Abstract: Tightly-coupled GNSS/INS integration has been shown to provide better performance than loosely-coupled approach in GNSS degraded environments. However, it is difficult to achieve reliable ambiguity resolution (AR) when the number of satellites is below four. In this case, ambiguity search in the ambiguity domain is not possible, which means other AR method should be considered. Also, it is likely to wrongly fix the ambiguities, which will lead to biased state estimate or even filter divergence. In this paper, a tightly-coupled RTK/INS algorithm based on centralized Kalman Filter (KF) is implemented, in which ambiguity parameters are augmented into the state vector. Ambiguity-fixed or -float carrier phase observables can be used to update the filter even if the number of satellites is less than four, and this improves the accuracy of the integrated system in complex environments. A field vehicular test was conducted to evaluate the performance of the tightly-coupled algorithm in terms of the position drift error during GNSS signal outages and ambiguity recovery time after outages. Test results indicate that by utilizing tactical-grade inertial measurement units (IMUs), the horizontal position drift error is about 0.3m during 30-second partial outage and the average time for reliable ambiguity recovery is 1 to 2 seconds when the number of satellites in view is three. Besides, the performance of float solution is only slightly worse than the fixed solution for shout-time (e.g. 60 seconds) partial outages, but both of them are much better than the case of full GNSS outages.