Deep-Stall dynamic analysis of V-Tail aircraft and recovery using a hybrid strategy with constrained reinforcement learning

Deep stall is an extremely dangerous flight state that poses a severe threat to flight safety. To address this challenge, this paper investigates a safety-constrained deep-stall recovery method, combining the reinforcement learning and predefined-time control. First, the deep-stall characteristics of V-tail aircraft are systematically analyzed using pitch aerodynamic analysis, bifurcation theory, and phase portrait methods. Then, a hybrid reinforcement learning-based recovery strategy is proposed by integrating predefined-time incremental control with the penalized proximal policy optimization algorithm. In this strategy, safety requirements are transformed into policy constraints, ensuring that the rapid recovery process remains compliant with flight safety requirements. Moreover, the return-to-normal-flight after deep-stall recovery is explicitly considered in this paper. A complete six-degree-of-freedom aircraft model is adopted, enabling the coordinated use of ailerons together with other control surfaces to assist the pitch-rocking recovery maneuver. Comparative simulations are conducted to evaluate its effectiveness, robustness, and generalization capability. Results demonstrate that the proposed recovery strategy can achieve fast and safe deep-stall recovery. Moreover, the generated recovery policy closely aligns with pilot operating habits, making the proposed strategy particularly suitable for practical engineering implementation.