JP2002163631A - Dummy creature system, action forming method for dummy creature for the same system and computer readable storage medium describing program for making the same system action - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a dummy creature system, an action forming method for dummy creature and a medium to be provided by a software application, with which a dummy creature itself newly generates a motion pattern by itself. SOLUTION: A robot 1 (dummy creature system) is provided with a reliability learning part 25, out-of-target learning part 26, other action learning part 27, condition control action learning part 28 and chain action learning part 29. The robot 1 is composed of a condition input part 21 for detecting a condition outside the robot 1 and holding a dummy emotion, relational database part 22 for storing plural motion patterns, action retrieving part 23 for extracting the motion pattern corresponding to the external condition and the dummy emotion, and action output part 23 for moving the robot 1 according to this motion pattern. The motion pattern, which is not stored in relational database part 22, is generated and stored by the reliability learning part 25 to the chain action learning part 29 so that motion patterns can be increased and speedily selected.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a simulated creature device, a method for forming simulated creature behavior in a simulated creature device, and a computer-readable storage medium containing a program for forming a simulated creature behavior.

[0002]

2. Description of the Related Art In recent years, various software applications in which a character appears on a display screen of a computer, and various robots having a character as an entity have been released.

[0003] As an example implemented as a software application for a personal computer, "TEO" (called "TEO":
"Registered trademark: Fujitsu""is a famous example, and as an example implemented as an independent robot, a dog type robot with a metallic design" AIBO (registered trademark: Sony) "
Is famous. Also, as an example implemented as a semi-autonomous robot remotely controlled from an external computer, a wheel running robot "R1" having an owl-like face
00 (NEC) "and so on.

[0004] These software applications and robots are systems aiming to make a character given as a model act as if it were alive. Therefore, the system is designed to output a response such as the emotional expression or action of the character in response to input information from the user.

Recently, some of these software applications and robots have the ability to recognize pattern information such as a user's face image and voice waveform, and can respond to such pattern recognition results as input information. Is becoming more sophisticated. Such a system is a so-called simulated creature system (or a virtual creature system) that attempts to give a simulated life to a character (not necessarily a real creature) to be produced.

[0006] The pseudo-biological system recognizes the surrounding environment of a computer or a robot, which is the substance of the system, including the action performed by the user of the system, and changes the inside of the pseudo-creature which changes according to the recognition result of the surrounding environment. Depending on the condition,
It is designed to return a predetermined response. For example,
If the user becomes bitten, the simulated creature will express joy, and if it is tampered with, it will become relentless or angry. Such a pseudo-creature with interactivity at the level of emotions and desires has only dry interactivity to perform practical tasks, unlike word processing software and industrial robots, for example.
Men and women of all ages are encouraged to bring a healing effect to the user, and that the creature creatures interest or interest the user, or that the user is happy to have an elaborate robot with such a character. It is being widely accepted.

[0007] By the way, the following three methods have heretofore been implemented as a method for determining the behavior of the simulated creature, and the conventional simulated creature system operates by one of these methods. I was going. I, Selective execution method II, Dispensing method III, Reinforcement learning method First, I is the method adopted by most simulated biological systems sold as toys, and is stored in advance in the storage unit built in the system. This is an action expression method in which one action corresponding to a signal input to the system is selected from a plurality of action patterns and executed. What kind of signal is inputted and what kind of action should be taken are built in the system in advance, and the correspondence (action pattern) between the input (situation) and the action leads to responsiveness like a living thing It is something that is built in.

[0008] The signal input to the system is a signal from an optical sensor or a pressure sensor provided in a computer or a robot which is a substance of the system, and a user passes in front of the optical sensor provided in the system. A predetermined signal is input, for example, when the user strokes a pressure-sensitive sensor provided in the system. Also, the behavior pattern is, for example, when the user passes in front of the optical sensor,
If the application is a software application, the creature calls the user; if it is a robot, the robot starts walking toward the user, or if the user strokes the pressure sensor, the creature makes a sound and barks. It is a combination of the input that triggers the action and the behavior of the pseudo-creature performed in response to the input.

[0009] However, in the case of I, in order to select a specific action pattern for a certain input signal and to cause the pseudo-creature to execute the selected action pattern, the same action is repeated every time in the same situation. turn into. Therefore, the more the user uses it, the more the user learns the behavior pattern of the simulated creature, sometimes predicts the next action to be performed, the interest in the simulated creature becomes faint and tired, and it becomes unpleasant because it does not feel like a creature And so on. In addition, mimics are programmed only to take certain actions in certain situations, react in an extremely ephemeral manner without any sense of what the consequences of that action will be. There was also a problem that they could only do things.

As an extension of I, there is a method in which a plurality of specific action patterns for a certain input signal are prepared without being limited to one, and one is determined on the spot by a random number. Among the limited options that have been
Since the action is still taken without considering the result, the above-mentioned problem still occurs.

Next, II is a system adopted by a pseudo-creature system that emits words, such as a pseudo-creature that emits words, such as "Farby (registered trademark: Tiger Electronics Limited)". Similar to I, a behavior pattern corresponding to a signal input to the system is selected and executed from among a plurality of stored behavior patterns.

