Compensating Star-Trackers Misalignments with Adaptive Multi-Model Estimation

This paper presents an adaptive multi-model framework for jointly estimating spacecraft attitude and star-tracker misalignments in GPS-denied deep-space CubeSat missions. A Multiplicative Extended Kalman Filter (MEKF) estimates attitude, angular velocity, and gyro bias, while a Bayesian Multiple-Model Adaptive Estimation (MMAE) layer operates on a discrete grid of body-to-sensor misalignment hypotheses. In the single-misalignment case, the MEKF processes gyroscope measurements and TRIAD-based attitude observations, and the MMAE updates a three-dimensional grid over the misalignment vector. For a dual-misalignment configuration, the same MEKF dynamics are retained, and the MMAE bank is driven directly by stacked line-of-sight measurements from two star trackers, forming a six-dimensional grid over the two misalignment quaternions without augmenting the continuous-state dimension. A novel diversity metric, $\Psi$, is introduced to trigger adaptive refinement of the misalignment grid around a weighted-mean estimate, thereby preventing premature collapse of the model probabilities and concentrating computation in the most likely region of the parameter space. Monte Carlo simulations show arcsecond-level misalignment estimation and sub-degree attitude errors for both estimation problems, with estimation errors remaining well-bounded, proving robustness and consistency. These results indicate that the proposed MEKF--MMAE architecture enables accurate, autonomous, and computationally efficient in-flight calibration for resource-constrained spacecraft, and establishes dual star-tracker misalignment estimation as a practical option for deep-space CubeSat missions.