CN113255914A - Method for structurally representing intelligent agent target implementation process - Google Patents

Abstract

The invention relates to a method for structurally representing the target implementation process of an intelligent agent, which comprises the steps of constructing a multi-element group G based on a target planning graph, wherein the multi-element group G comprises a set of nodes in the target planning graph and directed edge sets of different types among the nodes, generating the target planning graph based on the target of any intelligent agent, selecting a plan and constructing an example to realize the structural execution of the target implementation process of the intelligent agent; the structure for representing the intelligent agent target implementation process more flexibly is provided, a specific use mode of the structure is given, the graph structure is used for representing the partial order execution dependency relationship among the steps based on the directed acyclic graph, the strict full order relationship in the traditional target plan tree is replaced by the partial order execution dependency relationship, the steps without the order dependency relationship are allowed to be executed in an alternative execution order or even in parallel, a more accurate target implementation mode can be represented more objectively, and the repeated reproduction can be realized on the premise that all nodes are generated based on the target plan library.

Description

Method for structurally representing intelligent agent target implementation process

Technical Field

The invention relates to the technical field of computer systems based on specific computing models, in particular to a method for structurally representing an intelligent object implementation process in the field of artificial intelligence.

Background

A Multi-Agent System (MAS) is a System that implements complex intelligence through interaction and cooperation between agents, is used to solve a problem that a single Agent is difficult or impossible to solve, and is widely used in important fields such as urban transportation and industrial manufacturing.

As a common agent structure, a BDI model constructs the mental state of an agent by defining Belief (Belief), goal (Desire) and Intention (intent), and explains the autonomous behavior of the agent in the environment according to the change of the mental state, the BDI agent constructed based on the BDI model has strong autonomous ability, interaction ability and strain ability in a dynamic environment, and the autonomous behavior has good verifiability, and is widely applied to the fields of military, aerospace and the like.

The BDI agent selects different plans (Plan) from a predetermined Plan library to achieve the goal; plan (Plan) specifies the preconditions (preconditions) for its execution and the steps required for its completion, which may be actions (actions) that the agent can directly perform or sub-targets that need to be implemented by the Plan; in the process of achieving the goal, the step that the agent promises to be performed is called the intention (interaction) of the agent to achieve the goal, namely what needs to be done to achieve the goal.

Solving the intent selection problem and the plan selection problem requires a complete, objective, repeatable analysis of the implementation process and steps of the agent objectives.

The most common way to represent the target implementation process at present is a target plan tree (goal-plan tree) structure proposed by Thangarajah et al, and the method adopts an and or tree to represent the relationship between the intelligent object target and the plan, so that the plan selection problem in the deliberate process of the intelligent object is converted into the search and traversal problem of the tree, and through the target plan tree, the intelligent object can track the complete process of target implementation and acquire all possible target implementation paths.

However, the currently defined target planning tree is limited by its own structural features, so that each step in the plan must be executed strictly according to a preset sequence, and the flexibility and expressiveness are insufficient, and the real execution situation cannot be represented. In a real environment, an agent can adjust the execution sequence of target implementation steps at any time under the condition that conditions allow, when the operation sequences of some nodes are exchanged, the change of the sequence does not affect the final target implementation, and a target plan tree cannot be used for the representation, so that the application of the target plan tree in a real scene is limited finally.

Disclosure of Invention

The invention solves the problems in the prior art and provides an optimized method for structurally representing the intelligent agent target realization process.

The technical scheme adopted by the invention is that the method for structurally representing the intelligent agent target implementation process comprises the following steps:

step 1: construction of a target plan based tuple G = (V, E)_c,E_b,E_o) Where V is the set of nodes in the target planning graph, E_c、E_b、E_oRespectively representing different types of directed edge sets among the nodes;

in the invention, V comprises a target node, a plan node and an action node; for a goal, one or more plans may be included, with a goal being completed if any plan is completed; for a plan, which may include sub-objectives and actions, and one or more sub-plans under the sub-objectives, the plan is completed only if the sub-objectives and actions are completed; also, all sub-goals and actions are referred to as steps.

