Research Of Mobile Robot Localization And Mapping Based On Sensor Information Fusion

As the mobile robot applications environment become more complex and unstructured, exploration of the unknown environment has developed into an important research direction and basic problem in the field of mobile robot research. Being significant foundation method of mobile robot's intelligent navigation and environmental exploration research, SLAM (Simultaneous Localization and Mapping) has attracted many scholars'attention in this area. Simultaneous localization and mapping, refers to the process of the mobile robot localization itself based on pose estimation and sensor data, and build incremental environment map at the same time. Since SLAM solutions can realize the real robot autonomous navigation, it has become the research focu in the field of mobile robot in the past years.Based on collecting and collating relevant documents and literatures, this article made further research on mobile robot localization and SLAM based on sensor information fusion. The main research work are as follows:(1) Robot localization method based on GP-SRCKF.Since kalman filter and its variants over-reliant on priori knowledge of system dynamics model and system noise covariance, these methods have been greatly constraint on estimation performance. This article proposes GP-SRCKF algorithm based on Non-deterministic system model, by introduce GP (Gauss Process) regression into SRCKF (Square-Root Cubature Kalman Filter). This algorithm does not need deterministic system model, it can get accurate system noise covariance by learning sample data in GP regression, and make real-time adaptive adjustments in filter estimation. Apply this method into single robot localization, the simulation results show that GP-SRCKF can effectively improve positioning accuracy.(2) Robot localization method based on PSO-PF. As sample dilution occurs when there are fewer particles in high likelihood region in tradional PF (Particle Filter), this artcile propose new robot localization method based on PSO-PF (Particle Swarm Optimization Particle Filter). By intorduce PSO into prediction stage of PF and real-time updating the landmark's latest obervation information, this method drive all particles move to high likelihood probability region and reduce the risk of sampe dilution. PSO-PF adopts self-adaptive resampling method, reduce the number of resamp by doing it only in necessary, this can significant reduce the risk of good sampe been deleted and improve the performance. Apply this method into single robot localization and multi-robot cooperative localization, the simulation results show that PSO-PF can improve positioning accuracy.(3) Robot SLAM method based on DDPF-DDF.In basic FastSLAM method, EKF (Extended Kalman Filter) always cause larger linearization error when linearization the nonlinear system model in estimation, resampling process in PF always cause problems such as sample dilution. This article proposes new SLAM method based on DDPF-DDF (Divided Difference Particle Filter-Divided Difference Filter). This new method adopts DDPF in robot pose estimation, adopt DDF in landmark location estimation and map refreshing. Since calculation the jacobian equations of non-linear movement model and observation model is not necessary, DDPF-DDF can get higher estimation precision, and reduce the probability of sample degradation by using self-adaptive resampling strategy. Apply this method into single robot SLAM and multi-robot cooperative SLAM, the simulation results show that DDPF-DDF get higher precision and better continuity.(4) Robot SLAM method based on improved SEIF.SEIF (Sparse Extended Information Filter) use Markov blanket in data association and ignore the indirect connection of robot pose with landmarks, which will lead to over-confident likelihood estimation problem. This article propose an improved SEIF method, through augment linear system in data association and gain conservative edge covariance estimation which are consistent with respect to the actual covariance obtained by matrix inversion. This improvement will greatly increase the reliability of data association since it can effciently accessing and maintaining consistent covariance bounds. Apply this method into robot SLAM, theoretical analysis and simulation results show that it reduce the uncertainty in data association and improve estimation accuracy.The last part of this paper summarize the research works mentioned before, explain the innovations and the main research achievements, and point out issues need further study.