JPH04271467A - Device for preparing hierarchical schedule - Google Patents

Abstract

PURPOSE:To excessively reduce the labor of a man and to obtain a hierarchical schedule generating device which can generate hierarchical structure being always optimum by automatically generating the hierarchical structure of schedule generating knowledge. CONSTITUTION:The above device is provided with a hierarchical structure generating part 2. The generating part 2 analyzes the shard relation of additional condition among plural schedule generating knowledge, decides the attainment difficulty degree of assumption condition and generates the abstract hierarchical structure of schedule generating knowledge based on the result of analysis and decision. A schedule is stepwise generated from an abstract hierarchy to a minute hierarchy based on the generated abstract hierarchy structure.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a hierarchical plan generation device for automatically generating a work plan for, for example, an industrial robot.

0002

[Prior Art] As a knowledge representation for plan generation, STRIPS (Stanfordrese
arch institute problem so
The operator type of the lver) system is the basis. The STRIPS system operator consists of an operation name, preconditions, deletion conditions, and additional conditions.
Given the initial conditions and target conditions, the IPS system
Plan generation is executed by selecting and combining operators to achieve the target conditions from the knowledge base.

[0003] Also, based on this STRIPS,
ABSTRIPS (abstractSTRIPS), which introduces a hierarchical method to efficiently generate plans
system has been realized. The ABSTRIPS system basically has plan generation knowledge in the same expression format as the STRIPS system, but it structures the plan generation knowledge into an abstract hierarchy and generates plans step by step from the most abstract level to the most detailed level. It is. The abstract hierarchical structure is given based on the urgent value, which is the necessity of achieving the items appearing in the preconditions of the plan generation knowledge.

0004

However, in the conventional ABSTRIPS system described above, it is necessary for humans to create and provide an abstract hierarchical structure based on experience in order to handle plan generation knowledge in a hierarchical manner. This increases the burden on humans, and depending on the problem, calculation efficiency may deteriorate significantly.

[0005] Therefore, an object of the present invention is to automatically generate a hierarchical structure of planning generation knowledge, thereby greatly reducing the burden on humans, and creating a hierarchical structure that can always generate an optimal hierarchical structure. An object of the present invention is to provide a comprehensive plan generation device. Another object of the present invention is to provide a hierarchical plan generation device that can generate efficient plans without generating wasteful candidates.

[0006]

[Means for Solving the Problems] In order to achieve the above object, the present invention provides a plan for achieving the target condition from the initial condition when initial conditions, target conditions, and plan generation knowledge are given. A hierarchical plan generation device that uses generated knowledge is provided with a hierarchical structure generation means, which analyzes the shared relationship of additional conditions among multiple pieces of plan generation knowledge and determines the degree of difficulty in achieving the preconditions. , an abstract hierarchical structure of plan generation knowledge is generated based on these analysis results and judgment results, and a plan is created step by step from the abstract hierarchy to the detailed hierarchy based on this generated abstract hierarchical structure. It is.

[0007] In order to achieve the above-mentioned other objects, another aspect of the present invention provides a plurality of plan generation knowledge having object types and constraint conditions in addition to operation names, preconditions, deletion conditions, and addition conditions. When creating a plan by selectively using each plan generation knowledge stored in the plan generation knowledge storage means, the constraint resolving means calculates the above based on the object type and constraint conditions. This is designed to assist the selection process of plan generation knowledge.

[0008] Furthermore, in order to achieve each of the above-mentioned objects, another present invention is provided with a hierarchical structure generation means, which analyzes the shared relationship of additional conditions among a plurality of plan generation knowledge and determines the degree of difficulty of achieving the preconditions. An abstract hierarchical structure of plan generation knowledge is generated based on these analysis results and judgment results, and a plan is created step by step from the abstract hierarchy to the detailed hierarchy based on this generated abstract hierarchical structure. So,
In addition, the above plan generation knowledge is assumed to have object types and constraints in addition to operation names, preconditions, deletion conditions, and additional conditions, and when creating the above plan, the above object types and constraints are set using the constraint resolution means. Based on this, the selection process of the plan generation knowledge is assisted.

