Real-time POMDP algorithm based on belief states space compression
Min Wu · Kongzhi yu juece · 2007
Solving belief state space for partially observable Markov decision processes(POMDP) is an NP-difficult problem.Therefore,a belief state space compression(BSSC) algorithm is proposed,which compacts belief state space from high dimension to low dimension.State transition function,observation function and reward function are compressed by using dynamic Bayesian network to reduce the solving dimension and realize real-time decision.The test data show that the BSSC algorithm can quickly solve optimal policy and optimal value function in the real-time environment.