Improved Research SLAM Algorithm Based On Particle Filter

As the requirements on robots’ autonomy are becoming higher, more attention are gradually been paid to autonomous navigation these years. In order to provide robots with truly autonomous capabilities, two essential technologies-localization and mapping-should be seen as one integral problem to be solved, which is called simultaneous localization and mapping (SLAM). SLAM problems are studied in this dissertation and three improved SLAM algorithms are proposed to deal with the calculation of importance proposal distribution, particle degeneracy and particle impoverishment. The main contributions are as the following:Firstly, several models of the robot system including the motion model, observation model, map model, noise model, and the like are established. Based on these models, both the theoretical foundations and implementation steps of two traditional SLAM algorithms (EKF-SLAM, FastSLAM2.0) are introduced and the results of relevant simulations are compared and analyzed for evaluating the follow-up improved algorithms.Secondly, based on the compensated extended Kalman particle filter, an improved SLAM algorithm, CEKPF-SLAM, is proposed to compensate the linearization error resulted from the extended Kalman Filter. Thus, the importance proposal distribution will approach the true posterior probability density distribution to improve the accuracy of the algorithm. The validity of the CEKPF-SLAM algorithm is verified by simulations.Thirdly, on the basis of CEKPF-SLAM, two improved algorithms, FBPF-SLAM and WOCPF-SLAM, are proposed to improve the resampling method. A pre-processing of sorting, fission and normalizing (SFN) is implemented in FBPF-SLAM to slow the impoverishment of particles while the optimal combination is applied in WOCPF-SLAM to calculate the weight of particles so that fewer particles are removed and then the variation of the particle set is increased. As a result, FBPF-SLAM’s robustness is improved and WOCPF-SLAM’s robustness and accuracy are improved, which is verified by simulations.Lastly, the conclusions and the future research work are discussed.