Aiming at fixed-wing UAV attitude control, this paper proposes ASPPO algorithm based on action scaling mechanism and reward shaping method based on Sequential-dependent mechanism. Firstly, the UAV attitude control model is constructed, the state space and action space are set, and the general base reward function is designed. Subsequently, in order to optimize the training process of the agent, the ASPPO algorithm based on the action scaling mechanism is proposed. Further, we introduced a reward shaping method based on the Sequential-dependent mechanism to provide richer and more accurate feedback information for the agent by reconstructing the reward function. Ultimately, through the UAV attitude control simulation experiments, the results demonstrate that the Atitude_Control_raw+AS+SD model, with the addition of the ASPPO algorithm and the sequential-dependent reward shaping method, compared with the Atitude_Control_raw+AS model using only the ASPPO algorithm, improves the flight length and cumulative reward by 11% and 13%; compared to the Atitude_Control_raw model using only the base reward function, significant improvements of 39% and 48% were realized.
Deep Reinforcement Learning-Based Attitude Control Method for Fixed-Wing UAV
Lect. Notes Electrical Eng.
International Conference on Autonomous Unmanned Systems ; 2024 ; Shenyang, China September 19, 2024 - September 21, 2024
Proceedings of 4th 2024 International Conference on Autonomous Unmanned Systems (4th ICAUS 2024) ; Chapter : 33 ; 343-354
2025-03-28
12 pages
Article/Chapter (Book)
Electronic Resource
English