[0012] However, the behavioral pattern stored here is added with temporal information indicating the manifestation time and information on the input amount, and the time elapsed since the system was first activated and the given information are given. This is an action expression method in which the action pattern selected with the amount of input changes as needed.
For example, from the start up to 100 hours, there are three types of action patterns that can be expressed with respect to the signal input to the system, A, B, and C, and any one of them is executed according to the input signal, and 100 There are eight types of action patterns C to J up to 1000 hours, and any one of them is executed, and over 1000 hours there are action patterns C, F, G, J to Z, any of these Is executed. By doing so, the simulated creature changes its behavioral pattern according to the passage of time and the amount of bite, that is, it appears to the user to grow.

[0013] However, II differs from I in that the selectable action pattern changes according to the elapsed time and the input amount, but the result is considered in a limited option prepared in advance. Acting without any change is exactly the same as I. Therefore, II had the same problem as I.

[0014] Next, III is a system having a learning property that is used to give the simulated creatures individuality and further refine their behavior, and is stored in advance in a storage unit built in the system. This is a behavior expression method in which a parameter of the expression frequency is added to each of a plurality of behavior patterns, and a behavior pattern with a higher expression frequency is more often selected in a corresponding situation.

At this time, there are a plurality of action patterns that can be selected for one input signal, and when the user performs an action selected from the plurality of action patterns, what kind of reaction does the user take? The frequency of occurrence of the action changes depending on whether the action is performed.

For example, if the user responds by hitting the simulated creature after the action (when the input to the pressure-sensitive sensor is large in a short time), the simulated creature is reprimanded for the action just performed. And reduce the frequency of occurrence of this behavior. On the other hand, if the user responds by stroking the simulated creature (when the input to the pressure sensor is small for a long time), the simulated creature is judged to have been "praised" for the current action. Increase the frequency of this behavior.

As described above, when the user praise or scold for the action taken by the simulated creature, the user can use the simulated creature as an artificial life in the same natural manner as when interacting with dogs and cats. Can teach good and bad behavior. In this way, learning performed by giving only an evaluation of good or bad to the result is called reinforcement learning. By reinforcement learning to increase or decrease the frequency of occurrence of behavior, pseudo-creatures that initially performed irrespective of behaviors praised or scolded, perform only praised actions as learning progresses,
I will not be scolded. The selectivity of such an action pattern differs for each simulated creature depending on how the user has disciplined, and thus the individuality of the simulated creature can be created.

[0018] However, although the individuality of the simulated creature is formed by reinforcement learning, it is still the same that the action can be selected only from a limited set of options prepared in advance. When it reaches the top, only familiar behaviors with a high frequency of appearance will appear, so the problem that the user gets bored occurs I, I
Same as I.

As described above, when the pseudo-creature acts, one of the limited behavior patterns set (stored) in advance as being suitable for the situation is selected when the pseudo-creature acts. With the passage of time, the user can predict the behavior pattern of the simulated organism and lose interest in the simulated organism, and since the reaction of the simulated organism becomes unity, it is considered to be a category of toys after all. There was a problem that it was no longer used.

Even if it is possible to assign individuality, considering that the selected action is one of the stored action patterns, individuality determined only by a large or small selection range is considered. In other words, the pseudo-creature does not increase the behavior pattern beyond the preset category and does not act in a situation beyond the preset category, and the personality and growth of the In some cases, the user was not satisfied.

[0021]

As described above, in the conventional simulated creature, the action pattern to be executed is selected from pre-stored action patterns, so that as the user uses the simulated creature, the behavior of the simulated creature is reduced. There was a problem that it became predictable and interest became weak.

In view of the above, the present invention has been made in view of the above-mentioned conventional problems, and provides a simulated creature system in which a behavior pattern is autonomously increased with the passage of time so that a user can continuously use the simulated creature. It is an object of the present invention to provide a computer-readable storage medium describing a method for forming a behavior of a simulated creature in a simulated creature device and a program for forming the behavior of a simulated creature.

[0023]

Means for Solving the Problems To achieve the above object, a simulated creature device of the present invention is a simulated creature device for realizing a simulated creature that moves according to a desired motion pattern. An external situation input means for detecting a situation around the simulated creature as an external parameter value; an internal situation holding means for retaining a simulated emotion of the simulated creature as an internal parameter value; and the operation pattern being performed on the simulated creature. Action, the external parameter value and the internal parameter value of the simulated creature before this action, the external parameter value and the internal parameter value of the simulated creature after this action, the transition probability of this action, as a set of related information consisting of: A plurality of associated database means for storing the plurality of associated parameter values corresponding to the external parameter value and the internal parameter value, and An action searching means for selecting a pattern from the association information of the associated database means, and the action output means for movement on the basis of the selected associated information,
Wherein the external parameter value and the retained internal parameter value that are provided inside the pseudo-biological device and that are detected after executing the selected association information, and the selected association information Calculating the similarity between the external parameter value and the internal parameter value, and when the similarity is equal to or greater than a predetermined value, increasing the transition probability of the selected association information to determine the similarity. If the value is less than the value, a reliability learning means for reducing the transition probability of the selected association information is provided.

