关键词: 智能规划, 规划图, Markov决策过程, 不确定规划, 群体智能算法

Abstract: In the classical planning problems，to found a planning solution is a NP problem or even a NP complete problem.If the executing effects of the action take uncertainty，such as the planning problem of Markov Decision Processes（MDP），the problem will be more difficult.Some concepts such as the reflection action，reflection path-section，and reflection planning graph are presented based on the graph plan algorithm，and the ant colony planning algorithm will be designed based on them.In the algorithm，the actions can generate more nodes that each one is a representation of a proposition.It is proved that there is no less than a certainty probability that the solution of the ant colony planning algorithm will be reliability even in the uncertain action executing environments.

Key words: intelligence planning, planning graph, Markov decision processes, planning under uncertainty, swarm intelligence algorithm