[0009]

[Operation] As a result, according to the device of the present invention, when plan generation knowledge is received, the knowledge is classified hierarchically based on the shared relationship of additional conditions of each knowledge, and the difficulty level of achieving the preconditions is calculated and abstracted. A hierarchical structure is generated. In other words, an abstract hierarchical structure is automatically generated based on the given plan generation knowledge. Then, given the initial and target conditions,
Plans are generated step by step from the most abstract level to the most detailed level based on the abstract hierarchical structure. Therefore, there is no need for a human to create an abstract hierarchical structure in advance and provide it to the device, which greatly reduces the burden on the user. Furthermore, since the optimal abstract hierarchical structure is generated each time according to the contents of the plan generation knowledge, problems such as deterioration of calculation efficiency due to problems are reduced.

According to another aspect of the present invention, in addition to operation names, preconditions, deletion conditions, and addition conditions, object types and constraint conditions are prepared as elements constituting plan generation knowledge, and these are used when creating a plan. The selection operation of plan generation knowledge is performed with reference to the object type and constraint conditions. Therefore, compared to the case where object types and constraints are not used, it is possible to efficiently generate a plan without generating wasteful candidates during inference. Furthermore, compared to the case where, for example, information corresponding to object types and constraint conditions is provided separately from the plan generation knowledge in principle, a plan can be created efficiently with a smaller amount of calculation.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing the functional configuration of a hierarchical plan generation device according to an embodiment of the present invention. This device includes a basic plan generation knowledge storage section 1, a knowledge analysis section 2 for generating an abstract hierarchical structure of the basic plan generation knowledge stored in this basic plan generation knowledge storage section 1, and this knowledge analysis section 2. An abstract hierarchical structure storage unit 3 for storing the generated abstract hierarchical structure, a problem input unit 4 for inputting a problem P consisting of initial conditions and target conditions, a plan generation basic processing unit 5, and a plan generation basic processing unit 5. A constraint resolution processing section 6 for assisting the processing of the basic processing section 5, and a plan storage section 7 for remembering the plan generated by the plan generation basic processing section 5.
It is equipped with

The basic plan generation knowledge storage section 1 stores a plurality of basic plan generation knowledge pieces in advance. These basic plan generation knowledge includes an operation name, preconditions for realizing this operation, application results of the operation, conditions to be deleted, conditions to be added, and the type of operation target (object type) and constraints on the concrete value of this operation target. FIG. 2 shows an example of the specific contents.

[0013] The knowledge analysis section 2 analyzes the sharing relationship of additional conditions between the basic plan generation knowledge stored in the basic plan generation knowledge storage section 1, and extracts the basic plan generation knowledge from the analysis result. The above-described hierarchical structure is modified by hierarchically classifying and calculating the degree of difficulty of achieving the prerequisites. FIG. 3 shows an example of an abstract hierarchical structure obtained by the knowledge analysis unit 2, and each hierarchy is expressed by a combination of a condition name and a numerical value representing the hierarchy level.

When the initial conditions and target conditions are given by the problem input unit 4, the plan generation basic processing unit 5 generates a plan from an abstract hierarchical level based on the abstract hierarchical structure stored in the abstract hierarchical structure storage unit 3. A predetermined process is performed to generate a plan step by step at detailed hierarchical levels. The constraint resolution processing unit 6 refers to the object type and constraint conditions of the basic plan generation knowledge and performs auxiliary processing for efficiently selecting basic plan generation knowledge at the time of plan generation.

Next, the operation of the apparatus configured as described above will be explained with reference to a flowchart. First, the knowledge analysis section 2 performs the following processing. That is, this process is divided into three major routines as shown in FIGS. 3, 4, and 5. One is a routine 1 that classifies each piece of knowledge stored in the basic plan generation knowledge storage unit 1, a processing routine 2 that calculates an abstract hierarchy of conditions from the classification results, and a processing routine 3 that corrects the hierarchical relationship. .

First, in routine 1, step 3a is first performed.
The classification tree is S, OP is the selected basic plan generation knowledge, L
Initial settings are made to respectively define O as an operation located at a leaf of S. If the basic plan generation knowledge OP1 is now selected, the initial state of the classification tree S is empty, so the process moves from step 3b to step 3c, and the basic plan generation knowledge OP1 becomes S. Next, when the basic plan generation knowledge OP2 is selected from the basic plan generation knowledge storage unit 1, in step 3d, the operation located at the end of the classification tree S (LO
) and the additional conditions of basic plan generation knowledge OP2 are checked, and the shared condition becomes NA. Also this NA
If is not empty, the intersection of the preconditions of the basic plan generation knowledge OP2 and LO becomes NP in step 3e. If this NP is not empty, NP and N
Abstract plan generation knowledge ABOP is created from A. In step 3g, it is determined whether this abstract plan generation knowledge ABOP exists in S, and if it does not exist, LO and OP2
ABOP is placed above the ABOP, and OP from this ABOP
A link is created. On the other hand, if other abstract plan generation knowledge exists in S, the process moves to step 3i, where S is reconstructed. In other words, if any abstract hierarchy Sabi of S is more general than ABOP, then ABOP is Sa
It is placed below bi, or above bi if vice versa. When the reconfiguration of S is completed, LO is marked in step 3j.