Further, the behavior forming method of the present invention is a simulated creature apparatus that realizes a simulated creature that moves according to a desired movement pattern, and that has a linkage database means for storing the movement pattern as linkage information including a plurality of parameters. Detecting the situation around the simulated creature as an external parameter value, holding the simulated emotion of the simulated creature as an internal parameter value, and corresponding to the external parameter value and the internal parameter value. An action search step of selecting the action pattern to be performed by a living organism from the link information of the link database means, an action output step of exercising based on the selected link information, and after executing the selected link information The detected external parameter value and the retained internal parameter value,
Calculating a similarity between the external parameter value and the internal parameter value of the selected association information;
If the similarity is equal to or greater than a predetermined value, the transition probability of the selected association information is increased.If the similarity is less than a predetermined value, the transition probability of the selected association information is increased. And a reliability learning step of decreasing the reliability learning step.

Further, the storage medium of the present invention operates a pseudo-creature that moves according to a desired motion pattern, and forms a behavior in a pseudo-creature apparatus having an association database means for storing the operation pattern as association information including a plurality of parameters. Is a computer-readable storage medium storing a program for performing, the situation around the pseudo-creature is detected as an external parameter value, and the pseudo-feature of the pseudo-creature is held as an internal parameter value, Corresponding to the external parameter value and the internal parameter value, the operation pattern to be performed by the pseudo creature is selected from the association information of the association database means, and exercise is performed based on the selected association information. The external parameter value detected after the association information is executed and held The similarity between the internal parameter value and the external parameter value and the internal parameter value of the selected association information is calculated, and when the similarity is equal to or greater than a predetermined value, the similarity of the selected association information is calculated. The transition probability is increased, and when the similarity is less than a predetermined value, the transition probability of the selected association information is reduced.

Further, the computer program of the present invention operates a pseudo-creature that moves according to a desired motion pattern, and operates a pseudo-creature apparatus having an association database means for storing the operation pattern as association information including a plurality of parameters. A computer program, a function of detecting a situation around the simulated creature as an external parameter value, a function of retaining a simulated emotion of the simulated creature as an internal parameter value, and a function corresponding to the external parameter value and the internal parameter value. Then, a function of selecting the operation pattern to be performed by the simulated creature from the association information of the association database means, a function of exercising based on the selected association information, and the selected association information are executed. The external parameter value detected later and before being retained A function of calculating a similarity between an internal parameter value and the external parameter value and the internal parameter value of the selected association information; and, if the similarity is equal to or more than a predetermined value, the selected association information. And if the similarity is less than a predetermined value, a function of reducing the transition probability of the selected association information.

Further, the robot of the present invention is a robot which moves according to a desired operation pattern. The robot has an external situation input means for detecting a situation around the robot as an external parameter, and a simulated robot. An internal situation holding means for holding a target emotion as an internal parameter, and an action for causing the robot to perform the action pattern;
An association database means for storing a plurality of association information as a set of association information including an external parameter and an internal parameter of the robot before the action, an external parameter and an internal parameter of the robot after the action, and a success probability of the action; Corresponding to external parameters and the internal parameters, an action search means for selecting the action pattern to be performed by the robot from the link information of the link database means, and an action output means for exercising based on the selected link information. Wherein the external parameter and the stored internal parameter, which are provided inside the robot and are detected after executing the selected association information, and the external of the selected association information. Parameter and the confidence in calculating the similarity with the internal parameter (Similarity) calculating means, provided inside the robot, wherein when the similarity is equal to or more than a predetermined value, the transition probability of the selected association information is increased so that the similarity is equal to or less than a predetermined value. In the case of, the transition probability of the selected link information is reduced, and the external parameter and the internal parameter before executing the selected link information, the action of the selected link information, The external parameter and the internal parameter after the action and the transition probability of the action are generated as a set of link information. If the generated link information is not stored in the link database means, the link information is generated. Non-target action learning means for storing.

Further, the robot of the present invention is a robot that moves according to a desired operation pattern. The robot has an external situation input means for detecting a situation around the robot as an external parameter, and a simulated robot. Internal situation holding means for holding the emotion as an internal parameter, an action pattern to be performed by the robot, an action to be performed by the robot, an external parameter and an internal parameter of the robot before the action, and an operation of the robot after the action. An association database means for storing a plurality of sets of association information having external parameters and internal parameters, and the operation pattern corresponding to the external parameters and the internal parameters and performed by the robot by the association database means; Action search means for selecting from information; And a behavior output means for exercising based on the obtained association information, wherein the detected object is detected by the external situation input means, and when the detected object performs an action, the detected object is And the internal parameters having as parameters the emotion of the detected object predicted from the state of the detected object before and after the action. And a third party action learning unit that generates as a set of association information, and stores the generated association information unless the generated association information is stored in the association database unit.

