WO1992005479A1

WO1992005479A1 - Inference control system in artificial intelligence and robot control system

Info

Publication number: WO1992005479A1
Application number: PCT/JP1991/000051
Authority: WO
Inventors: Akira Tagagi; Yasuaki Abe; Kaname Nakazima; Junji Uenaka
Original assignee: Kabushiki Kaisha Csk
Priority date: 1990-09-14
Filing date: 1991-01-18
Publication date: 1992-04-02
Also published as: JPH04123231A

Abstract

When solving a problem, referring a knowledge base (4) for problem-solving and a meta knowledge base (5), a solution tree which analytically expresses a solution method of the problem is generated. An operation program which describes the procedure for executing the problem-solving by arranging execution functions positioned at ends of the solution tree in the execution order is produced and an inference engine (1) for executing the problem-solving in accordance with the procedure described in the operation program is provided. Thereby, when executing the problem-solving, an instruction can be executed by interruption between the executions of two execution functions constituting the operation program; and the capability of the problem-solving which reflects the result of recognition of the outside situation and the capability of interruption, break and resume of the program at any time can be realized to a high degree. The operation program produced is stored and retained in the knowledge base (4). Problems which can be solved with procedures similar to the one retained in the operation program are solved by the inference engine (1) according to the operation program. That is, problem-solving can be executed with a simpler processing quickly without repeating the same inference which is made to solve a problem previously.

Description

Inference control system and mouth-box control system in artificial intelligence

[Technical field ]

The present invention relates to an inference control method and a robot control method in artificial intelligence, and in particular, to an inference control method and a problem control method that solves a problem while recognizing the state of the outside world and its own state. It relates to a robot control method based on this.

[Background Art]

At present, research and development of human intelligence are actively conducted, and they can be applied to the operation control, diagnostics, and design support of mouth-boxes and other machines. It is considered.

On the other hand, in fields where artificial intelligence is expected to be applied, there are many cases where the ability to deal with dynamic environmental changes autonomously at any time is required. In such a case, inference control by artificial intelligence is expected to have the following capabilities in particular:

(1) Problem-solving ability that reflects the recognition result of the external situation

Instead of deciding the entire sequence of actions by a program or the like before, instead of taking into account the surrounding situation after a rough planning, Ability to determine actions according to the situation at that time.

Interruption / interruption / resumption of Lolo instruction Permits an instruction that has a higher priority than the instruction given earlier to be accepted later, interrupts the previous instruction, or terminates the interrupt instruction. Ability to resume execution of the previous instruction.

(3) I / O ability in natural language

The ability to input and respond to command statements in natural language.

At this point, in order to realize the above three capabilities, various types of problem solving by inference without inputting a program for problem solving in advance. Although inference control methods have been researched and developed, the conventional inference control method uniformly solves each problem by using the problem-solving knowledge base and the meta-knowledge base for each problem. The problem was solved while reading out the knowledge and processing methods and making inferences.

However, the above-described inference control method has the same problem even when the problem solving performed by the same processing procedure as in the basic operation is repeated many times. There is a lot of wasteful processing because the problem is solved while reading and inferring the solution knowledge and the processing method, and in such a case, the same problem is obtained by a simpler processing. An inference control method for solving the problem was desired.

The inference control method in the artificial intelligence according to the present invention has been made in view of the above-mentioned conventional problems, and is simpler when resolving a problem experienced in the past. The purpose is to provide a means to solve the above problem in the processing.

[Disclosure of Invention] The invention of the first item that achieves the above objectives is based on the inference engine as a control means, the self-state recognition unit that monitors the state of the self, and the state of the outside world. It shows the external environment recognition unit to be monitored, the problem solving that stores the knowledge rules for problem solving H, the knowledge base, the problem solving execution method, and the data processing method. It has a meta-knowledge base in which information is stored, and solves problems in accordance with the information in the above-mentioned knowledge base for problem solving and in the knowledge base. In the inference control method in artificial intelligence that recognizes the external situation based on information from the environment recognition unit, the inference engine solves the above problem and solves the above problem. Refer to the knowledge base for solution and the knowledge base for solving the problem. Generate an analytical solution tree and generate an action program to solve the problem by arranging the execution functions located at the end of the above solution tree in the execution order. The feature is to solve the problem according to the processing procedure shown in the operation program.

Thus, when executing a problem solving, an instruction can be interrupted between each of the execution functions constituting the operation program, and the recognition result of the external situation is reflected. The ability to solve problems and interrupt commands at any time. ♦ The ability to interrupt and resume operations to a high degree is effective.

In addition, the present invention stores the generated operation program in a knowledge base and saves the program, and solves the problem by a processing procedure similar to the processing procedure shown in the operation program. The problem that can make a decision On the other hand, when the problem is solved in accordance with the saved operation program and the problem solved in the past is solved again, the same inference is repeated. In addition, the above problem can be solved quickly with simpler processing.

The invention set forth in the second aspect is the invention according to the first aspect, wherein the self-state recognition unit, the external environment recognition unit, the knowledge base for problem solving, the meta knowledge base, and the input of a problem or an instruction. The means and the inference engine are equally positioned with respect to the inference engine, and the inference engine executes the self-state each time one of the execution functions constituting the operation program is executed. It is characterized by retrieving the recognition section, external environment recognition section, knowledge base for problem solving, and meta knowledge base cyclically, and allowing input of new information and problems.

This makes it easy to interrupt other instructions while the operation program is being executed, and to solve problems that reflect the recognition result of the external situation, and to interrupt, interrupt, and resume instructions as needed. Achieving the ability at a higher level will have a significant effect.

The invention of paragraph 3 is the same as the invention of paragraph 1 or 2 except that natural language input means is provided as a means for inputting operation data. It is characterized by inputting an operation command.

As a result, if the input / output capability in the natural language is dramatically improved, there will be a radiation effect.

The invention of paragraph 4 is based on the knowledge base for solving a problem as a control means for controlling the operation of a robot. The following describes the processing procedure for generating a solution tree by referring to the meta knowledge base and arranging the execution functions located at the end of the solution tree in the execution order to solve the above problem. An inference engine that generates an operation program and solves the problem according to the processing procedure shown in the operation program above is provided. An instruction can be interrupted between each execution function that constitutes the operation program that controls the operation of the program, and the problem that reflects the result of the external situation is solved. The ability to interrupt, interrupt, and resume functions and instructions at any time is highly effective.

In the present invention, the generated operation program is stored in a knowledge base and stored, and the problem is solved by the same processing procedure as the processing program shown in the operation program. In order to solve this problem, controlling the robot operation according to the stored operation program described above is similar to the operation experienced in the past. When the operation is performed again, the same inference is not repeated. <The operation control described above can be performed quickly with simpler processing.

[Brief description of drawings]

FIG. 1 is a block diagram showing a system configuration of a mouth socket control system according to an embodiment of the present invention.

FIG. 2 is a flowchart showing the procedure for solving a problem according to the present embodiment, and FIG. 3 is a diagram showing the contents of a knowledge base for problem solving used in the present embodiment. FIG. 4 is a diagram showing the contents of meta-knowledge used in this embodiment. FIG. 5 is a diagram showing the contents of a self-state recognition unit used in this embodiment.

FIG. 6 is a diagram showing an example of tree-structured data created when solving a problem according to the present embodiment.

Fig. 7 shows the solution tree generated from the data in Fig. 6.

FIG. 8 is a flowchart showing an operation program corresponding to the operation program generated from the solution tree of FIG. 7,

FIG. 9 to FIG. 11 are diagrams showing the specifications of the operation program in this embodiment.

[Best mode for carrying out the invention]

Hereinafter, embodiments of the present invention will be described with reference to the drawings.

In this embodiment, an example in which the inference control method is applied to the operation control of an autonomous mobile robot will be described.

FIG. 1 is a block diagram illustrating a system configuration of an inference control method according to an embodiment of the present invention.

As shown in the figure, the inference control method of this embodiment includes an inference engine 1 as a control means, a self-state recognition unit 2 as an input unit, and an external environment recognition unit 3 as an input unit. It has a natural language sentence analysis section 6, a problem solving knowledge base 4 as a problem solving section, and a meta knowledge base 5 as well as an output section 7. .

The natural language sentence analysis unit 6 converts the given natural language sentence into a sentence. On the other hand, a morphological analysis and a syntactic analysis were performed to extract a receiving structure in which a concept symbol was used as a node, and the result was converted to a frame structure. Is sent to the inference engine 1 as a natural language sentence interpretation.

In the natural language analysis unit 6, the morphological element analysis is realized by using a conventional method. In addition, as a means for parsing, conceptual symbols are analyzed by using an analysis means based on context free grammar, an augmented transition net, and the like. Anything can be used as long as it extracts a dependency structure that makes the dependency between the concepts as nodes.

