关键词: 基于实例的方法, 部分可观察马尔可夫决策过程（POMDP）, 启发式求解, 增强学习, 最近邻, 局部加权回归

Abstract: In this paper，with the idea of combining heuristic solution and reinforcement learning technique，the instance based approximate solution to POMDP problem is studied and Nearest Neighbor based algorithm NNI and its extended parameterized version ENNI and Local Weighted Regression based algorithm LWI are presented.With the performance analyzed and compared through experiments on common workbench，solving POMDP problems using instance based methods can produce good sub-optimal solutions.

Key words: instance based method, Partially Observable Markov Decision Process（POMDP）, heuristic solution, reinforcement learning, nearest neighbor, local weighted regression