Engine-Out Emergency Landing Based on Reinforcement Learning

Dual-engine failure leading to loss of thrust necessitates the development of emergency landing control for twin-engine aircraft. The existing methods can only generate landing paths based on simplified dynamic models and planning methods. However, the field of civil aviation lacks of safe emergency landing control systems for dual-engine failure, which demands more sophisticated dynamic modeling and enhanced real-time operation. In this paper, we propose a novel method for engine-out emergency landing, Multi-stage Pretraining and Hierarchical Fine-tuning (MPHF). It can achieve end-to-end control through task decomposition, stage training, and hierarchical fine-tuning. Additionally, we present a dual-engine failure environment to validate our method in the commonly used aircraft model. Experimental results demonstrate that MPHF achieves end-to-end real-time control in dual-engine failure scenes with the improved accuracy of emergency landing.