In the invention, the targets are divided into a top-level target and sub-targets, the top-level target represents the highest state that an intelligent agent wants to reach, all plans, sub-targets and actions in the graph are used for realizing top-level target services, and the sub-targets are generated in the execution process of the plans without specifying the realized plans in advance.

In the present invention, set E_cThe directed edge in (1) refers to pointing from a target node x to a planning node y, and satisfies (x, y) epsilon E_cMeaning that y is a possible plan to implement x,is a corresponding relation; specifically, for any one target G_iWhether it is a top level target or a sub-target, there is at least one related plan in the plan library, one of which needs to be selected_jTo implement G_iRepresented in the graph structure as a slave target node N_GiStarting from all for implementing G_iPlanning node N of_PjAll have a directed edge (N)_Gi,N_Pj)∈E_cAnd represents the correspondence between the target and the plan.

In the present invention, set E_bThe directed edge in (1) refers to pointing from a planning node x to an action or target node y, and satisfies (x, y) epsilon E_bY is the step in plan x, which is a dependency; in particular, a plan P is completed_jAll execution steps contained in the plan, plan P, need to be executed_jThe steps in (1) can be actions that can be performed directly or sub-objectives that require selection of a plan to implement, represented in the graph structure as a slave plan node N_PjGo out to all action or target nodes N it contains_SkAll have a directed edge (N)_Pj,N_Sk)∈E_bThe dependency between the step and the plan is indicated.

In the present invention, set E_oThe directed edge refers to the point from one target or action node x to another target or action node y in the same plan, and satisfies (x, y) epsilon E_oIndicating that x must be executed before y, and performing the relation in a partial order; in particular, there may be dependencies or order relationships between steps in the same plan, such as step S_k+1Need to be in step S_kAnd step S_k+2Then executed, assuming plan P_jAny one of the steps S_kAll have a set of pre-stagesφ(S _k )So that S_kIn thatφ(S _k )All steps in (1) are executed after being executed, thenφ(S _k )Step (2) is called S_kA preliminary step of (S)_kAll the preceding steps of (1) have partial order execution relation with the preceding steps, and the step nodes are arranged between the preceding stepsIs represented in the target plan graph structure as for plan P_jNode N of any step_SkIf, ifφ(S _k )If not, there is a directed edge slave node N_SmPointing to node N_SkAnd S_m∈φ(S _k )。

In the present invention, any group (x, y) can only belong to E_c、E_b、E_oOne kind of (1).

In the present invention, the target plan graph is a directed acyclic graph.

Step 2: any intelligent agent corresponds to a target plan library, and a target plan graph is generated based on the target of any intelligent agent;

and step 3: selecting a plan and constructing an example based on the target planning map;

and 4, step 4: structured execution of an agent object implementation process is achieved.

Preferably, in step 1, all V, E are in the initial state_c、E_b、E_oAre all empty sets.

Preferably, in step 1, the nodes are a set including all target nodes, plan nodes, and action nodes in the target plan graph, and the different types of directed edges include a corresponding relationship, a dependent relationship, and a partial order execution relationship.

Preferably, in step 2, the generating of the target plan map for any agent includes the following steps:

step 2.1: constructing an empty Plan Set (PS), and adding all plans in a target plan library of any intelligent agent into the PS;

step 2.2: judging whether the plan set PS is empty, if yes, proceeding step 2.6, otherwise, taking out a plan P from the PS_iGenerating a planning node N_PiAnd plan node N_PiAdding the data to a set V in the tuple G, wherein i is the serial number of the plan in the target plan set, and i is more than or equal to 0;

step 2.3: judgment plan P_iEach of the steps S;

if S is an action A_nThen generate the corresponding action node N_AnIs a reaction of N_AnAdding to set V while adding doublets (N)_Pi, N_An) Join to set E_b；