[0017] Finally, in step 3k, all L
It is determined whether or not the O has been marked, and if there is an LO that has not been marked yet, the process returns to step 3d and the above process is repeated. On the other hand, if all LOs have been marked, the process moves to step 3l, where it is determined whether there is any unselected basic plan generation knowledge OP. If there is basic plan generation knowledge that has not been selected yet, the process returns to step 3a and the above series of processes is repeated, and when there is no unselected basic plan generation knowledge, that is, the processing for all basic plan generation knowledge is completed. , the basic plan generation knowledge classification process is completed and the process moves to routine 2.

FIG. 7 shows the basic plan generation knowledge pushbox
An example is shown in which abstract plan generation knowledge abop1 is created for and pushtodoor. Figure 8 shows the basic plan generation knowledge gotobox in contrast to Figure 7.
This shows the final result when newly added. In the example of FIG. 8, gotobox's abstract plan generation knowledge is abop2, but this abop2 is further a
It is also the abstract plan generation knowledge of bop1.

Next, in routine 2, the plan generation knowledge classification tree S of routine 1 is used to perform processing for determining an abstract hierarchy between conditions. That is, step 4
In a, S=classification tree, i=1, and P=condition list are defined. Subsequently, in step 4b, the abstract plan generation knowledge located at the highest level of the classification tree S is set as OP1, and the set of preconditions for OP1 is set as PRE. and,
In step 4c, first, level 1 is set as a prerequisite for the abstract plan generation knowledge OP1 located at the top of the classification tree S.
is assigned. Next, in step 4d, the classification tree S is lowered by one level, and the preconditions of the abstract plan generation knowledge OP are examined. If there is a precondition that has not yet been assigned a value, level 2 is assigned to this precondition. At this time, if two or more abstract plan generation knowledge OP exist at the same level, the above leveling is performed for all preconditions. The above processing is performed until the end of the classification tree S, increasing the level by one each time the level of abstract plan generation knowledge is lowered by one. By the way, when determining the abstract hierarchy level in the example of Figure 8, level 1 is specified for type, inroom, which is a prerequisite for adop2.
, push and nextto are assigned level 2, and the rest are assigned level 3.

Next, in routine 3, the appearance frequency of the preconditions of each basic plan generation knowledge stored in the basic plan generation knowledge storage section 1 is calculated, and the preconditions are sorted in descending order of frequency. Then, as a result of Routine 2, if the same level is assigned to a plurality of conditions, the level values of these conditions are modified according to the above order. For example, if level 3 is assigned to conditions P1 and P2, but P1 appears more frequently, P1 is revised to level 3, and P2 is revised to level 4. At the same time, each condition assigned to level 4 or higher is 1
Increased by increments. In the case of FIG. 8, in the preconditions for basic plan generation knowledge, "inroom" appears more frequently than "type". Therefore, level 1 is assigned to inroom, level 2 is assigned to type, and push, ne
xtto is assigned level 3, and all others are assigned level 4. Steps 5a to 5g in FIG. 5 show specific processing contents for performing the above processing. Thus, an abstract hierarchical structure is obtained from each base plan generation knowledge.

Next, a procedure for creating a plan according to the abstract hierarchical structure generated as described above will be explained. Problem P consists of a set of initial conditions and target conditions, each of which is given as a set of predicates. FIG. 9 shows a specific example of the initial conditions and target conditions when creating a robot action plan. The initial conditions are "robot is in room R1", "box B
1 is in room R2" and "Box B2 is in room R2," and the target conditions are "The robot is in room R2" and "Boxes B1 and B2 are adjacent to each other." have.

When the problem P as described above is input by the problem input unit 4, the plan generation basic processing unit 5 basically
Plan generation processing is performed as follows according to the same procedure as STRIPS. That is, as shown in FIG. 10, first, in step 10a, processing is performed to set i representing the abstract hierarchy level to "1" corresponding to the highest level, and then in step 10c, the problem input unit 4 inputs After the problem P has been set as the target condition, the process shifts to the constraint resolution processing unit 6 in this state.

