The invention provides a vehicle-mounted navigation calculating method based on cascade adaptive robust federal filtering, and belongs to the technical field of vehicle-mounted navigation. The vehicle-mounted navigation calculating method comprises the steps: a navigation calculating equation and an error model are built; a three-stage weight reduction function is constructed by adopting the residual error as the statistical magnitude, and measuring noise containing a measuring abnormal value is reduced through the weight reduction factor; through a cascade mode, when the measuring noise R isstable, a system noise estimator is constructed through an improved adaptive filter, prior information Q is estimated in real time, and the optimal estimated values of the initial states of subsystemsare obtained; and the optimal estimated values of the two navigation subsystems are subjected to global convergence according to an information distribution principle, the final optimal estimated value is obtained, then the feedback process is conducted, and the steps from one to four are repeated. The problems that in a vehicle-mounted navigation system combined with a plurality of sensors, prior information of the system is uncertain, and the measuring value is abnormal, so that filtering is divergent and even loses efficacy are solved, the global optimal value is obtained, and thus the more accurate navigation parametric solution with better robustness is obtained.