The invention belongs to the field of an underwater vehicle and discloses a robust BSLAM method for AUV terrain matching navigation. The method comprises the steps of (1), inputting depth sounding andinertia navigation data; (2), carrying out data preprocessing to obtain filtered depth sounding data; (3), establishing an attitude graph, specifically, computing weak data association, establishinga new subgraph, carrying out closed loop detection through terrain positioning, and then carrying out invalid closed loop detection; and (4), optimizing a rear end graph, specifically, computing consistency functions through utilization of clustered closed loop data, finding out the minimum values of the consistency functions in all classes, integrating data of key statuses of inertia navigation,and outputting a modified BSLAM track. According to the method, through processing and closed loop detection of the depth sounding and inertia navigation data, global optimization of the data is realized, timeliness and consistency of positioning and graph establishment are improved, robustness is high, and influence of noises of measurement data on navigation precision can be reduced.