Further, the robot of the present invention is a robot that moves according to a desired operation pattern. Inside the robot, external situation input means for detecting the situation around the robot as external parameters, Internal situation holding means for holding the emotion as an internal parameter, an action pattern to be performed by the robot, an action to be performed by the robot, an external parameter and an internal parameter of the robot before the action, and an operation of the robot after the action. An association database means for storing a plurality of sets of association information having external parameters and internal parameters, and the operation pattern corresponding to the external parameters and the internal parameters and performed by the robot by the association database means; Action search means for selecting from information; A behavior output means for exercising based on the obtained association information, wherein the association information having the same behavior is extracted from the association information stored in the association database means, and the extracted association is extracted. Among the information, the relation information in which the pseudo parameter is improved by increasing the internal parameter after the action compared to the internal parameter before the action is compared with the relation information in which the pseudo emotion is reduced and darkened. The external or internal parameter included only in the relevant information is added to the relevant information to be darkened, and the external or internal parameter included only in the relevant information to be darkened is obtained from the relevant information to be darkened. Deleted, and the association information obtained by adding and deleting the external and internal parameters is stored in the association database means. ; And a status adjustment action learning means for storing as associated information otherwise.

The robot according to the present invention is a robot that moves according to a desired operation pattern. The robot has an external situation input means for detecting the situation around the robot as an external parameter, and a simulated robot. Internal situation holding means for holding the emotion as an internal parameter, an action pattern to be performed by the robot, an action to be performed by the robot, an external parameter and an internal parameter of the robot before the action, and an operation of the robot after the action. The association database means for storing a plurality of external parameters and internal parameters as a set of association information as a set of association information, and the operation pattern to be performed by the robot corresponding to the external parameters and the internal parameters. Behavior search selected from the association information of database means A pseudo-biological apparatus having a step and an action output means for exercising based on the selected association information, in the association information stored in the association database means, after the action of the first association information. Extracting the first and second association information in which the external parameters and the internal parameters of the robot and the external parameters and the internal parameters of the robot before the action of the second association information substantially match, and extracting the first association information; The external parameters and internal parameters of the robot before the action, the action of the first link information, the action of the second link information, and the external parameters of the robot after the action of the second link information, And an internal parameter are generated as a set of association information, and if the generated association information is not stored in the association database means, it is stored as the association information. Chain action learning means.

[0031]

Embodiments of the present invention will be described below with reference to the drawings.

The mimetic provided by the mimetic apparatus of the present invention has an anthropomorphic or imitative form as its outer shape. For example, if it is an animal, it may be a hamster, cat, monkey, dog, rabbit, It simulates parrots, bears, whales and the like, and if it is a plant, it simulates sunflower, maple, sweet basil and the like.

[0033] Therefore, the simulated biological device according to the present invention comprises:
As an embodiment thereof, a computer (electronic device) capable of displaying the shape of a simulated creature or a robot that imitates a simulated creature is used.

FIGS. 1 to 4 show a first embodiment in which a simulated biological system according to the present invention is implemented by a robot.

FIG. 1 is a perspective view of a dog-shaped robot.

The outer shape of the robot 1 is substantially a dog, and comprises a plurality of movable parts and a main body. The movable part is head 2, right arm 3,
It consists of a left arm 4, a right foot 5, a left foot 6, and a tail 7. The main body is the body part 8. The torso 8 has a head 2, a right arm
3, the left arm 4, the right foot 5, the left foot 6, and the tail 7 are connected via joints and can move with respect to the body 8. A motor that generates a driving force, a momentary posture (joint angle) of each part driven by the driving force of the motor transmitted to the joint,
Is provided.

The head 2 has a blinking LED 9 and a photographing camera 10 and an infrared distance sensor 11 having different emission colors at a position corresponding to an eye, a microphone 12 at a position corresponding to an ear, and a mouth. A speaker 13 is provided in the portion, and a pressure-sensitive sensor 14 is provided in a portion corresponding to the forehead. The portion corresponding to the mouth and the portion corresponding to the ear can move in a predetermined direction.

The right arm 3, the left arm 4, the right foot 5, the left foot 6, the tail 7, and the body 8 are also provided with pressure sensors 14, respectively.

Next, FIG. 2 is a block diagram of the robot.

A control section 15 is built in the body section 8. The control unit 15 includes a CPU 16 and a memory 17 including a ROM and a RAM. The CPU 16 receives signals from various sensors (an imaging camera 10, an infrared distance sensor 11, a microphone 12, a speaker 13, a pressure sensor 14, and an angle sensor) provided in each movable unit and drives a motor of each joint. It is connected so that the signal for making it output can be output.

Next, the function of the present invention will be described with reference to the block diagram of FIG. 3 for explaining the function of the present invention.

The function of this robot is called “behavioral expression ability”,
Two points of "action formation ability" will be described. -Action expression ability The configuration for making the action expression ability function
1, linked database section 22, action search section 23, action output section 2
4

The status input unit 21 is provided with various sensors (the robot 1
Receiving and processing signals from various movable parts, an imaging camera 10 provided on the main body, an infrared distance sensor 11, a microphone 12, a pressure-sensitive sensor 14, an angle sensor for each joint), and a CP for receiving and processing signals from various sensors.
U16 is realized by a part of the memory 17 that stores and holds signals and processing results thereof. That is, a process of recognizing signals from various sensors is executed by the CPU 16.