As shown in FIG. 11, the constraint resolution processing unit 6 defines OP=all basic plan generation knowledge and S=candidate in step 11a, and then first defines the operation target in each plan generation knowledge from the initial conditions in step 11b. Find concrete value candidates using the object type. Next, in steps 11c, 11d, and 11f, plan generation knowledge having the target condition as an additional condition is generated as a selection candidate. This selects conditions in which the condition names are the same and the target condition's target name and the additional condition's target name are the same. And step 1
1g, a comparison is made between the preconditions of the plan generation knowledge and the current state. Similarly, the condition name and target name are compared. However, only those that do not violate the constraints are generated. And the number of conditions satisfied n
is determined in step 11h, and the plan generation knowledge with the largest number n is selected in step 11i.

Now, when the constraint resolution processing unit 6 selects the plan generation knowledge for achieving the target conditions, the plan generation basic processing unit 5 determines that all the preconditions of the plan generation knowledge are satisfied in step 10f. It is determined whether or not they are satisfied, and if all the conditions are satisfied as a result, the plan generation knowledge is applied to the initial conditions in step 10h. On the other hand, if not, the unsatisfied preconditions are newly set as target conditions in step 10g, and the above process is repeated. The process continues until it is detected in step 10i that all target conditions have been achieved. For example, the goal "Boxes B1 and B2 are adjacent" can be achieved using the basic plan generation knowledge pushbox. At the most abstract level, of the six preconditions, only the inroom condition is considered. If, inroo
If all m conditions are achieved in the initial conditions, they are applied and the initial conditions are rewritten. Furthermore, if there is an inroom condition that is not satisfied, it is set as a new goal and the same process is continued.

Now, when the search for a certain abstract hierarchy level is completed, the search is performed at step 10k, lowering the level by one. At this time, the plan obtained at the upper level is handed over to the lower level as an outline plan in step 10d, and at the lower level, according to the abstract hierarchy, plan generation between each plan generation knowledge of the outline plan is performed while taking into account the added preconditions. Let's do it.

[0026] In this way, the search is carried out to the most specific level, and once this is confirmed in step 10j,
The plan generation basic processing unit 5 and the constraint resolution processing unit 6 complete the processing operations for plan generation and output the generated plan. This plan is stored in the plan storage section 17.

As described above, in this embodiment, the knowledge analysis unit 2 automatically obtains the abstract hierarchical structure from each basic plan generation knowledge given in advance, and according to this abstract hierarchical structure, the detailed information is calculated from the abstract level. Since the plan is generated step by step, it is no longer necessary for humans to create an abstract hierarchical structure empirically and provide it to the device, which greatly reduces the burden on the user. . Also,
Since the optimal abstract hierarchical structure can be generated each time according to the content of plan generation knowledge, problems such as deterioration of calculation efficiency due to problems are also reduced.

In addition, in addition to operation names, preconditions, deletion conditions, and additional conditions, object types and constraint conditions are prepared as elements constituting the basic plan generation knowledge, and these are processed by the constraint resolution processing unit 6 when creating a plan. Since the selection operation of plan generation knowledge is performed by referring to the object type and constraints of be able to. Furthermore, compared to the case where, for example, information corresponding to object types and constraint conditions is provided separately from the plan generation knowledge in principle, a plan can be created efficiently with a smaller amount of calculation.

[0029]

[Effects of the Invention] As described in detail above, the present invention includes a hierarchical structure generation means, which analyzes the shared relationship of additional conditions among a plurality of plan generation knowledge and determines the degree of difficulty in achieving the preconditions. Then, an abstract hierarchical structure of plan generation knowledge was generated based on these analysis results and judgment results, and plans were created step by step from the abstract hierarchy to the detailed hierarchy based on this generated abstract hierarchical structure. It is something.

Therefore, according to the present invention, it is possible to automatically generate an abstract hierarchical structure of basic plan generation knowledge that guides a hierarchical plan from existing plan generation knowledge without having to provide it empirically by humans. As a result, it is possible to provide a hierarchical plan generation device that can greatly reduce the burden on humans and always generate an optimal hierarchical structure.

