing node.Combined with the self starting technology,it speeds up the OODA data streaming processing and improves the intelligent degree of UAV system.After many tests,the TPR(target presence rate)value of the 100% pure domestic intelligent computer built in this paper is over 80%,the delay of the first two links of OODA is only 0.78 s,and the power consumption is 26.8w.Compared with the implementation effect of NVIDIA Jetson TX series,the FPS value of star 4K image processed by the physical edge intelligent computer is 24,which strongly reflects its powerful AI computing power.In addition,the paper also simulates the real battlefield environment,and designs the demonstration scene of UAV vs robot dog.The successful result of the battle between UAV and robot dog fully proves the reliability of the intelligent UAV system and meets the predetermined requirements of the project.