基于对抗图生成的动态蒙特卡洛强化学习

宋莉,宛袁玉,金苍宏,宋明黎

Dynamic Monte-Carlo reinforcement learning based on adversarial graph generation

Li SONG,Yuanyu WAN,Canghong JIN,Mingli SONG

表 2 GGAN-DQN、GGAN-MCRL、MA-DQN在多智能体驾驶环境中的结果

Tab.2 Result of GGAN-DQN, GGAN-MCRL and MA-DQN in multi-agent driving environments

环境 算法 los ls Qd Q Rc Psuc highway- fast-v0 MA-DQN 0.934 — −1.022 3.782 7.237 0.86 GGAN-DQN 1.005 −2.556 0.8467 4.748 15.119 0.82 GGAN-MCRL 0.490 −2.507 0.332 24.344 15.855 0.97 merge-v0 MA-DQN 0.214 — −2.018 3.136 8.970 0.64 GGAN-DQN 1.398 −5.100 1.079 2.691 34.317 0.60 GGAN-MCRL 0.423 −2.430 0.301 24.412 42.240 0.98 roundabount-v0 MA-DQN 0.114 — −0.751 4.777 9.853 0.84 GGAN-DQN 2.627 −4.733 1.356 5.045 26.555 0.73 GGAN-MCRL 0.347 −2.291 0.241 23.114 34.253 0.98