Here, the concept symbol is a symbol that represents the meaning of a word without depending on a specific language. For example, both the Japanese "book" and the English "hook" are words that represent the same concept. For example, this is represented by a conceptual symbol, such as C BOOK. Please. This means that the entire system can be translated into Japanese and English without changing the knowledge base by simply changing the morphological and parsing sections. It is possible to correspond to various languages such as French and French.

The inference engine 1 extracts necessary information from the input data, sets a problem solving goal, and performs necessary processing.

The above-mentioned problem-solving knowledge base 4 is based on the condition-result type, objective one-step type, time-based knowledge for solving individual problems. It is classified and stored according to rules such as ordinal types (hereinafter referred to as knowledge rules). Figure 3 shows the classification and actual examples of this knowledge rule.

The meta-knowledge base 5 stores meta-knowledge so that the control structure for solving the problem depends on the rule type of the problem. Specifically, it is the knowledge of how to interpret and execute knowledge for solving each problem, and how to process external data, self-data, and Japanese sentence interpretation data. Since the knowledge is placed outside the control structure, the control structure does not depend on the individual problem, and even if it is in the middle of processing a certain problem, it can be recognized by the outside world recognition. Interrupts such as information and external instructions are allowed. Even if a completely new type of problem-solving is required due to the realization of meta-knowledge, even if the knowledge becomes necessary, a knowledge rule that shows how to process that knowledge is used as meta-knowledge. By registering as a new knowledge base and applying a new knowledge base for problem solving, it is possible to expand the system without changing the control structure. Become .

Figure 4 shows the relationship between the rule type of the problem and the meta knowledge.

The self-state recognition unit 2 has a slot indicating the internal state of the robot in the form of a frame, and adapts to the slot of the self-state power s. Monitor for Figure 5 shows an example of the slot and its contents.

The external environment recognition unit 3 has a plurality of coordinate systems, A label is given to each coordinate of the target system, and the label is stored as a value specific to an object on that coordinate. When the robot searches the surrounding area and finds the rappelling power of the object to be detected, it simply reads the coordinates and recognizes the environment.

The output unit 7 outputs the progress and the result of the problem solving by the inference engine 1 by a graphic display and an inference process display.

Next, a problem solving procedure using the inference control method of the present embodiment will be described with reference to FIG. The problem solving according to the present embodiment is performed based on a cyclic processing process as shown in the figure.

First, when an instruction is input by a natural language sentence, the natural language sentence analysis unit 6 analyzes the instruction and interprets the instruction as a natural language sentence interpretation. Send to.

Hereinafter, the processing contents of the inference engine 1 will be described for each of processing 1 to processing 5 shown in the figure.

In process 1,

"If natural language sentence interpretation data is newly generated, the goal of action (goal) is generated from the interpretation data, and the result is used as a goal. It is stored in a storage area called a file recognition slot, and the area where the natural language sentence interpretation data is placed is cleared. "

And processing of the contents.

In process 2, "If there is a goal in the goal recognition slot, a knowledge rule with the goal as the goal is used as a knowledge base for problem solving. After searching, the contents and type of the knowledge rule are stored in the memory area called the rule slot, and the goal recognition slot is further cleared. "

The processing of the content is performed. In process 2, furthermore,

"If the knowledge slot is stored in the rule slot and stored in a rule slot, the rules for the rule interpretation can be stored in a meta knowledge base. Search and store it in a storage area called the meta-knowledge slot. "

The processing of the content is performed. ·

In process 3,

"If the meta-knowledge slot contains meta-knowledge for rule interpretation, the rules in the rule slot are followed according to the meta-knowledge. And store the result in a storage area called the solution tree subslot.

The processing of the content is performed. In one run of this process, it is not possible to generate a tree of solution cases in many cases. A knowledge rule represents a sub-goal to achieve a certain goal, and the sub-goal represents an action that can be immediately performed. In some cases, it is an execution function, and in other cases, the problem needs to be solved using another knowledge rule. For this reason, in the latter case, a different knowledge rule must be used, and the meta knowledge must be in accordance with the rule type. That is what happens. Therefore, the above processing is repeated cyclically until all the ends of the sub-targets become execution functions, and the respective results are combined. Thus, a single solution tree that analytically shows the solution of the problem is generated. This means that if the solution tree subslot has a solution tree (1), the solution tree (2) is also stored in a storage area called the solution tree slot. If so, the solution tree (1) inserted at the appropriate location in the solution tree (2) is taken as the new solution tree, otherwise the solution tree (1) is inserted. ) As a new solution tree, the new solution tree must be stored in the solution tree slot, and the solution tree subslot must be cleared. "

