Research On Simultaneous Localization And Mapping Algorithms Based On FastSLAM

Simultaneous Localization and Mapping(SLAM)is a key technology in the field of mobile robotics,which is crucial for achieving completly autonomy of robots.This thesis studys the Particle Filter-based FastSLAM algorithm based on a project of the National Key R&D Program(2018YFB600605),adopts the Clone Selection Algorithm and improved Particle Swarm Optimization Algorithm in order to solve the problems of particle degradation and insufficient estimation accuracy,completing the validation of the algorithm on Matlab platform and field environment.The research of this thesis are as follows:(1)The relevant theories and models of SLAM are studied.Firstly,the system model used in robot SLAM technology is constructed,the environment map is described,and the theoretical derivation of the map model is carried out.And then,the basic principle of FastSLAM algorithm based on Particle Fltering is studied,and the four processes of the algorithm are analysed.(2)Aiming at the particle degradation problem of FastSLAM,a resampling strategy based on Cone Slection Agorithm is adopted.Firstly,a particle reorganisation process is designed before particle initialisation to increase the diversity.Secondly,an adaptive clone number is designed,which can make the small-weight particles retained and increase the number of effective particles.Finally,an adaptive variation equation is designed,which can adjust the variation magnitude according to the degree of difference of the selected particles.Validated with Matlab platform,the experiments showed that the FastSLAM algorithm based on clone selection resampling significantly increases the number of effective particles and improves the particle degradation problem.(3)Aiming at the problem of insufficient accuracy of FastSLAM,a particle swarm optimization algorithm improved by Gravitational Search Algorithm is adopted.Firstly,the adaptive gravitational constant is designed for the Gravitational Search Algorithm to improve its optimization-seeking ability.Secondly,the speed and position updating methods of both Gravitational Search Algorithm and the Particle Swarm Optimization Algorithm are combined,effectively balancing the local search and global search,preventing particles from falling into local optimum.Finally,the particle fitness function is defined according to the robot's observatios.Simulation experiments showed that the accuracy of both the pose estimation and the landmark estimation of the FastSLAM algorithm based on gravity-improved particle swarm optimization is improved.(4)The clone selection resampling and gravity-improved particle swarm optimization are introduced to original Gmapping algorithm,field experiments showed that the improved algorithm could build complete and clear map with high estimation accuracy.