REAL-TIME TRAJECTORY PLANNING FOR HYPERSONIC GLIDING GUIDED PROJECTILES INTEGRATING PHYSICS CONSTRAINTS AND TRANSFORMER ATTENTION MECHANISM
Abstract
Supersonic gliding guided projectiles demand real-time trajectory planning with both computational speed and control smoothness. This paper introduces a third-order jerk penalty as physics-informed regularization within a causal transformer for online control sequence generation. The method was validated with a Gliding Guided projectile using Computational Fluid Dynamics aerodynamics and pseudospectral-generated optimal trajectories. The method suppressed high-frequency oscillations directly in the control space without simplified model approximations. Across independent and identically distributed, out-of-distribution, and label-noise scenarios, the jerk regularization reduced control jerk by approximately one order of magnitude and reduced the angle-of-attack total variation by 44%, doubling the closed-loop arrival rate from 13% to 27%. With receding-horizon model predictive control, both the regularized and the purely data-driven models achieved a 100% arrival rate. The regularized model further yielded 43% higher terminal velocity at short replanning intervals due to the smoother control profiles.
DOI
10.12783/ballistics26/37598
10.12783/ballistics26/37598
Refbacks
- There are currently no refbacks.