BDS outages and leads to steering failure is an acknowledged challenge when the agricultural machine is automatically navigating. Steering failure will lead to increased labor intensity and loss of agricultural materials. A kind of self-calibrating variable structure Kalman filter for tractor navigation during BDS outages was presented. A Kalman filter was designed to integrate BDS and INS information for improving positioning quality. The filter was real-time calibrated by a calibration method of initial heading based on auto regressive model and a calibration method of angular rate zero offset online, which could process BDS/INS information and combined navigation tracking control method, thus implementing the continuous navigation function. The method was tested for decimeter accuracy in both robotic vehicle navigation system and tractor in-field to follow paths with lines and rectangles during BDS outages. In the robotic vehicle test, the speed of vehicle was set 1.0m/s, the performance was compared with that of the variable structure Kalman method without self-calibrating. The results showed that the self-calibrating variable structure Kalman method reduced the actual lateral deviation and BDS between INS positioning distance during BDS outages by 34% and 44%, respectively. The average driving distance was increased by 80% when the lateral deviation reached 20cm, and then reduced. In the tractor in-field test, the speed of vehicle was set 1.0m/s, the average driving distance reached 16.65m when the lateral deviation reached 20cm. This method had guiding value for the emergency endurance processing of BDS outages in the process of cotton planting automatic navigation.