Simultaneous Localization And Mapping For An AUV Based On EKF

Whichever applications for unmanned submarines in military or civilian area, navigation technology is always one of the most essential. Navigation and localization with high precision has been one of most critical issues for the safety and effective completion of missions of autonomous underwater vehicles. Since the underwater environment is extremely complex and the external sensors of what can be used are limited to only sonar, as well as information obtained has too much noise and interference, thus all of the intractable will make data processing much more difficult. The method of Simultaneous Localization and Mapping allows the vehicle use on-board sensors such as sonar to perceive the environment and extract useful information to build up an augmented feature map of the underwater environment online while simultaneously using this map to locate itself without the help of prior environmental information map. The development of AUV applied in civilian and military area promotes the research of SLAM algorithm in underwater navigation and localization.As the most actual navigation system is nonlinear, to this fact, we use the Extended Kalman Filter to solve the underwater SLAM problem. The SLAM approach is significant in small environment, the EKF algorithm is robust and effective to deal with these problems. However, if the exploration environment becomes larger, the EKF SLAM approach has an important limitation: the linearization errors make the estimation out of consistency.In this paper we generalize the theory of EKF-SLAM algorithm, describe the consistency problem and point out the EKF algorithm how to make the estimation out of consistency. To overcome this problem, an EKF-based algorithm which builds feature map by using a robot centered representation is applied to an open-frame AUV, it delays the composition of the previous pose and the current vehicle motion until the feature map and the motion have been refined by using new observations of the environment. Proposed method results in a better linearization. The features currently observed have a small uncertainty in the order of the sensor error. Simulation results show that the vehicle can locate itself with an improved precision and the algorithm can be a feasible way for the AUV's navigation.