Robust autonomous orbit determination with dynamical uncertainty awareness: Application to Tianwen-2 at 311P

The Tianwen-2 spacecraft faces challenges in autonomous orbit determination near 311P/PANSTARRS due to its extremely weak gravity (~10⁻¹⁰ km/s² at 3 km) and stochastic accelerations from dust emissions. Traditional Extended Kalman Filters (EKF) degrade significantly in such highly dynamic environments. To address this, we develop the Uncertainty-Aware Filtering (UAF) framework, combining sigma-point propagation with physics-informed adaptive adjustment. UAF monitors normalized residuals and clearly distinguishes whether anomalies stem from measurement noise or system dynamics errors. It then selectively adjusts measurement or process noise to maintain filter consistency, enabling real-time robustness to unmodeled disturbances. Simulation results demonstrate that UAF achieves an order of magnitude improvement in orbit prediction accuracy and reduces computational cost by 42.5 % compared to EKF, even under a 500 m initial deviation and a constant perturbation of ~10⁻¹¹ km/s². The filter not only remains effective with initial deviations up to 0.7 km, but also demonstrates in controlled tests its ability to distinguish the effects of sensor noise and system dynamical uncertainty. Estimation degrades below 1 km altitude due to weak observability, while orbits above 3 km with 60–110° inclinations offer stable performance. These results provide valuable empirical guidance for mission design and validate UAF as a robust navigation solution for 311P proximity operations.