Research On Simultaneous Localization And Mapping Based On Distributed Particle Filter

Autonomous navigation technology is the fundamental guarantee for vehicleworking steadily in an unknown environment and localization and mapping are thekey problems for this technology. Localizing need the information of the environmentmap, at the same time, map will be built based on vehicle’s pose, so it is necessary tolocalize and mapping at the same time. This thesis focuses the research onsimultaneous localization and mapping(SLAM) methods for vehicle based on LiDAR.To reduce the calculate quantity and to improve the tolerance of centralized particlefilter SLAM. The research focus of this paper is distributed particle filter SLAM. Theresearch content of this paper including:First, based on autonomous navigation technology, this paper introduced thebasic principle of particle filter SLAM. First, the definition of SLAM problem isrendered, then, the principle of SLAM and its key process is introduced. Finally, thisarithmetic is been proved by simulation result.Then An algorithm of simultaneous localization and mapping based ondistributed particle filters (DPF-SLAM) is used in this thesis as the theoretical basis,and the traditional distributed particle filter SLAM system is improved in two aspects:1) An adaptive important weight, in this method, important weight was generate bythe variance of the particle, which made the particles gets a better weight. Theexperiment result proved that the novel system is more robust and stable.2) An novelfusion method, in this paper, we proposed an novel fusion method in the main filter.This method considered both the innovation method and the number of effectiveparticle method which improve the reliability of the system.Afterwards, this paper utilized the idea of RBPF (RaoBlackwellized particlefilter)in distributed particle filter system and proposed the distributed RBPF-SLAMsystem, this system divide the status of SLAM to two parts, linear part and non-linearpart, and applied Federal Kalman filter to evaluate the linear part to reduce thecomputation quantity.Finally, this paper adopted the the idea of UPF to distributed RBPF and proposeddistributed RBUPF-SLAM. And applied inherit particle initialization method in newadded local filter to improved the accuracy and convergence of the system. Finally,the experiment result proved that improved DRBUPF have a better stability andhigher convergence rate.