Simultaneous Localization And Mapping For AUVs In Large Scale Environments

Autonomous underwater vehicles (AUVs) will undoubtedly become the maintools for marine monitoring and investigation in the future. Currently, AUVsnavigation mainly depends on the acoustic baseline with high comprehensive-cost,limited operation range. It is a bottleneck hinders wider application of AUVs. Thispaper aims to explore the problem that AUVs break away from the external supportand adopt the simultaneous localization and mapping (SLAM) method to achieveautonomous navigation in unknown environments. SLAM has been successfullyapplied in small scale environments, however, there are still three key issues whichneed to be conquered for autonomous navigation and localization of AUVs in largescale unknown underwater environments: the data association, the consistency and thecomputational complexity and they were studied in this thesis.In order to overcome the limitations of individual compatibility nearest neighbor(ICNN) and joint compatibility branch and bound (JCBB), a novel iterativeclassification matching (ICM) data association method is proposed for SLAM in largescale environments. ICM applies the quaternion approach to extract a least squaresmatching vector from the measurement-map feature sets which are successful in theinitial association to weaken the influence of the inaccurate vehicle pose estimation.This vector is then used to update the map feature set which is failed in the initialassociation. The updated measurement-map feature sets are taken as a group of newinputs to perform the data association again until the check stage for mean squareerror is satisfied. Experiments of simulation and Victoria Park dataset demonstratethat the proposed ICM method is an efficient solution to data association in large scaleenvironments.Addressing the inconsistency problem of SLAM algorithm in large scaleenvironments, the classical EKF-SLAM system is analyzed based on the theory ofsystem observability. The relationship between system observability and systemconsistency is concluded, and simultaneously the underlying reason of inconsistencyproblem is revealed for the SLAM algorithm. LCEKF-SLAM, a novel locallyconsistency-constrained EKF-SLAM estimator, is designed by introducing theconcept of local consistency (LC). The proposed LCEKF-SLAM algorithm is experimentally compared with the traditional approaches as well as the popularestimators at present both in simulation and sea trial. Experimental results show thatLCEKF-SLAM performs well in aspects of consistency, accuracy and computationalefficiency.In order to satisfy the demand of real-time application for AUV-SLAM in largescale unknown underwater environments, a combined SLAM method based on theidea of submap is proposed. The main steps include: the local submap buildingemploys the locally consistency-constrained LCEKF-SLAM estimator and introducesthe multi-constrains submap building strategy; the local submap joining adopts theefficient SEIF algorithm and introduces the D&C submap joining strategy to reducethe computation complexity of the whole algorithm. The simulation, experiments ofVictoria Park dataset and sea trials verify that submap-based combined SLAMmethod can largely reduce the computation complexity of AUV-SLAM, andsimultaneously satisfy the accuracy and consistency.Finally, this thesis is concluded by summarizing the work and and points outcontributions and future work.