Further, the association database section 22 stores the memory 17
Is implemented using a part of

The action search unit 23 is realized by the CPU 16.

The action output unit 24 is realized by various output devices (LED 9, speaker 13, motor of each joint), a CPU 16 for controlling various output devices, and a part of a memory 17 for storing information for the control. Is done. That is, a process of outputting the operation and expression of the robot through various output devices is executed by the CPU 16.

First, the situation input section 21 obtains situation data inside and outside the robot 1. This situation data has M (M is a natural number) parameter values. The M parameters represent an external state parameter (external parameter value) relating to information obtained from the outside of the robot 1 and an internal state of the robot 1. And internal state parameters (internal parameter values) shown in FIG.

The external state parameters are detected by the imaging camera 10, the capacitance type distance sensor 11, the microphone 12, and the pressure sensor 14, and the presence of the detected object, the emotional state or behavior of the detected object, the detected object or The state of the surrounding environment detected by the robot 1 as a result of the action of the robot 1 itself. For example, the detected object is a human, etc., the emotional state of the detected object is human emotions, emotions and so on, the detected object's behavior is a human running behavior, and the result of the behavior is a human's Is a state where it ran into something when it ran, and the state of the environment is how many people exist in which direction around the robot 1.

The internal state parameters include the pseudo-emotions and desires unique to the robot 1 formed inside the situation input unit 21 under the influence of the external state parameters and the elapsed time, the result of the robot 1's judgment on the object to be detected, the robot 1 The action performed by 1 is whether or not the robot 1 is in action. For example, a desire is a desire such as a feeling of hunger that occurs with the passage of time, a pseudo-emotion is a frustrating feeling that is felt when having a feeling of hunger, and an action is that a detected object is a human. It is an action approaching when judged, and whether or not it is acting is whether it is approaching the detected object after detecting the detected object, and if the detected object is a human, This includes the feeling of satisfaction generated by the robot 1 and the memory of the stroked human face when the head 2 is stroked in close proximity (the pressing force detected by the pressure sensor 14 is small and there is contact for a long time). .

The external state parameter and the internal state parameter include symbol data such as an ID number for identifying an event, thing, action, or the like that can be handled by the robot, and a simulated emotion or desire intensity value of the robot. Is represented by numerical data representing

The symbol data is as shown in Table 1 below.

[Table 1] The numerical data is as shown in Table 2 below. The range of the intensity value is, for example, 256 levels from 0 to 255.

[Table 2] When there are K parameters (K ≦ M) describing numerical data, these parameters can be collectively handled as one K-dimensional vector composed of K scalar quantities. Furthermore, each parameter of the numerical data is defined in advance with a weight corresponding to its importance,
Each parameter value is combined as a vector after being multiplied by a predetermined weighting factor, and the length (norm) of the vector is normalized to 1. If the vector generated by this normalization is defined as a situation vector, the characteristics of the situation indicated by the numerical data can be represented by the direction of the situation vector.

The M parameters as the situation data are as follows:
Data composed of one of the K numerical data and one situation vector is referred to as situation information.

The status information is periodically detected and formed at a predetermined period T, and the latest L periods, that is, the period (T × L), are stored in the ring buffer provided in the memory 17. That is, when new status information is detected and formed, the oldest status information in the ring buffer is discarded,
Replaced by the latest one. A list of the status information for the latest L cycles stored in the ring buffer will be referred to as a status information sequence.

In order to form the behavior of the robot described later, the situation input unit 21 is provided with three ring buffers for storing a situation information sequence.

The first is a ring buffer 1 for storing the latest status information sequence updated as the status information is regularly formed by the status input unit 21. The second is a ring buffer 2 for storing situation information until immediately before the action of the robot starts. The third is a ring buffer 3 for storing status information immediately after the completion of the action of the robot.

The ring buffer 1 indicates the current situation inside and outside the robot. The ring buffer 2 is a copy of the entire contents of the ring buffer 1 at the time when the robot starts to behave.
c is read as the contents of the ring buffer 2. The ring buffer 3 is a copy of the status information for L cycles from the time when the robot has completed the action along with the update of the ring buffer 1, and the actual status Jdc after the action described later is the ring buffer 3 Is read as the contents of In other words, the ring buffers 2 and 3 are buffers for storing the situation before and after the action, and the robot should read and use the actual situation before and after the action of the robot from here. Can be.

The association database unit 22 stores the association information R
(Described later) is stored. The association database unit 22 is formed by using a part of the memory 17, and the memory 17 is used for the robot to recognize an event or thing such as a user or a desk based on signals from various sensors. Holds pattern dictionary information, body control information and output sound data necessary for robot actions (movement and expression of facial expressions), and a status information string formed from external status parameters and internal status parameters formed by the status input unit 21 Ring buffer 1 for
3, and other various temporary or permanent information for operating and learning the robot.

