Localization And Loop-closing Problems In Depth-fixed Navigation For AUV

AUV is an underwater robot without cable which can navigate itself in the complex undersea environment. Autonomous navigation and localization are the most manifestation for AUV, and are also hot problems in the research area. When AUV is moving, the simultaneous localization and mapping algorithm (SLAM) can sense the environment by on-board sensors such as sonar, extract useful information to locate itself without the help of prior environmental information maps, at the same time, build the feature maps of the environment.In this paper, we talk about the SLAM algorithm based particle filter(PF-SLAM).The basic algorithm exist deficiency, if the motion noise is bigger than the observation noise, it will cause the filter diverse. So we improve the algorithm. At the same time, loop-closing strategy is used in the AUV localization process, by actively re-entering the visited area, the error of the position can be reduced.We first reviews AUV motion model and observation model, and the PF-SLAM posterior can be factored into robot path posterior and landmark posterior conditioned on the robot's path. Then we discuss about sampling a new pose, updating the landmark estimates, calculating importance weights and data association and so on. In the data association, if the landmark is unobserved, the map will be enlarged, else the observed landmark will be updated. Based on the theories, we improve the algorithm, when sampling a new robot pose, the new proposal distribution will rely not only on the current control, but also on the most recent sensor measurement. Since the important weights are defined as the ratio of the target distribution over the proposal distribution, it must also be updated to reflect this change.In the actual environment, due to the accumulated error, when the AUV travels back to the visited terrain along different trajectories, it can not determine whether the region has been detected, so the loop-closing detection is needed. If the loop-closing is detected, the AUV current pose can be mended by the built maps. At the same time, in order to prevent the depletion of particles, we take the information entropy as the criterion, when it is smaller than the given threshold, AUV will stop following the loop and start to explore new terrain. Simulation experiments improve that the loop-closing can reduce the position error and build more accurate maps.