Design of Longitudinal Automatic Carrier Landing System Based on Reinforcement Learning

Abstract

AbstractThis paper mainly aims at automatic carrier landing task under high sea conditions, and proposes a method for designing the landing guidance law based on reinforcement learning, and the effect of which is also tested and analyzed. This paper first established the longitudinal small perturbation equation, control law model, ship surface atmospheric environment model and aircraft carrier deck motion model of F-18 carrier-based aircraft as the guidance law training environment. For better docking engineering practice, the outer loop of the automatic carrier landing system (ACLS) designed in this paper adopts landing guidance law based on reinforcement learning, and the inner loop adopts traditional \(\dot{H}\) tracking control law. It is found that the landing accuracy of the guidance law based on reinforcement learning is significantly better than that based on the PID algorithm. When facing steady-state disturbance, it can anticipate disturbances and resist them to ensure the accuracy of the mean value of landing point. In the face of random disturbance, it can also generate more appropriate instructions and reduce the distribution range of landing points. In this paper, the Markov decision process modeling method of ship landing mission is also studied from the perspective of flight dynamic. The result shows that it is very important to select the appropriate state quantity as the input. And after adding random factors into the state transition function, the training effect of guidance law is significantly improved. The random environment should consider not only random distribution, but also amplitude. Adding random amplitude disturbance can significantly improve the training effect.KeywordsCarrier-based aircraftAutomatic landingReinforcement learningMarkov decision processRandom disturbance