导航

Acta Aeronautica et Astronautica Sinica

Previous Articles     Next Articles

Research on DQN-based active arrest and recovery technique for UAVs

  

  • Received:2024-10-28 Revised:2024-12-07 Online:2024-12-10 Published:2024-12-10
  • Contact: Yi-Ming PENG

Abstract: The success rate of hooking the cable is one of key indicators of safety and reliability when recovering a UAV using arresting cables. In response to the challenge of improving the success rate under restricted UAV recovery areas pro-vided by vehicle-mounted mobile platform, this paper proposes an active cable engagement method for UAV arresting and recovery. The method enhances the success rate by having the arresting system actively move to the optimal ca-ble engagement position under the guidance of AI computation results. First, a UAV landing and recovery dynamics model is established to calculate the cable engagement failure boundary. The Support Vector Machine (SVM) method is applied to identify the dynamics simulation results and generate a proxy model for UAV cable engagement analysis. Then, the cable-moving process is simplified into a Markov decision process. Using the cable-moving device model as the training environment, a six-degree-of-freedom UAV landing and descent model is employed to generate the dataset, and the cable engagement analysis proxy model is used as the reward function. A deep Q-network (DQN) is applied to train a cable-moving strategy that dynamically computes and guides the arresting system to adjust to the optimal en-gagement position in real time. Simulation results show that, under constrained recovery space, the proposed active cable engagement method improves the success rate by 29% compared to the traditional passive recovery method. This approach significantly enhances the safety and reliability of UAV landing and recovery, providing new theoretical support and practical solutions for the development of intelligent recovery technology.

Key words: Fixed-wing UAV, landing recovery, success rate of hooking cable, DQN, artificial intelligence

CLC Number: