Similarity: entity, packaging state, can perform some actions and methods; Communicate through messages;
Difference:
1) When deciding whether to execute the method of the object, the decision-making power is different; In the object-oriented system, the decision is made by the object who actively calls the method, while in the agent system, the decision is made by the agent who accepts the request; .
2) Flexible behavior ability (reactivity, initiative and sociality) is considered as one of the basic attributes of Agent;
3) The parallelism of agents is the most basic feature; Traditional objects are not like this;
Second, an overview of service-oriented architecture SOA
Agent-oriented system has more and more extensive application value. In the formalization process of rational agent, it is generally believed that the thinking state of agent includes three attributes: belief, desire and intention, so BDI model has always been the focus of agent modeling research.
BDI model
At present, the BDI model used by agents is generally developed on the basis of the intention model of Cohen and Levesque's normal modal logic and the BDI computational logic model of Rao and Georgeff, that is, it focuses on the formal description of beliefs, desires and intentions, or BDI for short. The essential problem to be solved is how to determine the agent's goal and how to achieve it.
The BDI model consists of three basic components:
(1) Belief is a set of beliefs related to the world, beliefs related to the thinking tendency of other subjects and self-beliefs. Belief is the subject's cognition of the world, including data describing environmental characteristics and data describing their own functions, and it is the basis of the subject's thinking activities.
(2) Desire is the initial motivation of the subject, and it is a set of states that he hopes to achieve or maintain. The state that the subject wants to achieve can stimulate the planning and action of the system. Generally speaking, it can be expressed as the subject's expectation and judgment of the environmental state, that is, judging whether the state is established is a sign of whether the wish is realized. Agents can have incompatible wishes, and they don't need to believe that their wishes are absolutely achievable.
(3) Intention is the most necessary or suitable thing to accomplish at present, and it is the goal that the current subject is about to achieve, which belongs to the intentional direction of psychological state. The current intention plays a guiding role in the agent's current action.
The agent research based on BDI model is generally divided into two levels, one is theoretical level, also called logical level, which mainly studies how to give a formal semantic description of BDI agent by logical method; The other level is the practical application level, including the design and programming of the system architecture.
2.BDI logic
2. 1 modal logic
Modal logic is about inevitability and possibility, or about "must be" and "may be". Modal logic not only considers the truth and falsehood of the actual existence of things, but also considers what will be true or false if things are in a different existence from the actual existence. Logic focuses on truth and falsehood, while modal logic focuses on truth and falsehood in the real world and other possible worlds. In this sense, a proposition is inevitable in a world, and it is only possible to be true in all possible worlds that may replace this world, and to be true in a possible world that may replace this world.
Usually, □ stands for inevitability operator and △ stands for possibility operator.
Definition: A frame F is represented as an ordered pair.
Definition: A model M is represented as a ternary ordinal group.
Definition: If K is a minimal normal modal system, then many modal systems are obtained by adding axioms to K, such as T, S4, S5, KD, etc.
2.2 ? BDI logic
BDI logic is a multi-modal logic, which models beliefs, wishes and intentions into normal modal operators under possible world semantics, which are expressed as BEL, DES and INT respectively, where (agent)s).
BDI: B, D, I D, I satisfy reflexivity;
BDI: B, D and I satisfy reflexivity and transitivity;
B D I: B satisfies Euclid, continuity and transitivity, and D satisfies reflexivity, transitivity and symmetry; I am satisfied with reflexivity.
3. Combinatorial logic
The goal of combinational logic is to make comprehensive use of familiar existing logic tools. At present, there are two kinds of combination skills, one is fusion and the other is fiber. Through detailed comparative analysis, this paper thinks that fiber method is more feasible. (Liu Yang is responsible for the explanation)
4. Extension of BDI model with compound function.
In this chapter, compound action is considered, and two operators, RES (result) and OPP (opportunity), are added to the BDI model extended by predecessors, and the semantic description is given.
4. 1 Research background and motivation
We can define things in the world as right and wrong, and we can also define actions and events as success and failure. There are no concepts related to actions and events in BDI model.
The subtle differences between actions and events are as follows: actions refer to actions taken by the subject when one state is transformed into another; An event is a performance result of a specific agent performing a specific operation.
Rao? Georgeff gives a semantic description of the event based on the proposed temporal framework, but does not explain the actions involved in the specific event.
4.2 Research Foundation
(1) Brown Theory-Action and Ability
Considering the factors of action and ability, this paper points out a main measure of ability: reliability, which can be measured by the number of repeated executions.
Definition: model M=(W, f, v), where w is the set of possible worlds; V is the assignment of atomic propositions in each world; F is a function that connects a single world with a group of worlds.
m，w？ Ability? iff？ And then what? :M，w '？
Semantics are: in the model, there is a world set, so that any world in the world set can satisfy the ability of the agent? Then we think the agent is capable.
Brown theory also gives the definitions of operators such as will and strength.
(2) Unintentional theory of ability;
On the basis of unintentional and objective goal orientation proposed by Sommerhoff, Elegesem holds that the subject maintains a specific goal state in a dynamic environment.
Definition: focus condition f (,) = o, semantics: it is believed that to achieve the goal, there must be at least a certain point in time, at least an action variable and at least an environment variable, so that the goal can be achieved by executing actions at any time and under any conditions.
Definition: model m = (w, v), w is the world set; = … ; V is a function. Is the dynamic set of W.
Definition:? (w,), I refers to the agent; W refers to a world; Is a collection of worlds that make the goal g true. Semantics of (w,): It is a collection of worlds in which subject I has the ability to achieve goal G. ..
Definition:? (w,), if and only if w' satisfies f (,) = o.
Definition: Success? (W，X) X if x w。
Theoretical significance: In all worlds, if a subject has the ability to achieve the goal, it can be defined as success.
Elegesem's unintentional theory also distinguishes between "the agent has the ability to achieve the goal" and "the agent really achieved the goal by taking action" and gives a semantic description.
(3) ability-BDI;
Padgham believes that an agent has the ability to achieve a certain goal G, which is always related to a plan. In fact, every plan can be regarded as a trigger event to achieve this goal G, which means that the agent has at least one way to achieve this goal G. 。
Definition: belief-ability compatibility axiom: CAP() BEL (), semantics: If agent I has the ability to realize it, agent I believes it.
Definition: axiom of ability-goal compatibility: GOAL() CAP (), semantics: If the goal of agent I is, agent I has the ability to achieve it.
The following table shows the BDI model and Rao after adding CAP. Comparison of BDI prototype proposed by Geoff;
Rao? Geoff Padgam?
I system IC system?
A 1: GOAL( ) BEL( ) CAP( ) BEL()？
A2: INT( ) GOAL( ) GOAL( ) CAP()？
a3:INT(does())does()CAP()BEL(CAP())？
a4:INT()BEL(INT())GOAL()CAP(GOAL())？
a5:GOAL()BEL(GOAL())INT()CAP(INT())？
A6: INT( ) GOAL(INT())？
A7: done( ) BEL(done())？
A8: INT () must ◇( INT ())?
4.3 Introduction of RES Operator and OPP Operator
The above BDI theory considers adding other operators to model the agent. However, there are still some problems that have not been involved. For example, A3: INT(does( )) does () in the above table, if under the condition of compound actions, we get: int (does (; )) Indeed (; )
In this way, the influence of the execution of the action on the execution of the subsequent action is not considered.
In order to solve this problem, this paper introduces two operators, RES and OPP, and gives the semantic description.
OR 1:
CAP(does(； )) ? [Bell (does ()) opp (does ())] RES (done ())?
Semantics: If an agent has the ability to perform composite actions? Ability of
Then at a certain moment, the agent believes that he can perform the action, has the opportunity to perform the action, believes that he can perform the action, has the opportunity to perform the action, and the result of performing the action will not lead to illegal state.
OR2:
CAP(does(； )) ? [CAP(does())OPP(does())]RES(done())？
Semantics: If an agent has the ability to perform composite actions? Ability of
Then the agent has the ability to execute the action and has the opportunity to execute the action, and the result of executing the action will not lead to illegal status.
OR3:
CAP(does(； )) ? [Bell (Cap (does ())) OPP (does ())] RES (done ())?
Semantics: What if an agent has the ability to perform composite actions? Reach the goal,
Then the agent believes that it has the ability and opportunity to execute the action, and the result of executing the action will not lead to illegal state.
……
OR7:
CAP (if? then what else))？
[ ? BEL(CAP(does()))OPP(does())]RES(done())]？
[ ? Bell (CAP (DOES ()) OPP (DOES ())] RES (Done ())]
Semantics: If an agent has the ability to perform an action, otherwise, perform an action?
Then, it is established and the agent thinks that it has the ability and opportunity to execute the action, and the result of executing the action will not lead to illegal state; Or it is not established and the agent thinks that he has the opportunity to execute the action, and the result of executing the action will not lead to illegal state.
5 Intention reasoning based on revocable logic
5. 1 problem proposition
Rao? Georgeff is modeled as a normal modal operator, so there are inevitably logical omniscient problems and side effects under the tautological implication problem.
(1) logic omniscient problem
INT()
(2) The side effects implied by tautology.
INT( ) INT(R)
(3) disjunctive expansion problem
INT( ) INT(R？ )
(4) Conjunctive separation problem
INT(R？ )INT(R)？ INT()
5.2 Theoretical basis
(1) Conolly lattice? Pollack intention modeling
Konolige and Pollack think that normal modal logic is not suitable for intention. They use scenes to express the mental state of agents. The scheme of the formula is a subset of W truth. The scheme set I in the model is the set of all unequal intention formulas, and only the formula equivalent to one of the formulas in I is the agent's intention. So, if int () can be represented by int in this model,
2 Rao? George is going to be a model.
Rao? In order to overcome the above-mentioned problems (3) and (4), the unique modal form of BDI is introduced, which is essentially similar to KonoligePollack theory.
5.3 introduction of de feasible logic
From another angle, this paper studies the intention model based on policy by using the revocable logic based on non-monotonic logic.
Non-monotonic logic is a kind of reasoning: the theorem set of reasoning system does not increase monotonously with the reasoning process, and the new theory is likely to deny and change some original theorems. Nonmonotonic reasoning is a tool to deal with incomplete knowledge.
Revocable logic is a non-monotonic reasoning. The application of revocable logic in intention reasoning is to allow agent to use existing incomplete knowledge to reason without fully understanding the environment.
6 Summary and the next step
Modeling of (1) multi-agent system
(2)tableaux system
(3) Dynamic maintenance of multi-agent system
More information:
"ERP Chief Designer Forum" ()