Neural Network Observer-Based Nonsingular Practical Predefined-Time Control for Laterally Symmetric Vehicle During Boost Phase

This paper addresses the attitude tracking control problem for laterally symmetric vehicles during the boost phase under aerodynamic parameter variations and high-altitude wind disturbances. A neural disturbance observer-based nonsingular predefined-time sliding mode control scheme is proposed. First, a Lyapunov-based predefined-time stability criterion is established, which facilitates the design of an adaptive predefined-time observer using radial basis function neural networks. Without requiring prior knowledge of disturbance bounds, this observer ensures that disturbance estimation errors converge to a neighborhood of the origin within a predefined time parameter. Second, a novel nonsingular predefined-time sliding surface is constructed using hyperbolic tangent functions, leading to an integrated predefined-time sliding mode controller. The proposed scheme guarantees that the upper bound of the convergence time for initial attitude tracking errors is independent of the initial boost-phase states and can be arbitrarily predefined. Unlike conventional predefined-time control methods, the proposed approach eliminates controller singularity issues while avoiding the introduction of piecewise continuous functions or double-integral terms in either the sliding surface or the control law, thereby reducing structural complexity. Theoretical analysis confirms the boundedness of all closed-loop signals during attitude tracking. Numerical simulations demonstrate the effectiveness of the proposed control strategy under complex flight conditions.