The Research On Navigation Algorithm Based On Particle Filter For AUV

Autonomous Underwater Vehicle (AUV) is one of current hot issues thatresearchers are working on recently. Due to intelligence, mobility, and autonomy,robots can complete tasks such as detection, adventure and operation, in verycomplicated and dangerous environment. The ability of autonomous navigation isthe key the precondition for AUV to achieve underwater tasks in unknownenvironment. To truly achieve autonomous navigation, Simultaneous Localizationand Mapping (SLAM), become the problem that must be settled urgently in the robotfield.Whether the robot can truly autonomously navigate itself depends on the abilityof simultaneous localization and mapping (SLAM).SLAM is the key issue to realizethe AUV autonomous navigation problem.And it reflects the robots’ perceptionability and intelligence. SLAM is the problem of estimating the position of the robotand simultaneously constructing the map where the robot is moving based on thecontrols input and observations. In recent years, many researchers have made theremarkable progress in indoor, outdoor, underwater environment have However, thereare still a lot of problems need to solve.This paper first introduces the development of AUV and its present researchstatus; Secondly, this paper introduces several wide-applied SLAM algorithms, thatwe also need to know for the paper. Based on establishing the system model, theprinciple of SLAM algorithm and process are introduced in detail, and point out thatthese algorithms are the problems which need to be solved are pointed out.Aiming at solving the important limitations of standard FastSLAM, thederivation of the Jacobian matrices and the linear approximations of non-linearfunctions, the paper proposes an unscented FastSLAM algorithm based on theparticle swarm optimization based on the particle filter. This algorithm combined unscented particle filter(UPF) and unscented particle kalman filter (UKF) to estimatethe robot poses and landmarks more accurately compared to the standard FastSLAMalgorithm. This avoids introducing the error from linearization and derivation ofJacobian matrices and improves the estimate accuracy. In addition, to prevent theparticle degeneracy and impoverishment, we used particle swarm optimization tooptimize particles after resampling. Choose all the positions of particles to presentthe target particle, and PSO was executed to update the robot position.At last, to verify the effectiveness and accuracy of proposed algorithm, westudy the simulation experiments and the application of PSO-UFastSLAM inC-ranger AUV navigation via sea trial experiment. Simulation results reveal thatPSO-UFastSLAM shows better accuracy in terms of estimation of robot and featurepositions than other algorithms such as UFastSLAM algorithm and standardFastSLAM and meet the requirements of the feasibility and availability for SLAM.