Research And Implementation Of SLAM System Based On Global Optimal Distributed Estimation

Simultaneous Localization and Mapping(SLAM)has become a key technology to solve the problem of robot autonomous navigation in unknown environment because of its high positioning accuracy and the ability to obtain unknown environmental information.However,due to the changing environment of the robot,the traditional SLAM system with centralized structure has the problems of large computation and poor fault tolerance,which seriously restricts the actual realization and application of SLAM technology.This paper focuses on the existing problems of SLAM system,analyzes the advantages of EFK-SLAM and distributed structure,analyzes the distributed heading-aid SLAM system model,improves and optimizes the performance of SLAM system,and constructs SLAM system suitable for practical application.The main research completed as follows:Firstly,for the existing centralized and distributed SLAM system dynamic fault-tolerant performance is poor and large computation,aiming at to tackle these limitations,the heading information is introduced into the distributed SLAM system,and a distributed EKF-SLAM system based on heading assistance is built.This paper mainly introduces the distributed heading-aid EKF-SLAM system form four aspects: system modeling,performance analysis,algorithm implementation and platform construction,and the experiment verify the performance of the distributed heading-aid EKF-SLAM system.Secondly,aiming at the first-order linearization of the EKF algorithm results in a larger model error,which makes the estimation accuracy of the algorithm decrease and even causes the filtering divergence,and the system has the problem of suboptimal estimation when the curve or dynamic change is large,the fusion algorithm of the distributed system is optimized and improved.From the concept of information sources,the fusion algorithm utilizes the full information sources of the local measurements,local prediction and global prediction to achieve the global optimality.The experiment verifies the performance of the algorithm.Thirdly,in the fusion process of sub-filter,the global optimal estimation of the system is affected by the coupling of the state noise and the absence of state information,so the distributed global optimization algorithm is further improved.By introducing the state relation matrix between the local filter and the main filter,the distributed filtering equation is brought into the centralized filtering equation to obtain the sub-optimal fusion result of each local filter,then uses the suboptimal solution as observation feedback to correct the one-step prediction state and yield the optimal solution of the global filtering.The experimental results demonstrate the reliability and efficiency of the improved algorithm.