Research On Robotic SLAM Algorithm In Indoor Unknown Environment

Simulation localization and map building(SLAM)is a process in which a robot uses its own sensors to provide environmental information to construct a consistent representation of a map,and determines the position and pose of the robot based on the map.Starting from the algorithm of robot trajectory planning,localization and map construction,this paper mainly studies the SLAM algorithm of robot as follows:For extended kalman filtering(EKF)algorithm,because of the state and measurement model of Jacobi matrix to nod at each time step,according to the latest estimates,state estimation,cause the EKF algorithm state linearization of nonlinear system model,nonlinear SLAM system has higher than the actual number of observable subspace of the problem,put forward the improvement can view EKF algorithm,the first available estimate of each state variable to calculate jacobian matrix of the filter,so as to ensure the observable subspace of the state error of system model,with the actual nonlinear SLAM system observable subspace with the same dimension.When the covariance of measurement noise is unknown,the EKF algorithm uses the initialized exponential moving average estimation to measure the covariance of measurement noise.Experimental results show that the improved EKF algorithm is more accurate than the traditional EKF algorithm.In the RBPF algorithm,the Gaussian distribution of the odometer information is taken as the proposed distribution,and a large number of sampling particles are needed to ensure the positioning accuracy,which leads to the increase of calculation amount.Therefore,an improved RBPF algorithm is proposed.The annealing coefficient is used to adjust the proportion of observation information and odometer information in the proposed distribution so that the proposed distribution is closer to the target distribution.ICP is used to match the lidar scan data,and the matching result is used to replace the odometer reading with large error,which can effectively reduce the number of sampling particles and improve the efficiency of the algorithm.Aiming at the problems of high entropy and large error in raster map construction algorithm,two improved algorithms are proposed.Based on the improved algorithm of inverse sensor raster map,the prior occupancy probability is combined with the measurement information of lidar in the Bayesian framework to reduce the entropy value of raster map,and the result of ICP data matching is taken as the pose variation of robot,so as to improve the accuracy of raster map.A raster map construction algorithm based on branch-and-bound algorithm is proposed.The branch-and-bound method is applied to locate the robot position,and the one-dimensional Gaussian model is applied to transform the noise data into occupant information along the measured distance of the measured ray,which reduces the entropy value.