The Research Of Mobile Robot SLAM Based On Particle Filters

The ability of simultaneously localizing a robot and accurately map its surroundings is a key prerequisite of truly autonomous robots. EKF-based SLAM algorithms suffer from two well-known shortcomings that complicate their wide application to large, real-world environments: quadratic complexity and sensitivity to failures in data association.This paper introduces the universal framework and theoretical model of SLAM, establishes the motion and observation model of AS-RF mobile robot, and enumerates related formulas about sampling the pose, updating the state, constructing and managing the map. We make use of a Rao-Blackwellized Particle Filter-based Algorithm, and establish a two-dimension simulation model. Changing the observation and control noises, the performance of FastSLAM is compared against EKF-based approaches. Simulation results demonstrate that the estimate errors of the pose and the landmarks calculated by FastSLAM are far less than the EKF-based approaches under the same noise condition. FastSLAM can produce more accurate maps in extremely large environments and in environments with substantial data association ambiguity.To avoid the inborn problem of the odometry whose angle error will accumulate ceaselessly with time growing, a mixed localization approach which fuses the odometry and laser information is developed. Hough transformation is adopted to extracted line features, which are used to match the consecutive scan data to correct the angle error of the odometry.Using VC++ platform,many experiments are performed based on AS-RF robot with proposed method. Mixed localization approach can enhance the precision of the constructed map. Experimental results demonstrate that the proposed method can successfully create a consistent map of our laboratory aisle environment while locate itself using the map.