How do you design and tune a Kalman filter for multi-sensor navigation?

Navigation Kalman filter fuses measurements from GPS, INS, air data, and other sensors. Design steps: State vector selection - typically 15+ states (position, velocity, attitude errors, accelerometer biases, gyro biases, possibly GPS errors); Dynamics model - error state propagation based on sensor characteristics and vehicle dynamics; Measurement model - how each sensor relates to states (GPS provides position/velocity, magnetometer provides heading, etc.). Tuning: Process noise (Q) reflects sensor drift and unmodeled dynamics; too low causes filter lag, too high causes noise. Measurement noise (R) reflects sensor accuracy; use Allan variance for inertial sensors. Techniques: Schmidt-Kalman for consider states, adaptive filtering for varying dynamics, fault detection using innovation monitoring. Validation: Monte Carlo simulation across scenarios, covariance analysis for consistency, and flight test data validation. Real-time considerations: Computational load, numerical stability (square-root or UD factorization).