If S is a sub-target G_nThen, the set V is searched to determine whether there is a representation G_nNode N of_Gn(ii) a If not present or if present but N_GnIf the degree of income is 0, the corresponding target node N is generated or reserved_GnIs a reaction of N_GnAdd to set V and add doublet (N)_Pi, N_Gn) Join to set E_bIf present and satisfies N_GnIf the degree of income is greater than 0, then N is added_GnCopying one part of all types of nodes and edges in the whole connected set, generating new nodes and edges with different names and the same expression content, and respectively adding the new nodes and edges into the sets with corresponding types in the multi-element group G;

step 2.4: connecting the step nodes with the partial order execution relation according to the partial order execution relation among the step nodes;

step 2.5: find plan P_iObject G to be achieved_jGenerating a target node N_GjJudging the target node N_GjIf not, the target node N is determined to be in the set V_GjAdding into the set V, if existing, only one target node N exists_GjThen a doublet (N) of edges will be represented_Gj, N_Pi) Join set E_cIf there is more than one target node N_GjThen N is added_PiCopying one copy of all types of nodes and edges in the connected set to generate new nodes and edges with different names and the same type, adding the new nodes and edges into the set of corresponding types in the multi-element group G respectively, and representing the two-element group (N) of the edge_Gj, N_Pi) Join set E_c(ii) a Completion P_iDeleting P in the plan set PS_iReturning to the step 2.2;

step 2.6: and checking the target plan graph, deleting logic redundant edges, and finishing the generation of the target plan graph.

Preferably, in the step 2.4, if there is a different step node N_SmAnd N_Sk，N_SmAnd N_SkHas a partial order execution relation therebetween, and S_mIs S_kThe precondition of (2) is to use the binary group (N)_Sm, N_Sk) Join to set E_o。

Preferably, the step 3 comprises the steps of:

step 3.1: the intelligent agent determines a top target G according to the requirement₀；

Step 3.2: search all generated target planning graphs for G₀If not with respect to G₀The target planning map discards the current target and selects a new target G₀Repeating the step 3.2, otherwise, carrying out the next step;

step 3.3: tuple G = (V, E) based on selected target plan_c,E_b,E_o) For G₀Instantiation is carried out;

step 3.4: from the target G₀Corresponding target node N_G0Initially, a set of executable steps EX is established_NG0Adding currently executable target node and/or action node to EX_NG0；

Step 3.5: random slave N_G0Selects an executable plan node N from the plan child nodes_Pk。

Preferably, any one of the nodes is provided with a state value; the status values include default, in execution, success, and failure.

Preferably, the initial values of all nodes are defaults.

Preferably, the step 4 comprises the steps of:

step 4.1: will N_PkUpdate to executive status of (EX), delete EX_NG0N in (1)_G0；

Step 4.2: will P_kAll step nodes S without pre-step in the set of executable steps EX_NG0Performing the following steps;

step 4.3: selection of EX_NG0Any step node in the step is executed, and the state of the step node is updated to be in execution;

if the step node is an action node, directly executing;

if the node in the step is the sub-target node N_GiTo establish a portable deviceColumn step set EX_NGiThe current sub-target node N_GiIn executable target node or action node joins EX_NGiFor EX_NGiRepeat step 3.5 and add N_GiFrom EX_NG0Removing EX_NGiAdding EX_NG0Performing the following steps;

step 4.4: updating executable step set EX_NG0；

If step node S_jIf the execution is successful, the step node S is executed_jIs updated to be successful and removed from the corresponding set of executable steps, updates the set of executable steps EX_NG0(ii) a Judging with step node S_jIf the step node of the pre-step has other non-executed pre-steps, directly repeating the step 4.3 if the step node is not the other non-executed pre-step, otherwise, directly using the step node S_jAdding a step node to EX for a step of a preceding step_NG0In step (2), repeatedly executing step 4.3; set of executable steps EX if updated_NG0If the node is empty, setting the states of the plan node and the corresponding target node as successful to finish the target;

