Iterative Belief Revision in Partial Observable Non-deterministic Planning

Any information about the current state is precious in Partial Observed Nondeterministic Planning (PONDP). Since the system do not exactly know the current state, new observation information is helpful to make it clearer. Although delayed effects are common in real-world domains, they have never been addressed in PONDP. Hence we propose a novel method for reasoning about belief states in PONDP, especially in the case of delayed effects. Addressing delayed effects need to revise not only the current belief state but also the whole belief history. The core algorithm is called Iterative Belief Revision algorithm (IBR), which bridges the gap between PONDP and belief change for the first time. IBR first finds out all action candidates for a newly known fact, and then determines which effects have happened, and finally revise the belief history along with the current state. Examples show that IBR fulfills its duty.