The relation information R is defined by a condition Ji from another condition Ji.
When there is an action Ac that changes the situation to d (Ji + Ac →
Jd), a situation information sequence representing Ji, a situation information sequence representing Jd, and one set of action information representing Ac, and R = [Ji, Ac, Jd].
The statuses Ji and Jd included in the association information are a sequence of status information (status information sequence) for L cycles that the status input unit 21 will observe under each status, and such a status information sequence is observed. It is sample data indicating that it is Ji or Jd.

At this time, Ji is the initial situation, Jd is the objective situation,
The action Ac that causes a transition from Ji to Jd will be referred to as a situation transition action. The situation transition action Ac includes ID information (for example, 0002 “walk” in Table 1) for identifying the unit action (action unit) illustrated in Table 1 above, and a unit action of the action identified by the ID information. An action element comprising timing information (for example, 0 seconds later) for executing (action element) and operation parameter information indicating an object of an action to be added as necessary (for example, 0001 “user” in Table 1) It is configured as a list including one or more pieces of piece information. For example, the association information is as follows.

Linked information R0001: Initial situation Ji = "I am lonely without the user"
(Loneliness = 221) Situation transition action Ac = "Search for the user on the spot" (behavior element ID = 0004 + 0 seconds later) Purpose situation Jd = "I am happy to find the user" (Detected object ID = 0001, Joy = 255) Related information R0002: Initial situation Ji = "Scary with stranger" (Detected object ID = 0009, fear = 194) Situation transition action Ac = "Barking at that person" (Action element ID =
0003 + subject ID = 0009 + 0 seconds later) Purpose situation Jd = "I am relieved that there is no stranger" (fear = 25) Related information R0003: Initial situation Ji = "I am lonely without a user"
(Loneliness = 221) Situation transition action Ac = "barking" (action unit ID = 0003 + 0)
Seconds) Purpose Situation Jd = "I'm glad the user came."
(Detected object ID = 0001, joy = 255) Related information R0004: Initial situation Ji = "Hearing voice" (Detected object ID = 00
10 + azimuth = 40 °) Situation transition action Ac = “Search for the main voice while heading there” (action unit ID = 0002 + azimuth = 40 ° + 0 seconds later) (action unit ID = 0004 + 0 seconds later) Purpose situation Jd = "I am happy to find the voice's master (user)" (detected object ID = 0001, joy = 255) In addition, the situation transition action Ac is action segment information that instructs unit actions such as walking, barking, and looking around. Contains one or more,
When the action unit information is plural, for example, "Search for the voice main while walking (the action unit information is walking, search for the main voice 2
)). further,
Depending on the situation Ji, just waiting without doing anything may automatically lead to the situation Jd.
The action information Ac having only one indicates an action of doing nothing as a result.

Next, the action search unit 23 determines the current situation Jic output from the situation input unit 21 and the initial situation Ji of the association information R = [Ji, Ac, Jd] stored in the association database unit 22. Compare.

The comparison is performed by obtaining score information Sc calculated from the current situation Jic and the initial situation Ji. Current situation
The score information Sc (Jic, R) of the association information R with respect to Jic is the situation suitability A (Jic, R), the situation improvement effect E (R), the transition probability S (R), and the ease of execution W (R). Sc (Jic, R) = [A
(Jic, R), E (R), S (R), W (R)]. Score information Sc (Jic, R) = [A (Jic, R), E (R), S
(R), W (R)].

Here, the four parameters A (Jic, R), E (R), S
A method for calculating (R) and W (R) will be described. (1) Situation conformity A (Jic, R) Situation similarity Sj (Ja, Jb) is defined as a scale for evaluating the similarity between two situations Ja and Jb. Situation similarity Sj between situations Ja and Jb
(Ja, Jb) is the similarity Sk (Ja, Jb) of ((MK) × L) symbol data of each of Ja and Jb, and (K × L) number of similarities of Ja and Jb. The product of the similarities Sv (Ja, Jb) of the numerical data, ie, Sj (Ja, Jb) = Sk (Ja, Jb) × Sv (Ja, Jb).

The similarity Sk (Ja, Jb) of the symbol data information is Ja
Is calculated as a value obtained by normalizing ((MK) × L) the number of times that the symbol data of L periods coincide with each other at each time (MK) held by each and Jb. The range of the value of Sk (Ja, Jb) is ｛Sk
(Ja, Jb): 0 ≦ Sk (Ja, Jb) ≦ 1｝. At this time, Sk (J
As (a, Jb) is closer to 1, the symbol data of the two situations Ja and Jb match, and conversely, the closer they are to 0, the more different they are.

