Based On The Overall Effectiveness Of Agent Intention To Reconsider The Strategy Study

When designing agent that working in real environments, the BDI model is a popular agent architecture .In so-called Belief-Desire-Intention(BDI) agent architecture, agent select some desire (goal)as its intention based on its beliefs about the world. While the real world are usually dynamic,unpredictable and non-deterministic, so the intention of a agent will not be permanent and static. For example, when an intention cannot be achieved anymore because the circumstances have changed, or the agent might benefit more from other intentions, agent should reconsider its intention.In BDI agents, decision making is composed of two processes: deliberating (decide which intention to achieve) and means-ends reasoning (decide how to achieve this intention).The process of intention reconsideration is a process of re-deliberating the desires. In the past, researchers usually treat deliberation as this: agents deliberate every single goal and then select one certain goal as its intention by some regulations. In this paper, a concept of multi-goals' utility was defined. An agent regard multi-goals as a set and deliberate many sets based on multi-goals' utility, and then agent select one set as its intention set and will be restricted by this intention set. This idea was inspired by the overloading method of Bratman. Some intention reconsideration policies based on this method were given and assessed by experiments in the TILEWORLD domain. The results of our experiments showed that our idea is effective and efficient.Deliberation is a computationally costly process, and in order for a BDI agent to operate effectively, it should choose to deliberate only when necessary, this requires an appropriate intention reconsideration policy to make agent to decide adaptively whether to deliberate or not when running in some environment. In this paper, multi-goals' utility and set restriction were integrated with the DDS process, and an adaptive intention reconsideration policy was presented, where Schut's method of using DDS was referenced. Then this policy was assessed by experiments in the TILEWORLD domain, and the results showed that this policy is effective.