if step node S_jIf the execution fails, the step node S_jIs updated to failure, the state of the corresponding planning node is also updated to failure, and S is cleared_jIs located in the set EX_NGi（S_j∈ EX_NGi) (ii) a Selecting a target node N belonging to the same_GiGo back to step 4.1 for N_GiStep 4.1 and step 4.2 are executed, and step 4.3 is normally executed again; repeating step 4.3 until there are no step nodes that can be executed;

and if all plans fail, updating the state of the corresponding target node to fail.

The invention relates to a method for expressing the goal realizing course of intelligent agent structurally and optimally, construct the multivariate group G based on goal planning chart, the multivariate group includes the set of node in the goal planning chart and different kinds of directed edge sets among the node, the goal based on any intelligent agent generates the goal planning chart, choose the plan and construct the example and then realize the structural execution of the goal realizing course of the intelligent agent; the structure for representing the intelligent agent target implementation process more flexibly is provided, a specific use mode of the structure is given, the graph structure is used for representing the partial order execution dependency relationship among the steps based on the directed acyclic graph, the strict full order relationship in the traditional target plan tree is replaced by the partial order execution dependency relationship, the steps without the order dependency relationship are allowed to be executed in an alternative execution order or even in parallel, a more accurate target implementation mode can be represented more objectively, and the repeated reproduction can be realized on the premise that all nodes are generated based on the target plan library.

The invention has the beneficial effects that:

(1) the method carries out graph structural representation on the targets, plans and actions according to the logical relationship of the targets, plans and actions, and sets the front step nodes of the step nodes in the plans to be a plurality of rather than one nodes, thereby forming a graph structure, eliminating the constraint of strict sequential linear execution and better simulating the implementation process of the intelligent agent target;

(2) because no strict linear execution sequence exists among the step nodes, and no strict sequence constraint limitation exists when the logical relation is expressed, the method can be widely applied to different complex scenes, and the problem of poor expressiveness of the traditional target planning tree is solved;

(3) according to the accessibility of each node in the plan body, whether the nodes in each step in a plan can be parallel or not can be clearly known, so that various different choices are provided for the execution sequence, the flexibility is greatly improved, and the method can be used in the artificial intelligence field such as AI development and the like.

Drawings

FIG. 1 is a schematic view of a target plan of the present invention; in the figure:

the dashed edges represent dependencies between steps and plans, with directions pointing from the plan node to the step node (action or sub-goal);

the solid line edge with double arrows represents the partial order relation among the steps, and the direction is pointed to the node depending on the step by the step node;

the solid line edge with a single arrow represents the corresponding relation between the target and the plan, and the direction is pointed to the plan node by the target node;

FIG. 2 is a flow chart of the present invention.

Detailed Description

The present invention is described in further detail with reference to the following examples, but the scope of the present invention is not limited thereto.

The invention relates to a method for structurally representing the realization process of an intelligent object, wherein all subscripts are used for identifying the objects, sub-objects, plans and actions which are different under the current condition, and are not serial numbers.

As shown in FIG. 1, there is a top level target G₀The top level goal may be accomplished through plan P₀Or plan P₁To achieve plan P₀Or plan P₁Any of which are considered to reach the top level goal; to plan P₀For example, action A needs to be completed₁Action A₀Action A₂And sub-target G₁To complete action A₁Before completing action A₀To complete action A₂Before completing action A₀And G₁So by analogy, the step that actually needs/can be executed first is action A₃And action A₀At this time, plan P₂Completion sub-goal G₁Is achieved, then action A is completed₁And action A₂Based on the executed action A₁Action A₀Action A₂And sub-target G₁Execution plan P₀Finally reach the top level target G₀(ii) a Plan P₁The same is true.