The similarity Sv (Ja, Jb) of the numerical data is Ja
And Jb are obtained as a value obtained by averaging the inner product {Ip = (Va · Vb):-1 ≦ Ip ≦ 1} of the situation vectors Va and Vb at each time for L periods. Since the value of the numerical data constituting the situation vector is 0 or more, all the situation vectors are always in the first quadrant of the K-dimensional hyperspace, and two such situation vectors Va and
The angle formed by Vb is from 0 degrees to 90 degrees. Therefore, the range of the inner product of Va and Vb is actually ｛Ip = (Va · Vb): 0 ≦ Ip ≦
1｝, and Sv (Ja, J
The range of the value of b) is also {Sv (Ja, Jb): 0 ≦ Sv (Ja, Jb) ≦ 1}. As Sv (Ja, Jb) becomes closer to 1 (that is, the angle of the formed angle is exactly 0 degree), the numerical data of the two situations Ja and Jb match, and conversely, 0 (ie, the angle of the formed angle of 90 degrees completely matches). The difference is closer to (orthogonal).

The situation similarity Sj (Ja, Jb) is Sk (Ja, Jb) × Sv (J
a, Jb). The range of the value of the situation similarity Sj (Ja, Jb) is {Sj (Ja, Jb): 0 ≦ Sj (Ja, Jb) ≦ 1}.
As (Ja, Jb) is closer to 1, the two situations Ja and Jb match both symbolically and numerically, and conversely, the closer they are to 0, the more different they are.

Here, the situation similarity Sj between the actual situation Jic before the transition and the initial situation Ji of the association information R = [Ji, Ac, Jd].
(Jic, Ji) is defined as situational fitness A (Jic, R). This A (Jic,
If (R) is equal to or greater than the predetermined threshold, the current situation Jic is very similar to the initial situation Ji of the association information R, and by performing the action Ac indicated by R in this situation Jic, the target situation J indicated by R is obtained.
There is a possibility of reaching d. Note that the situation similarity Sj (Jdc, Jd) between the actual situation Jdc after the transition and the target situation Jd of the association information R = [Ji, Ac, Jd] is referred to as a transition achievement degree B (Jdc, R). I will. This is a measure of how much the actual situation Jdc after the transition matches the intended situation Jd, that is, how well the objective transition has been achieved. If this B (Jdc, R) is equal to or greater than the predetermined threshold, the situation after the transition
Jdc is very similar to the goal situation Jd of the association information R, and by performing the action Ac indicated by R, the goal situation J indicated by R is obtained.
That is, d has been reached, that is, the transition has been successful. (2) Situation improvement effect E (R) Situation improvement effect E (R) related to certain related information R = [Ji, Ac, Jd]
Is the robot 1 before and after the situation transition action Ac indicated by R (Ji and Jd).
How the pleasant emotion intensity C changes. The pleasant emotion intensity C is a parameter indicating a pleasant emotion of the robot obtained from the pseudo emotion and the desire in the internal state parameter, and is described for the situation J as C (J).

The pleasant emotion of the robot 1 changes in accordance with the result of good / bad judgment on the object to be detected existing in the outside world and the state of satisfaction. For example, if the robot finds a favorite thing or a favorite partner, cooperates with a favorite partner or opposes a disliked partner, if the desire is actually satisfied after detecting a target that can satisfy the desire, Increase the value of pleasant emotion intensity C. At this time, the upper limit of C is 1. Also, in the opposite situation, if the robot finds something you dislike or a person you dislike,
If you can not cooperate with your favorite partner or benefit from a disliked partner, if you detect a target that can satisfy your desire and then disappear before it is satisfied, the value of pleasant emotion intensity C Lower. At this time, the lower limit of C is -1.

As described above, the pleasant emotion intensity C (J) is
It functions as a scale to evaluate the preference of situation J for a person. Then, the difference 快 E (R) between the intense emotions C (Jd) and C (Ji) in the target situation Jd of the relation information R = [Ji, Ac, Jd] and the initial situation Ji.
= C (Jd) -C (Ji): The greater the value of −2 ≦ E (R) ≦ 2｝, the greater the preference of the situation is improved by the situation transition behavior Ac. (3) Transition Probability S (R) The transition probability S (R) for a piece of related information R = [Ji, Ac, Jd] is
It represents the probability that the association information R = [Ji, Ac, Jd] is correct (is established). This is because, due to an uncertain element in the outside world, it is not known whether or not the situation transition action Ac described in the association information R can reliably reach the target situation Jd even when executed in the initial situation Ji.
When the situation transition action Ac was performed and the target situation Jd was reached, the transition was successful (association information was established), that is, the association information R correctly indicated the transition (Ji + Ac → Jd) that actually occurred. It turns out that.

The transition probability S (R) is calculated for each piece of association information in the range of {S (R): 0 ≦ S (R) ≦ 1}. Described as the transition probability S (R) for the association information R, the quotient S of the number of executions Nt (R) of the association information R and the number Ns (R) of times that the transition actually succeeded
It is calculated as (R) / = Ns (R) / Nt (R). However, related information
In the initial state where R has not yet been executed, Nt (R) = 0,
Since Ns (R) = 0, the quotient Ns (R) / Nt (R) cannot be defined. It is assumed that the transition probability in this case is S (R) = 0. Related information R
Nt (R), Ns (R), and S (R) are combined and transition probability information Si (R) = [Ns
(R), Nt (R), S (R)]. (4) Ease of execution W (R) Ease of execution W (R) related to certain relation information R = [Ji, Ac, Jd]
Is a scale for measuring how low the cost related to the execution of the action Ac specified by the association information R, that is, how easy the action Ac is to execute. The degree of execution W (R) is the reciprocal of the number of action segments (at least one) constituting the action Ac ｛U (Ac):
It is calculated as 0 ≦ U (Ac) ≦ 1｝. Here, act U (Ac)
It is called the labor-saving index of Ac, and it becomes larger as the action Ac is completed with less operations. The range of the value of the ease of execution W (R) is ｛W (R): 0
<W (R) ≦ 1｝, which indicates that the larger the W (R), the more the action Ac is a simple action that can be completed with fewer actions.

