BriefGPT.xyz
Sep, 2011
Perseus: POMDPs随机基于点的价值迭代
Perseus: Randomized Point-based Value Iteration for POMDPs
HTML
PDF
M. T. J. Spaan, N. Vlassis
TL;DR
介绍了一种基于点集采样的算法——Perseus,使用该算法可以解决大规模的部分可观测马尔可夫决策过程问题,其通过随机选择子集进行值迭代,提高信念集中每个点的值,特别适用于连续动作空间。
Abstract
partially observable markov decision processes
(POMDPs) form an attractive and principled framework for agent planning under uncertainty.
point-based approximate techniques
for POMDPs compute a policy based on a
→