The method comprises the following steps:

step 1: construction of a target plan based tuple G = (V, E)_c,E_b,E_o) Where V is the set of nodes in the target planning graph, E_c、E_b、E_oRespectively representing different types of directed edge sets among the nodes;

in step 1, all V, E are in the initial state_c、E_b、E_oAre all empty sets.

In the step 1, the nodes are a set including all target nodes, plan nodes and action nodes in the target planning graph, and the different types of directed edges include a corresponding relationship, a subordinate relationship and a partial order execution relationship.

Step 2: any intelligent agent corresponds to a target plan library, and a target plan graph is generated based on the target of any intelligent agent;

in step 2, generating the target plan map for any agent includes the following steps:

step 2.1: constructing an empty Plan Set (PS), and adding all plans in a target plan library of any intelligent agent into the PS;

step 2.2: judging whether the plan set PS is empty, if yes, proceeding step 2.6, otherwise, taking out a plan P from the PS_iGenerating a planning node N_PiAnd plan node N_PiAdding the data to a set V in the tuple G, wherein i is the serial number of the plan in the target plan set, and i is more than or equal to 0;

step 2.3: judgment plan P_iEach of the steps S;

if S is an action A_nThen generate the corresponding action node N_AnIs a reaction of N_AnAdding to set V while adding doublets (N)_Pi, N_An) Join to set E_b；

If S is a sub-target G_nThen, the set V is searched to determine whether there is a representation G_nNode N of_Gn(ii) a If not present or if present but N_GnIf the degree of income is 0, the corresponding target node N is generated or reserved_GnIs a reaction of N_GnAdd to set V and add doublet (N)_Pi, N_Gn) Join to set E_bIf present and satisfies N_GnIf the degree of income is greater than 0, then N is added_GnCopying one part of all types of nodes and edges in the whole connected set, generating new nodes and edges with different names and the same expression content, and respectively adding the new nodes and edges into the sets with corresponding types in the multi-element group G;

step 2.4: connecting the step nodes with the partial order execution relation according to the partial order execution relation among the step nodes;

in step 2.4, if there are different step nodes N_SmAnd N_Sk，N_SmAnd N_SkHas a partial order execution relationship therebetween, andS_mis S_kThe precondition of (2) is to use the binary group (N)_Sm, N_Sk) Join to set E_o。

Step 2.5: find plan P_iObject G to be achieved_jGenerating a target node N_GjJudging the target node N_GjIf not, the target node N is determined to be in the set V_GjAdding into the set V, if existing, only one target node N exists_GjThen a doublet (N) of edges will be represented_Gj, N_Pi) Join set E_cIf there is more than one target node N_GjThen N is added_PiCopying one copy of all types of nodes and edges in the connected set to generate new nodes and edges with different names and the same type, adding the new nodes and edges into the set of corresponding types in the multi-element group G respectively, and representing the two-element group (N) of the edge_Gj, N_Pi) Join set E_c(ii) a Completion P_iDeleting P in the plan set PS_iReturning to the step 2.2;

step 2.6: and checking the target plan graph, deleting logic redundant edges, and finishing the generation of the target plan graph.

In the present invention, step 2.1, the tuple G is actually needed to be constructed initially based on the agent's target plan library, in the present invention the default G exists, and in the output state, V, E in G_c、E_bAnd E_oAre all empty sets; the target plan library is a plurality of current individual plans, and when a target needs to be realized, the corresponding plan capable of realizing the target is searched in the target plan library.

In the invention, step 2 is to find out all the associated nodes which can achieve the goal, including plan and actions and sub-goals under the plan.