This is achieved by processing the contents. In this case, the above processing is not repeated continuously, but is performed only once per cycle of the inference engine 1. As a result, even during the generation of the solution tree, the self-state recognition based on the input from the self-state recognition unit 2 and the external environment recognition unit 3 and the external environment recognition become possible, and the solution It is possible to cope with knowledge rules that require the result of recognition when generating a tree, and it is also possible to accept another statement in the meantime. .

In Process 3, furthermore,

"If the solution tree slot has a solution strength s', there is no power, and there is no unprocessed subgoal in the solution tree, that is, the solution tree If all the terminals are expanded to the execution function, create an action program and execute the action program slot. It must be stored in a storage area called, and the solution tree slot needs to be cleared. J

The processing of the content is performed. When creating the action program, proceed in steps of one cycle of the inference engine 1 in the same way as when creating the solution tree. You can do it. However, in order to create an operation program, self-state recognition and external environment recognition are not particularly necessary, and even if interrupts are permitted during this period, it is still a merit. In this embodiment, the operation program creation is realized by preparing a function to be performed at once, calling it, and processing it.

In process 4,

"If the operation program slot has an operation program power, if the execution has not been completed, one step at a time If it is running and completed, you need to clear the operating program slot. "

The processing of the content is performed. The execution of the operation program slot is performed by taking out the operation contents sequentially from the beginning of the operation program and executing the operation program, and advances the point of regard. However, in the case of a jump command, shift the point of regard to the jump destination, and skip the label in the case of a label.

In process 5,

"If there is no value in each recognition slot and there is recognition information that has been stacked (evacuated), the latest evacuation information is used. Restores the original storage area from the original There is a need . "

The processing of the content is performed. Also, if you need to report on the achievement of your goals, do so here.

The following is a specific example of the problem solving process when a statement in the autonomous transfer port with the inference control method of this embodiment is input in Japanese. You

For example, suppose that the following statement is input.

"Go ahead of the vending machine."

Upon receiving this sentence, the natural language sentence analysis unit 6 performs a morphological analysis on this sentence. The result is as follows.

Go to Z before vending machine Z

When the morphological analysis is completed, the parsing is performed in consideration of the relation between the morphemes, and the following dependency structure is created.

C GO: Verb (semantic information: action)

(p dest): Case-rib (place end case)

C P FRO NT: Noun (semantic information: location)

(KITAI): Adjunct particle (substrate attribute)

C BEND M: Nouns (information: substance), the structure of the dependency is represented by a conceptual symbol. In the above Japanese sentence, "C BEND" is replaced by "C BEND M". "Vending machine", "CP FRONT" corresponds to "before", and "CGO" corresponds to "go".

Next, this dependency structure is read, and its meaning is described in a frame structure. Describe in frame structure As a result, the result described in the frame structure where the form of the analysis result is less affected by the style of the natural language sentence is as follows:

(doushi (C GO semantic information: act))

meirei-flag t

basyo-syuuten C PFR0NT 0

kakar eru nil) (1) (m e i sh i (C PF 0NT (semantic information location

kitai C BE D M 1

kakar er u nil) 2) (m e i s h i (C ΒΕΝϋ Μ (semantic information entity

kakar erunil) (3) The statement converted to the frame structure in this way is passed to the next inference engine 1.

The inference engine 1 first extracts the necessary information from the frame structure resulting from the interpretation of the instruction and the query in the processing 1, and outputs the necessary information to the robot. Do it in the shape of a powerful goal. From the above frame structure,

(C GO: p dest C PFR0NT: kitai C BE D M)

A goal force s of the form is generated.

When the goal is generated, the inference engine 1 searches the knowledge rule for achieving the goal from the knowledge base for problem solving 4 while circulating through the processes 1 to 5. , Sub-goal based on Meta-Knowledge Base 5 By doing so, a solution tree is finally generated, and an action program is generated from the tree.

First, a solution tree is generated. For this purpose, from the knowledge base for problem solving, which stores the purpose-specific one-way knowledge rules, there is a knowledge rule that has a purpose part that matches the goal. Search for rules (assuming that there is exactly one knowledge rule). In this example, the following knowledge rule is obtained.

