Decision making is one of the central problems in artificial intelligence and specifically in robotics. In most cases this problem comes with uncertainty both in data received by the decision maker/agent and in the actions performed in the environment. One effective method to solve this problem is to model the environment and the agent as a Partially Observable Markov Decision Process (POMDP). A POMDP has a wide range of applications such as: Machine Vision, Marketing, Network troubleshooting, Medical diagnosis etc. We consider a new technique, called Recursive Point Filter (RPF) based on Incremental Pruning (IP) POMDP solver to introduce an alternative method to Linear Programming (LP) filter. It identifies vectors with maximum value in each witness region known as dominated vectors, the dominated vectors at each of these points would then be part of the upper surface. RPF takes its origin from computer graphic. In this paper, we tested this new technique against the popular Incremental Pruning (IP) exact solution method in order to measure the relative speed and quality of our new method. We show that a high-quality POMDP policy can be found in lesser time in some cases. Furthermore, RPF has solutions for several POMDP problems that LP could not converge to in 24 hours.
展开▼
