The Study On Simultaneous Localization And Mapping Based On Extended Information Filter And Consistency

Navigation technology of autonomous underwater vehicle is one of hot issuesthat researchers are working on currently. The robots’ itself localization is one of mostbasic and essential functions. But the simultaneous localization and mapping (SLAM)is the key to realize the true autonomous navigation for AUV in unknownenvironment. SLAM is the process of building a map of an environment andconcurrently generating an estimate of the robot pose, which is considered as the basicability to expore unknown environment.The SLAM has been one of the notable successes of the robotics’ navigationcommunity over the past twenty years. At a theoretical and conceptual level, SLAM isconsidered as a solved navigation problem, it has been formulated and solved in anumber of different forms. SLAM has also been implemented in a number of differentdomains from indoor robots to outdoor, underwater and airborne systems. The SLAMalgorithm based on extended kalman filter is the basic solution to SLAM problem.EKF-SLAM can get optimal solution of SLAM through first-order Taylor expansionof nonlinear function. But EKF-SLAM describes the uncertain information in SLAMby maintaining a covariance matrix. The computational complexity is proportional tothe square of the number of environmental features, which limits the application inlarge-scale environment. Afterward, sparse extended information filter was proposedto solve the problem, which is the information form of EKF. Through pruning theweak links between robot and features, all update formula can be implemented inconstant time, and the computational complexity is significantly reduced.However, owing to the SEIF’s sparsification strategy and its own linearizationerrors, the problem of inconsistency always exists in the estimator algorithms and is abit more complicated. In this paper, we employ a consistency-constrained method toimprove SEIF-SLAM algorithm. The method studies consistency from the perspectiveof observability, with preserving the SEIF’s sparseness. Due to one of the main causes of inconsistency in the linearized error-state system model is the Jacobians of theobservation model having higher dimension of observable subspace than that in thenonlinear SLAM system, in this paper we use the first-ever available estimates tocalculate filter Jacobians, which can make the linearized error-state system keep theidentical observable subspace with the nonlinear SLAM system. The method is calledthe “First Estimates Jacobian”(FEJ), which represents a better consistency in SEIF.Simulation results are shown FEJ-SEIF can improve estimation performance interms of accuracy and consistency by comparing with SEIF. Furthermore, theapplication of autonomous navigation with the FEJ-SEIF for an autonomousunderwater vehicle (AUV), C-Ranger, is verified by sea trial.