Pickyour favorite single player computer game, for example, Wechat ‘Jump a jump’.How would you model the problem oflearning to play this game (without needing to know the physics of the game) asa reinforcement learning problem? Define statespace, action space, transition model, reward function and goalof the underlying MDP.
Markov Decision Process(MDP)
Reinforcement learning
Markov decision process
马尔可夫决策过程依赖于马尔可夫假设,即下一个状态si + 1的概率仅取决于当前状态si和动作ai,但不依赖于前面的状态或动作。
A good strategy for an agent would be to always choose an action that maximizesthe (discounted) future reward.Q(s,a)
基于Simple_Q_table_Q-learning的改进二维空间 环境+强化+提升循环效率
效果展示 Deep Q-Network(DQN) Deep Q-Network Algorithm:
Network Architecture(网络体系/架构) ![]() 效果展示: ![]() |