In the present invention, deleting the logical redundant edge in step 2.6 is a conventional technique in the art, that is, for a partial order execution relationship structure of the logical relationship of each step node in a plan, if there is a directed edge e satisfying the basic construction condition of G, and even if it is deleted, the logical constraint of the sequential execution between each step node is still unchanged, then e is called asIs a logical redundant edge; for example, step node A in FIG. 1₅、A₆And G₃，A₆The pre-step node of (A) is₅And G₃And A is₅The pre-step node of (2) is G₃At this time A₅To A₆The directed edge can be deleted, and the deletion of the edge does not change the logic constraint relation between nodes; in order to simplify the logic constraint relation between nodes executed successively, all the logically redundant edges need to be deleted. After all the deletable edges are deleted, the simplest target planning graph can be obtained, the simplest target planning graph is more concise in expression, and after the redundant logic edges are deleted, when each node updates the completion condition of the preposed node, the nodes related to the redundant logic edges do not need to be processed.

And step 3: selecting a plan and constructing an example based on the target planning map;

the step 3 comprises the following steps:

step 3.1: the intelligent agent determines a top target G according to the requirement₀；

Step 3.2: search all generated target planning graphs for G₀If not with respect to G₀The target planning map discards the current target and selects a new target G₀Repeating the step 3.2, otherwise, carrying out the next step;

step 3.3: tuple G = (V, E) based on selected target plan_c,E_b,E_o) For G₀Instantiation is carried out;

any node is provided with a state value; the status values include default (default), executing (executing), success (success), and failure (fail).

All nodes have default initial values.

Step 3.4: from the target G₀Corresponding target node N_G0Initially, a set of executable steps EX is established_NG0Adding currently executable target node and/or action node to EX_NG0；

Step 3.5: random slave N_G0Selects an executable plan node N from the plan child nodes_Pk。

In the present invention, EX is used when execution is started_NG0= N_G0And N is_G0Is updated to be in execution.

And 4, step 4: structured execution of an agent object implementation process is achieved.

The step 4 comprises the following steps:

step 4.1: will N_PkUpdate to executive status of (EX), delete EX_NG0N in (1)_G0；

Step 4.2: will P_kAll step nodes S without pre-step in the set of executable steps EX_NG0Performing the following steps;

step 4.3: selection of EX_NG0Any step node in the step is executed, and the state of the step node is updated to be in execution;

if the step node is an action node, directly executing;

if the node in the step is the sub-target node N_GiEstablishing a set of executable steps EX_NGiThe current sub-target node N_GiIn executable target node or action node joins EX_NGiFor EX_NGiRepeat step 3.5 and add N_GiFrom EX_NG0Removing EX_NGiAdding EX_NG0Performing the following steps;

step 4.4: updating executable step set EX_NG0；

If step node S_jIf the execution is successful, the step node S is executed_jIs updated to be successful and removed from the corresponding set of executable steps, updates the set of executable steps EX_NG0(ii) a Judging with step node S_jIf the step node of the pre-step has other non-executed pre-steps, directly repeating the step 4.3 if the step node is not the other non-executed pre-step, otherwise, directly using the step node S_jAdding a step node to EX for a step of a preceding step_NG0In step (2), repeatedly executing step 4.3; set of executable steps EX if updated_NG0If the node is empty, setting the states of the plan node and the corresponding target node as successful to finish the target;

if step node S_jExecution failureThen step node S_jIs updated to failure, the state of the corresponding planning node is also updated to failure, and S is cleared_jIs located in the set EX_NGi（S_j∈EX_NGi) (ii) a Selecting a target node N belonging to the same_GiGo back to step 4.1 for N_GiStep 4.1 and step 4.2 are executed, and step 4.3 is normally executed again; repeating step 4.3 until there are no step nodes that can be executed;

and if all plans fail, updating the state of the corresponding target node to fail.

