How would you design a GPS/INS integration architecture with optimal performance?

GPS/INS integration exploits complementary characteristics: INS provides high-rate, smooth, autonomous navigation but drifts; GPS provides absolute position with noise. Integration approaches: Loosely coupled - GPS position/velocity updates INS through Kalman filter, simple but suboptimal; Tightly coupled - GPS pseudoranges/Doppler directly integrated, better in degraded GPS (fewer satellites); Ultra-tight/Deep integration - GPS tracking loops aided by INS, enables operation in jamming/dynamics. Kalman filter states include position, velocity, attitude errors, sensor biases, and GPS errors. Design considerations: Filter tuning (process noise, measurement noise), Fault detection and isolation, Integrity monitoring (solution separation), and Reversion modes when GPS unavailable. Typical INS drift without GPS ~1 nm/hour; with GPS <10m continuous.