Research On Low-latency Laser Synchronous Localization And Mapping Technology

In recent years,robots have rapidly penetrated into social production and daily life,The core technology of robot perception,Simultaneous Localization And Mapping technology,has become the focus of research.In many SLAM applications,especially in the field of autonomous mobile robots,the real-time performance of the algorithms is required.due to the high computational complexity of SLAM algorithms,running on conventional CPUs or embedded systems is slow and requires high-performance CPUs to meet its requirements,however,it is difficult to deploy high-performance CPUs on mobile robots due to size and cost constraints.What is more,robotic systems need to handle many other tasks,so sometimes robotic systems cannot run SLAM algorithms in real time.So there is an urgent need for a solution for autonomous mobile robots.To address this problem,we studied the most commonly used laser-based SLAM algorithm in the field of mobile robots in depth,and a laser SLAM accelerator is implemented based on the Zynq family of FPGAs,which is tested in a real environment with certain results.Firstly,the paper introduces the significance of SLAM and the status and development of domestic and international research,and briefly describes the basic framework of SLAM algorithm;then the key technologies of SLAM are introduced.Then the Gmapping algorithm is studied in depth,and experimentally analyze the Gmapping algorithm to determine the effect of the algorithm parameters on the performance of the algorithm and the time consumption of each part of the algorithm,so as to provide preparation for the subsequent hardware design.Based on this,an improved scheme for multi-threaded implementation of the Gmapping algorithm is proposed.Next,the accelerator is designed based on Zynq.A hardware-software co-design approach is used to map the most time-consuming part of the original algorithm to the PL part for implementation,and the other parts are implemented in the PS part.The implementation of the gas pedal is optimized by:(1)the interface part is implemented by DMA and RAM to reduce the CPU occupation and access time of the memory,so as to reduce the CPU burden;(2)the optimization scheme of map cutting and compression is proposed for the limited hardware storage;(3)the optimization scheme of simplifying the calculation of the weight is proposed for the complex calculation of the weights.Finally,experiments are conducted.The implementation results show that after improving the Gmapping algorithm in this paper,processing speed of the whole algotithm is improved by about 4 times compared to the original algorithm.The accelerator designed in this paper was implemented on a Zynq series development board,and then the board was loaded on a robot platform for running tests in a real environment.The test results show that the accelerator designed in this paper is able to achieve the basic design goals: in addition to significantly reducing the use of general-purpose computing resources,the hardware PL part is 5 times faster in processing speed compared with the general computing resources;the maximum error of the whole system in the indoor10m*10m test environment is 30 cm,the average error is 10 cm,which has certain engineering value.Finally,the research content of this paper is summarized and the outlook for the subsequent work is proposed.