A POMDP Approximation Algorithm That Anticipates the Need to Observe

This paper introduces the even-odd POMDP, an approximation to POMDPs in which the world is assumed to be fully observable every other time step. The even-odd POMDP can be converted into an equivalent MDP, the 2MDP, whose value function, V 2MDP , can be combined online with a 2-step lookahead search to provide a good POMDP policy. We prove that this gives an approximation to the POMDP's optimal value function that is at least as good as methods based on the optimal value function of the underlying MDP. We present experimental evidence that the method gives better policies, and we show that it can nd a good policy for a POMDP with 10,000 states and observations.

• Type: Conference
• Year: 2000
• Where: PRICAI
• Authors: Valentina Bayer Zubek, Thomas G. Dietterich
• Keywords: Artificial Intelligence, Even-odd Pomdp, Optimal Value Function, PRICAI 2000, Value Function