In the present invention, EX is updated_NG0And adding N_PkNode of executable step (1), N_PkUpdate to in-flight and delete N_G0Starting execution plan N_PkWhen N is present_PkSome of the steps S are without pre-stages, when these steps S are executable, all step nodes are added to the executable step set EX in turn_NG0In the method, step nodes with default states are all executable, namely, no precedence order constraint exists, and EX is selected to be executed_NG0Any one of the step nodes in (1); for action nodes it can be executed directly, while for sub-target nodes it is necessary to establish a new, included set of executable steps EX_NGiAfter completion, EX is_NGiAdding EX_NG0And (4) performing updating.

In the present invention, in action A₀Action A₂Sub-target G₁For example, when action A₀After the execution is completed, action A₂Still cannot be executed, must be taken as the child target G₁After the execution is finished, action A can be executed₂Therefore, all unexecuted prestage nodes need to be added to EX_NG0And step 4.3 is performed.

In the invention, the first step is continuously and repeatedly completed, and a new executable set related to a possibly existing sub-target is added until EX_NG0An empty set means that all execution step nodes in a plan node that completes the top-level target are successfully executed, and a successful execution of the corresponding plan means that the state of the target node is changedAnd the new execution is successful.

In the invention, in the actual operation, a counting mark can be set, namely the top target is G₀The first sub-target is G₁And numbered sequentially when G is complete_iAnd i is 0, the top level goal is achieved.

Claims (9)

1. A method for structured representation of an intelligent agent object realization process, characterized by: the method comprises the following steps:

step 1: construction of a target plan based tuple G = (V, E)_c,E_b,E_o) Where V is the set of nodes in the target planning graph, E_c、E_b、E_oRespectively representing different types of directed edge sets among the nodes;

step 2: any intelligent agent corresponds to a target plan library, and a target plan graph is generated based on the target of any intelligent agent;

and step 3: selecting a plan and constructing an example based on the target planning map;

and 4, step 4: structured execution of an agent object implementation process is achieved.

2. A method for structured representation of an intelligent object realization process according to claim 1, characterized in that: in step 1, all V, E are in the initial state_c、E_b、E_oAre all empty sets.

3. A method for structured representation of an intelligent object realization process according to claim 1, characterized in that: in the step 1, the nodes are a set including all target nodes, plan nodes and action nodes in the target planning graph, and the different types of directed edges include a corresponding relationship, a subordinate relationship and a partial order execution relationship.

4. A method for structured representation of an intelligent object realization process according to claim 1, characterized in that: in step 2, generating the target plan map for any agent includes the following steps:

step 2.1: constructing an empty Plan Set (PS), and adding all plans in a target plan library of any intelligent agent into the PS;

step 2.2: judging whether the plan set PS is empty, if yes, proceeding step 2.6, otherwise, taking out a plan P from the PS_iGenerating a planning node N_PiAnd plan node N_PiAdding the data to a set V in the tuple G, wherein i is the serial number of the plan in the target plan set, and i is more than or equal to 0;

step 2.3: judgment plan P_iEach of the steps S;

if S is an action A_nThen generate the corresponding action node N_AnIs a reaction of N_AnAdding to set V while adding doublets (N)_Pi, N_An) Join to set E_b；

If S is a sub-target G_nThen, the set V is searched to determine whether there is a representation G_nNode N of_Gn(ii) a If not present or if present but N_GnIf the degree of income is 0, the corresponding target node N is generated or reserved_GnIs a reaction of N_GnAdd to set V and add doublet (N)_Pi, N_Gn) Join to set E_bIf present and satisfies N_GnIf the degree of income is greater than 0, then N is added_GnCopying one part of all types of nodes and edges in the whole connected set, generating new nodes and edges with different names and the same expression content, and respectively adding the new nodes and edges into the sets with corresponding types in the multi-element group G;

step 2.4: connecting the step nodes with the partial order execution relation according to the partial order execution relation among the step nodes;

