Unverified50% confidenceFactExact time
In a partially observable Markov decision process (POMDP), the agent cannot directly observe the true state and can only maintain a belief state through the history of observations, and the optimal policy is a function of the belief state.
1
Sources
50%
Confidence
Long-term
Relevance
10/22/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedPOMDP introduces the belief state, i.e., a probability distribution defined over all possible hidden states, based on which the agent makes decisions rather than from a single deterministic state.80% similarUnverifiedMDP假设当前状态包含做出最优决策所需的全部信息（马尔可夫性质），而POMDP中智能体无法直接观测完整的环境状态78% similarUnverified贝叶斯滤波是POMDP中维护信念状态的核心机制，通过状态转移模型预测先验概率、用观测模型计算似然、归一化得到后验信念72% similarUnverifiedPOMDP在MDP基础上增加了观测函数O(o|s,a)，智能体只能获得状态的部分信息，需要维护对真实状态的置信度72% similarUnverified解决POMDP的经典方法是维护信念状态（belief state），但在高维空间中计算上不可行；现代深度强化学习通过LSTM或GRU等循环结构隐式维护历史信息70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1084012API
curl https://kongchang.com/api/v1/knowledge/claims/1084012MCP
get_claim(id=1084012)