This paper presents the design of a tightly coupled inertial/global navigation satellite system (GNSS) hybrid navigation system tailored to rocket launch applications. The tight architecture, selected for its design flexibility and robustness, uses time-differenced GNSS carrier phases to support the pseudorange measurements in the correction of strapdown inertial propagation and in-flight inertial sensor calibration. The robust filter design departs from a high-order inertial sensor calibration model, accounting for multiple error sources. Thorough order reduction is then performed through impact and observability analysis of each inertial sensor uncertainty for several sensor grades and under a set of launch trajectories. On-the-pad alignment sequence is also analyzed. Schmidt–Kalman filtering is used to account for sensor errors without their explicit estimation. The GNSS filter model is also reduced, maintaining robustness against receiver clock, atmospheric delays, and channel biases. Testing is done using simulated and real GNSS measurements produced from Vega launcher real-flight data. The effectiveness of the design is demonstrated against a full-order filter under a variety of inertial error sources. Comparison to a loosely coupled setup shows clear advantage under partial GNSS outage conditions, whereas comparison to inertial-only solutions reveals stable position/velocity performance and attitude estimation improvement equivalent to a one-step inertial sensor upgrade.