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 4 MAPPO和PPO算法的超参数设置

Tab.4 Hyperparameter settings of MAPPO and PPO algorithms

参数 数值 MAPPO PPO M 2 048 1 024 Bs 256 128 $ {l}_{{\mathrm{r}}} $ 2×10−4 10−4 $ \gamma $ 0.99 0.99 $ \epsilon $ 0.2 0.2 $ {\lambda }_{\text{GAE}} $ 0.95 0.95 $ {\alpha }_{\text{H}} $ 0.01 0.01 Tmax 300 300 Eepoch 10 3 τclip 0.5 — nh 2×256 2×128