In another aspect of the present invention, a plurality of pieces of plan generation knowledge including object types and constraint conditions in addition to operation names, preconditions, deletion conditions, and additional conditions are stored in a plan generation knowledge storage means. When creating a plan by selectively using each piece of plan generation knowledge stored in the plan generation knowledge storage means, the constraint resolving means assists in the process of selecting the plan generation knowledge based on the object type and constraint conditions. This is how it was done. Therefore, according to another aspect of the present invention, it is possible to provide a hierarchical plan generation device that can efficiently generate a plan without generating wasteful candidates.

[Brief explanation of the drawing]

FIG. 1 is a block diagram for explaining the functional configuration of a hierarchical plan generation device in an embodiment of the present invention.

FIG. 2 is a diagram showing a specific example of representation of basic plan generation knowledge in an embodiment of the present invention.

FIG. 3 is a flowchart showing the flow of processing for generating an abstract hierarchical structure from a basic plan generation knowledge base in the apparatus shown in FIG. 1;

FIG. 4 is a flowchart showing the flow of processing for generating an abstract hierarchical structure from a basic plan generation knowledge base in the apparatus shown in FIG. 1;

FIG. 5 is a flowchart showing a process flow for generating an abstract hierarchical structure from a basic plan generation knowledge base in the apparatus shown in FIG. 1;

FIG. 6 is a diagram showing a specific example of an abstract structure inferred from the basic plan generation knowledge base.

FIG. 7 is a diagram showing an example of the content of abstract plan generation knowledge generated from basic plan generation knowledge.

FIG. 8 is a diagram showing another example of the content of abstract plan generation knowledge generated from basic plan generation knowledge.

FIG. 9 is a diagram showing an example of a problem.

10 is a flowchart showing the flow of processing for hierarchically generating a plan according to an abstract hierarchical structure in the apparatus shown in FIG. 1;

FIG. 11 is a flowchart of processing performed by a constraint resolving unit in the apparatus shown in FIG. 1;

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1... Basic plan generation knowledge storage unit, 2... Knowledge analysis unit, 3... Abstract hierarchical structure storage unit, 4... Problem input unit, 5... Plan generation basic processing unit, 6... Constraint resolution processing unit, 7... Plan storage unit.

Claims (3)

[Claims] [Claim 1] Initial conditions and target conditions are given,
And when plan generation knowledge consisting of at least an operation name, preconditions, deletion conditions, and additional conditions is given, a hierarchical method that uses the above plan generation knowledge to create a plan to achieve the target condition from the initial conditions. The plan generation device analyzes the shared relationship of additional conditions among multiple pieces of plan generation knowledge, determines the degree of difficulty in achieving the preconditions, and creates an abstract hierarchical structure of the plan generation knowledge based on these analysis and judgment results. A plan for creating a plan for achieving a target condition from the initial condition step by step from the abstract hierarchy to the detailed hierarchy based on the hierarchical structure generation means and the abstract hierarchy generated by the hierarchical structure generation means. A hierarchical plan generation device characterized by comprising: generation means. 2. A hierarchical plan generation device that uses the plan generation knowledge to create a plan for achieving a target condition from the initial conditions when initial conditions, target conditions, and plan generation knowledge are given; a plan generation knowledge storage means that stores a plurality of plan generation knowledge constituted by names, preconditions, deletion conditions, additional conditions, object types, and constraint conditions; A plan generation means that selectively uses each plan generation knowledge stored in the plan generation knowledge storage means to create a plan, and assists the plan generation knowledge selection in this plan generation means with reference to the object type and constraint conditions. A hierarchical plan generation device characterized by comprising a selection assisting means for selecting. 3. A hierarchical plan generation device that uses the plan generation knowledge to create a plan for achieving a target condition from the initial conditions when initial conditions, target conditions, and plan generation knowledge are given; Analyze the sharing relationship of additional conditions between the plan generation knowledge storage means that stores multiple plan generation knowledge consisting of names, preconditions, deletion conditions, additional conditions, object types, and constraint conditions, and the multiple plan generation knowledge. together with determining the difficulty of achieving the prerequisites,
A hierarchical structure generating means that generates an abstract hierarchical structure of plan generation knowledge based on these analysis results and judgment results, and a step-by-step process from an abstract hierarchy to a detailed hierarchy based on the abstract hierarchical structure generated by this hierarchical structure generating means. a plan generation means for selectively using each plan generation knowledge stored in the plan generation knowledge storage means to generate a plan for achieving the target condition from the initial conditions; and plan generation in the plan generation means. A hierarchical plan generation device comprising: selection assisting means for assisting knowledge selection based on the object type and constraints.