The invention discloses a DVL-assisted SINS robust on-moving initial alignment method. The DVL-assisted SINS robust on-moving initial alignment method comprises the following steps of step (1), acquiring real-time sensor data; step (2), establishing a vector apparent motion parameter equation; step (3), on the basis of establishing the vector apparent motion parameter equation, estimating a parameter matrix through adaptive robust Kalman filtering; step (4), establishing reconfiguration vectors based on estimated parameters; step (5), through an optimal base quaternion attitude determining method, computing the error angle between a determined attitude and a real attitude; step (6), setting an initial alignment moment as M and a real-time data acquiring moment as k, if k=M, outputting an initial alignment result and completing an initial alignment process, and if k