Counter-Encirclement of UAV in Pursuit-Evasion Environment via Improved RL

This paper proposes a counter-encirclement method for evasive Unmanned Aerial Vehicle (UAV) within a successful capture environment via reinforcement learning. An improved Deep Deterministic Policy Gradient (DDPG) algorithm was utilized to train the actions of the evasive UAV, enabling it to successfully evade capture while simultaneously avoiding collisions. Additionally, adversarial modeling and the design of reward functions are conducted for both the pursuing UAVs and the evasive UAV. Finally, the improved DDPG algorithm was evaluated through simulations for its effectiveness in enhancing the escape performance of evading UAVs, demonstrating its efficacy in evasion scenarios within encirclement contexts.