Using drones to neutralize the threat of drones is an active defense method. The key is to generate countermeasures against drones, in which reinforcement learning is a popular and effective method. However, drones need to pay attention to the number of enemies and friends in intense real-time confrontation to adopt policies challenging for learning stable policies. This paper introduces the self-attention method into the MADDPG algorithm, which adapts to the dynamic changes in the number of friends and foes during the policy learning process to make the learned policy more stable. We create a simplified 2D simulation environment and three collaboration and confrontation scenarios for policy learning and experiments. Experiments show that the method can adapt to the dynamic changes of enemy-to-friend confrontation during the confrontation process and has higher performance than the baseline method.
Anti-drone Policy Learning Based on Self-attention Multi-agent Deterministic Policy Gradient
Lect. Notes Electrical Eng.
International Conference on Autonomous Unmanned Systems ; 2021 ; Changsha, China September 24, 2021 - September 26, 2021
Proceedings of 2021 International Conference on Autonomous Unmanned Systems (ICAUS 2021) ; Kapitel : 225 ; 2277-2289
2022-03-18
13 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
DOAJ | 2022
|Deep Deterministic Policy Gradient (DDPG) Agent-Based Sliding Mode Control for Quadrotor Attitudes
DOAJ | 2024
|