A Hostile Plan Recognition based on Plan Semantic Tree-Graph

First of all, we define the hostile plan, hostile plan domains and hostile plan recognition. Then we address a new plan representation which is based on plan knowledge graph representation, namely plan semantic tree-graph representation. The representation using a new object, i.e. dashed connecters, makes up disadvantages of plan knowledge graph which only expresses the total ordering plans. Thus, the representation can express not only the total ordering plans but also the partial ordering plans. Later by introducing the relevant action and the irrelevant action, crises degree and dangerous plan, we design a sort of hostile plan recognition algorithm, namely hostile plan recognition algorithm of duplex search with flags. The advantages of the algorithm are as follows. Firstly, it can infer unobserved actions that are executed actually by observed actions and make marks on actions which will be executed in the future and have been executed. Secondly, with respect to the new observed actions, the number of the nodes it generates is less than Kautz´s method and plan knowledge graph method. This enhances the efficiency of the plan recognition. Thirdly, it can forecast the next actions that the agent will perform when fewer actions are observed. The design of the algorithm takes account of the characteristics of the hostile plan recognition world so that it can dispose partial observability and concurrence in hostile plan recognition domains.