This paper presents a detailed robustness analysis of three nonlinear filtering algorithms: the unscented Kalman filter, the cubature Kalman filter, and the ensemble Kalman filter, applied to aircraft state estimation for fixed-wing flight dynamics. The study focuses on estimating critical longitudinal flight parameters such as true airspeed, angle-of-attack, pitch angle and pitch rate using pitch angle measurements. A nonlinear aircraft model is formulated, and each filtering technique is implemented and evaluated under multiple scenarios, including sensor noise, initial state mismatches and plant-model uncertainties. Simulation results across four cases, ranging from ideal conditions to $95{\mathrm{\% }}$ mismatch, demonstrate that the unscented Kalman filter consistently delivers the most accurate and robust estimates, especially for velocity and pitch rate. The cubature Kalman filter offers a trade-off between estimation accuracy and computational efficiency, while the ensemble Kalman filter shows significant sensitivity to uncertainties but performs relatively better in estimating the angle-of-attack under severe mismatch conditions. This comparative study provides valuable insights for selecting appropriate filtering strategies in aerospace applications, particularly where robustness and reliability under uncertainty are crucial.

This paper investigates worst-case analysis of a moving obstacle avoidance algorithm for unmanned vehicles in a dynamic environment in the presence of uncertainties and variations. Automatic worst-case search algorithms are developed based on optimization techniques, and illustrated by a Pioneer robot with a moving obstacle avoidance algorithm developed using the potential field method. The uncertainties in physical parameters, sensor measurements, and even the model structure of the robot are taken into account in the worst-case analysis. The minimum distance to a moving obstacle is considered as an objective function in automatic search process. It is demonstrated that a local nonlinear optimization method may not be adequate, and global optimization techniques are necessary to provide reliable worst-case analysis. The Monte Carlo simulation is carried out to demonstrate that the proposed automatic search methods provide a significant advantage over random sampling approaches.