Multi-Agent_DRL

文章目录李宏毅笔记DDPG 李宏毅笔记 drl课程完整笔记 DDPG DDPG理论推导

文章目录Critic(通过学习使得自己估值越来越准确) Critic(通过学习使得自己估值越来越准确)

文章目录1. Q-learning 1. Q-learning an algorithm which produces a Q-table that an agent uses to find the best action to take given a state.

文章目录参考的大佬们numpyReplayBuffer 参考的大佬们 很详细的参考 numpy print(np.random.normal(size=6)) # 默认标准正态分布 μ=0, σ=1, shape = 6 [ 1.18306191 0.42100423 1.40213637 -0.17240589 0.02877852 -1.5534178 ] ReplayBuffer self.mem_cntr = 0 # the last memory index you store .

文章目录actions(not the reword)grid system前车与后车, 和自己的距离 actions(not the reword) grid system 前车与后车, 和自己的距离 前车是三个格子

文章目录Imitation Learning Imitation Learning 训练采用了imitation learning，这也是目前的主流吧，在这种FPS复杂环境下，高纬度的state and action space，维度灾难是显而易见的，所以他们先让Agent观看人类玩家30分钟的游戏操作之后，再开始训练，加快了policy的训练和收敛速度。 ...

文章目录Advantage func(how good a given action is compared to the average.) Advantage func(how good a given action is compared to the average.)   Can be thought of as a measure of how good a given action ...