基于自适应课程强化学习的多无人艇对抗围捕决策

陈浪,刘增力,赵宣植

Decision-making for multi-USV adversarial encirclement based on adaptive curriculum reinforcement learning

Lang CHEN,Zengli LIU,Xuanzhi ZHAO

表 5 各算法在2种目标策略下的性能指标对比

Tab.5 Comparison of performance metrics of different algorithms under two target strategies

目标策略 算法 $ {C}_{\text{u}} $ $ {C}_{\text{T}} $ $ {P}_{\text{u}} $ $ {P}_{\text{T}} $ $ {D}_{\text{avg}} $/m $ {R}_{\text{avg}} $ RL ACL-MAPPO 194 40 46 2 302.324 55.253 CL-MAPPO 364 58 71 22 367.474 42.437 NOCL-MAPPO 1 401 134 432 38 421.161 17.367 ACL-MADDPG 389 67 83 31 381.193 40.152 Random ACL-MAPPO 942 301 154 54 285.391 37.113 CL-MAPPO 1 224 304 205 52 321.482 30.172 NOCL-MAPPO 1 558 150 367 56 343.514 13.627 ACL-MADDPG 1 287 317 221 61 337.643 25.241