Research On Robot Navigation Based On POMDP

Partially Observable Markov Decision Process (POMDP) is a general sequential decision-making model where the effects of actions are nondeterministic and the information about world states is partially available. So, POMDP is very effective on the problem of sequential decision-making under uncertainty. In this paper, theory of POMDP is studied, complexity and theoretical result of POMDP are analyzed, and its value iteration algorithm is modified properly. The robot navigation task based on the research of framework of POMDP is performed by simulation, and the simulation results are analyzed.Firstly, some general algorithms in robot navigation are introduced, and the current research status of POMDP is given in detail.Secondly, the Mathematics framework and theory of Partially Observable Markov Decision Process (POMDP) are discussed in detail. In order to explain the framework and background perspicuously, a gradual study process is given: first, the Markov Decision Process (MDP) is amply discussed; second, on the base of MDP, the definition and solution of POMDP model are given. The standard algorithms of POMDP are studied, an exact value iteration algorithm is analyzed thoroughly, and the value iteration algorithm of POMDP is modified properly.Finally, the design of POMDP robot navigation model is given, and lots of simulation experiments are carried out for robot navigation in complicated simulation environment. For the robots which run in real world, the uncertainty of the information on world states shows mainly in the robot perception on its environment by sonar. In this paper, probability form is applied to show the uncertainty of sonar, and make it combined with the belief update of POMDP, that gives an improvement on the application of POMDP in robot navigation. The results of experiments show that this approach can perform robot navigation task in unknown complicated environment effectively.