Rule 1:

(Rule-1 T-10D

(C GO: dest C PFRONT

: Kitai C BEND M)

((C KNOWN: d ob j C LOCATE

: Kitai C BEND M))

((C MOVE: dest C PFRONT

: Kitai C BEND M)))

The knowledge rule is represented by a conceptual symbol, "C KNOWN" corresponds to "defined" in Japanese, and "C MOVE" corresponds to "move" in Japanese. .

This rule 1 says, "To go to the vending machine," in order to achieve the regaul, "Know the location of the vending machine. Under the conditions where the sub-goal has been achieved, it is only necessary to achieve the sub-goal of "move to the vending machine". ”. It is a knowledge rule for "Conditional execution operation instruction type" (see Fig. 4). Since this knowledge rule includes sub-goals, another knowledge rule is needed to achieve that sub-goal.

Rule 2:

(Rule-2 Τ-ϋ 1 K

(C KNOWN: d ob j C LOCATE

: Kitai C BEND M)

(OBJECT-PO ITION C BEND M)

((C LOOK: d obj C LOCATE

: Kitai C BEND M)))

Rule 2 is based on the execution function OBJECT-POSITION (the robot itself) in order to achieve the goal of "" A function that obtains the position information of the object from the memory of the memory), and if the result is ni1, the sub-goal of "visualize the vending machine (in front of)" It is good to achieve J. It is a knowledge rule of << Unconditional execution confirmation instruction type >>.

Rule 3:

{Rule-3 T-01 D

(C LOOK: d obj C LOCATE

-.kitai C BEND M)

((SEARCH POSITION C BEND M)))

Rule 3 requires the execution function SEARCH to achieve the goal of "seeing (in front of) vending machines". POSITION (a function to obtain the position information of an object from the external force using the robot's visual unit) may be activated. ”. It is a knowledge rule of << Unconditional execution operation instruction type >>.

Rule 4:

(Rule-4 T-01 L

(C MOVE: dest C PFR0NT

: Ki tai C BEND)

(SEARCH OBJECT C BEND: len 0)

((CM 0 V E 1: p org (get 'robot' position)

: P dest C BEND M

: N e ar t

: Speed 'normal)))

Rule 4 requires the execution function SEARCH in order to achieve the sub-go, "Move before vending machines".

OBJECT (function that determines whether an object is within a fixed distance range using a robot's visual unit) is something other than ni1 Until then, the execution function CM0VE1 (the function that activates the robot's walking unit and performs the movement in the specified direction) may be repeated and started. ". It is a knowledge rule of << Unconditional execution loop instruction type >>.

When the goal is developed according to these knowledge rules, the hierarchical relationship between the goal, sub-goals, and execution functions is obtained as shown in Fig. 6. It is possible to obtain tree structure data. Here, Fig. 6 (a) shows the state in which the rule is applied, Fig. 6 (b) shows the state in which rule 1 is applied, and Fig. 6 (c) shows the state in which rule 2 is applied. Fig. 6 (d) shows the state where rule 3 is applied, and Fig. 6 (e) shows the state where rule 4 is applied.

However, in practice, it is necessary to add control information such as branching and repetition. The type of control information that must be added depends on the type of relay. The tree-structured data in which the jump instruction of the control information is marked at the position where the jump instruction is to be processed is the solution tree.

FIG. 7 shows the solution tree in this embodiment. In the figure, "rn" stands for "RETURN-TO-NEXT" (jump instruction),

"rb" indicates "RETURN-TO-BEFORE" (jump instruction). Once the solution tree has been determined, an action program is generated based on the tree. Near the executable function at the end of the solution tree, a jump instruction is placed at the position where the jump instruction can be applied. Add a bell.

Here, the specification of the operation program in this embodiment will be described.

The operation program generated by the system when solving a problem is an indefinite list of control information and execution functions.

Here, the control information includes a jump instruction in the operation program and a label indicating a jump destination. It is. A list of control information in this embodiment is shown in Fig. 9 (note that num, num-A, and num-B shown in the figure are positive integers).

In addition, an execution function represents a primitive action for a robot, or a combination thereof. FIG. 10 shows a list of the execution functions set in the present embodiment.

There are some functions that fall into the category of execution functions, but have slightly different handling methods. This is called a meta function, and in this embodiment, three types shown in FIG. 11 are set.

Based on the above specifications, the operation program generated for the problem solving rule shown in Fig. 7 is shown below.

