Trajectory optimization learning for fixed-wing aircraft

doi:10.1016/j.cja.2026.104388

Trajectory optimization learning for fixed-wing aircraft

2026-07-17

0

PRE

AI

DOI：10.1016/j.cja.2026.104388

Original

Original request for help

Save

Abstract

En 中文

Trajectory optimization for six-degree-of-freedom fixed-wing aircraft is challenging due to the conflicting demands of real-time performance, feasibility, and optimality. To address this, we introduce a self-supervised trajectory optimization learning framework with balanced trajectory feasibility and optimality. This framework directly imitates the augmented Lagrangian method through deep neural networks, transferring the time-consuming optimization process to the training stage. After training, the neural network, due to the adopted sparse trajectory parameterization, can directly perform temporally stable trajectory-level inference. In simulations of a supermaneuverable aircraft performing maneuvers at low airspeed and high angle-of-attack, our framework generated trajectories with an average time of 0.156 ms, with trajectory feasibility and optimality close to those achieved by advanced pseudo-spectral methods.

Keywords:

Trajectory optimization learning

Six-degree-of-freedom fixed-wing aircraft

Augmented Lagrangian method

Deep neural networks

Self-supervised learning

Trajectory optimization learning for fixed-wing aircraft

Abstract

Journal

Organization