An Online Trajectory Planning Framework for Hypersonic Morphing Vehicles Incorporating Physical Constraints

In this paper, a neural network-based online trajectory planning framework incorporating physical constraints is proposed and applied to the reentry phase of a variable-wingspan Hypersonic Morphing Vehicle (HMV). This framework ensures real-time performance while reducing violations of physical constraints in the planned trajectories. The framework consists of three components: offline dataset construction, offline training, and online planning. During offline dataset construction, an optimal trajectory dataset is generated based on the developed HMV reentry trajectory optimization model, considering initial state disturbances and parameter deviations in vehicle configuration. In the offline training phase, the physical constraints of the HMV reentry phase are transformed into penalty terms in the loss function, and multi-scale loss terms are regularized and balanced to ensure training stability and equilibrium. This approach achieves the integration of physical constraints and data-driven methods. In the online planning phase, the trained network is deployed for real-time trajectory planning. Comprehensive experiments demonstrate the superior flight performance of the HMV compared to traditional vehicles and validate the effectiveness of the proposed online trajectory planning framework.