(OBJECT-POSITION C BEND M)

(RETURN-TO-NEXT 001)

(SEARCH POSITION C BEND M)

(001)

(SEARCH OBJECT C BEND M: 1 en 0)

(RETURN- TO-NEXT 001)

(CMOVE 1: p org (GET 'ROBOT' POS ITION)

: P dest C BEND M

: N e ar t

: Speed 'NORMAL) (RETURN-TO-BEFORE 001)

(001))

The flow chart corresponding to this operation program is as shown in FIG. The function 1 in Fig. 8 is (OBJECT-POSITION C BEND M), the function 2 is (SEARCH POS ITION C BEND M), and the function 3 is (SEARCH OBJECT C BEND M: 1 en 0 ) And function 4 corresponds to (CM0VE1: porg (GET 'ROBOT' POSITION): dest C BEND M: near t: speed 'NORMAL).

When the operation program is generated, the operation program is executed in process 4. The execution of the operation program is performed for each execution function. In other words, each time one execution function is executed, each process shown in Fig. 2 is cycled, and each time the problem is to be solved prior to the problem currently being executed (for example, (Such as sub-goals for the goal in question or processing to respond to changes in the external environment). Then, if there is an input, similarly, an operation program for solving the problem is generated and executed. When there is no input, the next execution function of the original operation program is executed.

By the above processing, the robot completes the execution of the instruction "go before the vending machine". In this embodiment, the generated operation program is stored in a knowledge base and is stored as new knowledge. The problem is solved by the same procedure as "Go to the vending machine".

Factory

0 ^ shaking π 葸 ¾ ¾ ¾ i π ¾ d

^ 熥 0 ί ¾ ¾:

籁 31 ¾ ί

^ ϋ 0 ¾ σ 1 ^ or ί-

¾ ^^^ 鏽 $ M S ΠI ¾ i τ¾ i 7 α *

Plate> s ¾ ¾ ¾ ϋ¾ s

Claims

) An inference engine as a control means, a self-state recognition unit to monitor the state of our own, an external environment recognition unit to monitor the state of the outside world, and a knowledge rule for problem solving. The stored problem

It has a knowledge base for solution and a knowledge base for storing information indicating a method of executing a problem solution and a data processing method, and the knowledge base for problem solving and the meta knowledge base described above. Information car

The self-state recognition unit and the external environment

In the inference control method in the artificial intelligence that recognizes the external world situation based on the information from the environment recognition unit,

The inference engine generates a solution tree that analytically shows the solution of the problem with reference to the problem-solving knowledge base and the meta-knowledge base in order to solve the problem.

An execution program for solving the above problem is generated by arranging the execution functions located at the end of the tree of the above solution in the order of execution, and

In addition to solving the problem according to the processing procedure shown in the above operation program,

The generated operation program is stored and saved, and for a problem that can be solved by a processing procedure similar to the processing procedure shown in the operation program, An inference control method in artificial intelligence characterized by solving a problem in accordance with the saved operation program.

(2) Self-state recognition unit, external environment recognition unit, problem solving-knowledge base, meta-knowledge base, and problem or instruction input means are equivalent to inference engine. Located in

The inference engine executes the self-state recognition unit, the external environment recognition unit, the knowledge base for problem solving, and the program each time one of the execution functions constituting the operation program is executed. The inference in the artificial intelligence described in claim 1 characterized in that the knowledge base is searched cyclically and new information 2 and the input of a problem are allowed. control method.

(3) A natural language input means is provided as a means for inputting the operation data, and the method is characterized in that an operation command in a natural language is input. Inference control method in artificial intelligence described in paragraph 1 or 2.

(4) An inference engine as a control means for controlling the operation of the robot, a self-state recognition unit for monitoring the self-state, and an external environment for monitoring the state of the outside world A recognition unit, a knowledge base for problem solving that stores knowledge rules for problem solving, and a meta-knowledge base that stores information indicating how to execute the problem solving and how to process data. With

The problem is solved in accordance with the information in the above knowledge base for problem solving and the knowledge base, and the external situation is determined in accordance with the information from the self state recognition unit and the external environment recognition unit. In the robot control system that recognizes

The inference engine described above is required to solve the above problem. A solution tree that analytically shows the solution of the problem is generated with reference to the knowledge base for problem solving and the knowledge base, and

The generated operation program is stored and saved, and the problem can be solved by the same processing procedure as the processing program shown in the operation program. A robot control method characterized by solving problems in accordance with saved operation programs.