飞行包线
自动驾驶仪
控制系统
计算机科学
空气动力学
控制理论(社会学)
强化学习
控制工程
比例导航
导弹
模拟
工程类
航空航天工程
控制(管理)
人工智能
电气工程
作者
Brian Gaudet,Roberto Furfaro
出处
期刊:
[American Institute of Aeronautics and Astronautics]
日期:2023-01-19
被引量:2
摘要
View Video Presentation: https://doi.org/10.2514/6.2023-1628.vid We use reinforcement meta-learning to optimize and evaluate two integrated guidance and flight control systems, the first being fully integrated and the second being partially integrated. The fully integrated system maps navigation system outputs directly to commanded control surface deflections. In contrast, in the partially integrated system a guidance system maps navigation system outputs to a commanded acceleration, with the flight control system mapping this acceleration to commanded deflections. The latter approach is integrated in the sense that the flight control system consists of a single control loop, and is optimized for a particular guidance system. We optimize and evaluate the two approaches in an environment modeling a point defense interceptor defending against a hypersonic threat. We find that both systems adapt to a large flight envelope and off-nominal flight conditions that include perturbation of aerodynamic coefficient parameters and flexible body dynamics. Similarly, both systems are robust to the parasitic attitude loop induced by radome refraction and imperfect seeker stabilization. We then repeat the experiment by evaluating the two approaches in an air-to-air missile environment. Although our original hypothesis was that the fully integrated system would provide the highest performance, in both environments we found that the converse was true.
科研通智能强力驱动
Strongly Powered by AbleSci AI