step 2.5: find plan P_iObject G to be achieved_jGenerating a target node N_GjJudging the target node N_GjIf not, the target node N is determined to be in the set V_GjAdding into the set V, if existing, only one target node N exists_GjThen a doublet (N) of edges will be represented_Gj, N_Pi) Join set E_cIf there is more than one target node N_GjThen N is added_PiCopying one copy of all types of nodes and edges in the connected set to generate new nodes and edges with different names and the same type, adding the new nodes and edges into the set of corresponding types in the multi-element group G respectively, and representing the two-element group (N) of the edge_Gj, N_Pi) Join set E_c(ii) a Completion P_iDeleting P in the plan set PS_iReturning to the step 2.2;

step 2.6: and checking the target plan graph, deleting logic redundant edges, and finishing the generation of the target plan graph.

5. The method for structured representation of intelligent agent object realization process according to claim 4, characterized in that: in step 2.4, if there are different step nodes N_SmAnd N_Sk，N_SmAnd N_SkHas a partial order execution relation therebetween, and S_mIs S_kThe precondition of (2) is to use the binary group (N)_Sm, N_Sk) Join to set E_o。

6. A method for structured representation of an intelligent object realization process according to claim 1, characterized in that: the step 3 comprises the following steps:

step 3.1: the intelligent agent determines a top target G according to the requirement₀；

Step 3.2: search all generated target planning graphs for G₀If not with respect to G₀The target planning map discards the current target and selects a new target G₀Repeating the step 3.2, otherwise, carrying out the next step;

step 3.3: tuple G = (V, E) based on selected target plan_c,E_b,E_o) For G₀Instantiation is carried out;

step 3.4: from the target G₀Corresponding target node N_G0Initially, a set of executable steps EX is established_NG0Adding currently executable target node and/or action node to EX_NG0；

Step 3.5: random slave N_G0Plan child node ofTo select an executable plan node N_Pk。

7. The method for structured representation of intelligent agent object realization process according to claim 6, characterized in that: any node is provided with a state value; the status values include default, in execution, success, and failure.

8. A method for structured representation of an intelligent object realization process according to claim 7, characterized in that: all nodes have default initial values.

9. The method for structured representation of intelligent agent object realization process according to claim 6, characterized in that: the step 4 comprises the following steps:

step 4.1: will N_PkUpdate to executive status of (EX), delete EX_NG0N in (1)_G0；

Step 4.2: will P_kAll step nodes S without pre-step in the set of executable steps EX_NG0Performing the following steps;

step 4.3: selection of EX_NG0Any step node in the step is executed, and the state of the step node is updated to be in execution;

if the step node is an action node, directly executing;

if the node in the step is the sub-target node N_GiEstablishing a set of executable steps EX_NGiThe current sub-target node N_GiIn executable target node or action node joins EX_NGiFor EX_NGiRepeat step 3.5 and add N_GiFrom EX_NG0Removing EX_NGiAdding EX_NG0Performing the following steps;

step 4.4: updating executable step set EX_NG0；

If step node S_jIf the execution is successful, the step node S is executed_jIs updated to be successful and removed from the corresponding set of executable steps, updates the set of executable steps EX_NG0(ii) a The step of judgmentNode S_jIf the step node of the pre-step has other non-executed pre-steps, directly repeating the step 4.3 if the step node is not the other non-executed pre-step, otherwise, directly using the step node S_jAdding a step node to EX for a step of a preceding step_NG0In step (2), repeatedly executing step 4.3; set of executable steps EX if updated_NG0If the node is empty, setting the states of the plan node and the corresponding target node as successful to finish the target;

if step node S_jIf the execution fails, the step node S_jIs updated to failure, the state of the corresponding planning node is also updated to failure, and S is cleared_jIs located in the set EX_NGi(ii) a Selecting a target node N belonging to the same_GiGo back to step 4.1 for N_GiStep 4.1 and step 4.2 are executed, and step 4.3 is then executed; repeating step 4.3 until there are no step nodes that can be executed;

and if all plans fail, updating the state of the corresponding target node to fail.

Images