Using MDP to Find the Best Path in Multi-floor World

This paper discusses the stochastic path (SSP) problem in an uncertain grid world. In a chaotic environment where lots of factors, like people, can disturb performances of an agent, like an indoor service robot, the agent needs to consider the uncertainty to make sequential decisions so that accomplishes tasks. Markov Decision Process (MDP) is a widely used mathematic model as well as reinforce learning model for dynamic algorithms solving problems in an environment with uncertainty. Based in MDP, an algorithm was proposed to find the best path in a multi-floor grid world where the agent can move upstairs or downstairs by lifts and stairs. The results show that, given the grid map, the algorithm can find the path with the lowest energy or time cost considering in such environment. The algorithm can be applied to mobile agents in stochastic environment like service robots in shopping malls.