简体繁体 English

在强化学习中，agent 和 environment 有不同的状态还是只有一种状态？

[英]In Reinforcement learning , do both agent and environment have different states or there is only one state?

原文 2020-02-17 06:51:32 8 1 machine-learning/ reinforcement-learning/ markov-decision-process/ mdp

Is the state of the environment or the agent ?是环境状态还是代理状态？ Somewhere i read environment also has state .我在某处阅读环境也有状态。 How do both of them differs ?两者有何不同？

1 个解决方案

The state represents the environment as seen from the agent.状态代表从代理看到的环境。

If you want to say that some attribute is part of the agent (let's say you have a car and you declare that its speed belongs to the agent), then the state contains this attribute but it is part of the environment as your agent exists in the environment.如果你想说某个属性是代理的一部分（假设你有一辆车并且你声明它的速度属于代理），那么状态包含这个属性但它是环境的一部分，因为你的代理存在于环境。

Then, when you have multiple agents, all of them can "visualize" the environment through a state but each of those can be different.然后，当您有多个代理时，所有代理都可以通过状态“可视化”环境，但每个代理都可能不同。

For example例如

If you have multiple bots playing cards, each bot will represent the environment as the visible cards and the cards in its hands.如果您有多个机器人在打牌，则每个机器人将环境表示为可见的牌和它手中的牌。 Therefore, all of your agents (the bots) will have a different perception of the environment, resulting in different states to represent them.因此，您的所有代理（机器人）将对环境有不同的感知，从而产生不同的状态来表示它们。 In this case, you could say "the state of the agent", it is just a matter of phrasing.在这种情况下，您可以说“代理的状态”，这只是措辞问题。

强化学习中同一状态的不同奖励 - Different rewards for same state in reinforcement learning

强化学习适用于 RANDOM 环境吗？ - Is reinforcement learning applicable to a RANDOM environment?

用于连续状态，离散动作的强化学习算法 - Reinforcement learning algorithms for continuous states, discrete actions

在强化学习中定义状态，Q和R矩阵 - Defining states, Q and R matrix in reinforcement learning

强化学习-如何摆脱“粘性”状态？ - Reinforcement Learning - How to get out of 'sticky' states?

强化学习 - 当游戏的输入只有像素时，我们如何决定对代理的奖励？ - Reinforcement Learning - How to we decide the reward to the agent when the input to the game is only pixels?

在任意较大的动作/状态空间中进行强化学习 - Reinforcement Learning in arbitrarily large action/state spaces

DQN（强化学习）：状态应该标准化吗？ - DQN(Reinforcement learning) : should state be standardized?

在每个州都处于终结状态的强化学习 - Reinforcement Learning where every state is terminal

增强学习以获取连续的状态和动作空间 - Reinforcement learning for continuous state and action space

暂无

暂无

声明:本站的技术帖子网页，遵循CC BY-SA 4.0协议，如果您需要转载，请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.

相关问题 强化学习中同一状态的不同奖励 - Different rewards for same state in reinforcement learning 强化学习适用于 RANDOM 环境吗？ - Is reinforcement learning applicable to a RANDOM environment? 用于连续状态，离散动作的强化学习算法 - Reinforcement learning algorithms for continuous states, discrete actions 在强化学习中定义状态，Q和R矩阵 - Defining states, Q and R matrix in reinforcement learning 强化学习-如何摆脱“粘性”状态？ - Reinforcement Learning - How to get out of 'sticky' states? 强化学习 - 当游戏的输入只有像素时，我们如何决定对代理的奖励？ - Reinforcement Learning - How to we decide the reward to the agent when the input to the game is only pixels? 在任意较大的动作/状态空间中进行强化学习 - Reinforcement Learning in arbitrarily large action/state spaces DQN（强化学习）：状态应该标准化吗？ - DQN(Reinforcement learning) : should state be standardized? 在每个州都处于终结状态的强化学习 - Reinforcement Learning where every state is terminal 增强学习以获取连续的状态和动作空间 - Reinforcement learning for continuous state and action space

相关标签

粤ICP备18138465号 © 2020-2024 STACKOOM.COM