| 摘要 |
PROBLEM TO BE SOLVED: To achieve a device and method for constructing an altered POMDP. SOLUTION: As for the respective information of a state, action, observation and reward defined in a partially observable Markov decision process(POMDP), a plurality of different factors are respectively set, and a factored POMDP clarifying the relations of those factors is constructed. As for status transition prediction or action decision processing to which the factored POMDP is applied, processing clearly discriminating the causal relations of factor units is made possible, and more accurate state prediction or effective action determination processing is made possible. COPYRIGHT: (C)2008,JPO&INPIT
|