After obtaining (1) to (4), one piece of solution relation information R is extracted. Specifically, situational conformity A (R)
Is greater than or equal to a predetermined threshold, and the situation improvement effect E (R) and the transition probability
The product of S (R) and the ease of execution W (R), that is, the degree of expectation for situation improvement ｛Ep (R) = E (R) × S (R) × W (R): −2 ≦ Ep (R) ≦ 2 One in which｝ is equal to or greater than a predetermined threshold value and maximum is selected. In other words, the selected solution relation information is the most likely to be the situation that can be improved with less effort among the applicable actions that the robot knows for the current situation Ji (situation suitability A is equal to or greater than a predetermined threshold). (Situation improvement expectation Ep is maximum when it is equal to or greater than a predetermined threshold).

Next, the action output section 24 executes the contents of the action segment information at a predetermined timing based on the solution relation information selected by the action search section 23.

For example, suppose that the robot was alone in the room and missed (Situation Jic). The above-described score information Sc is calculated, and association information (candidate association information) having an initial situation similar to the situation Jic is extracted from the association database unit 22. It is assumed that there are, for example, the following three pieces of extracted association information.

Related information R1 = [Ji1: I am lonely because the user is not on the side Ac1: Look around and search for the user Jd1: I am happy to find the user] (Ep (R1) = E (R1) × S (R1) × W (R1) = 1.0 × 0.3 × 0.8 = 0.24) Related information R2 = [Ji2: lonely because the user is not on the side Ac2: barking Jd2: I am happy that the user is coming] (Ep (R2) = E (R2) × S (R2) × W (R2) = 1.0 × 0.5 × 1.0 = 0.5) Association information R3 = [Ji3: Lonely without user Ac3: Search for user while walking Jd3: I am happy to find the user] (Ep (R3) = E (R3) × S (R3) × W (R3) = 1.0 × 0.7 × 0.5 = 0.35) At this time, the threshold of the situation improvement expectation Ep is 0.1. If so, the maximum situation improvement expectation Ep =
The link information R2 having 0.5 is selected as the solution link information that most improves the situation Jic.

The robot performs a barking motion by moving the motors of the movable parts while outputting a bark from the speaker 13 in accordance with the selected solution relation information R2.・ Action formation ability The configuration for making the action formation ability work is the reliability learning section.
25, an untargeted action learning unit 26, another person action learning unit 27, a situation adjustment action learning unit 28, and a chain action learning unit 29.

The reliability learning unit 25, the untargeted behavior learning unit 26, the other behavior learning unit 27, the situation adjustment behavior learning unit 28, and the chain behavior learning unit 29 are exclusively a part of the CPU 16 and the memory 17 in the control unit 15. Is realized by: That is, the processing performed by these components is executed by the CPU 16.

The behavior formation ability has the following four abilities. 1. Action search ability 2. Action replication ability 3. Situation adjustment ability 4. Action planning ability Each of the above abilities is for the reliability learning unit 25 for 1 and the behavior of others for the untargeted action learning units 26 and 2. The chain behavior learning unit 29 performs operations corresponding to the situation adjustment behavior learning units 28 and 4 with respect to the learning units 27 and 3, respectively.

First, each capability will be described. (Action search ability) The action search ability is based on the action search unit described above.
23, a reliability learning 25 described later, and an untargeted behavior learning unit 26.

A robot that encounters a certain situation Jic is applicable in that situation (the degree of conformity A is equal to or greater than a predetermined threshold).
If no action (association information) that can expect a situation improvement effect with a small amount of effort (situation improvement expectation Ep is equal to or more than a predetermined threshold) is not stored in the association database unit 22, the action search unit 23 Unable to select disassociation information. Because the relation database section 22 contains the situation Ji
Only actions that are unlikely to be able to improve the situation with less effort even when applicable in c (situation improvement expectation Ep is less than a predetermined threshold) or that are not applicable in situation Jic (situation conformance A is less than a predetermined threshold) This is because they are not stored. Therefore, only in such a case, the action search unit 23 selects one piece of association information from the association database unit 22 using random numbers as (temporary) solution association information.

The selected association information R = [Ji, Ac, Jd] is the status Ji
c is applicable (situation conformity A (Jic, R) is equal to or greater than a predetermined threshold), and the situation Jdc after the execution of the action Ac is sufficiently similar to the goal situation Jd (the transition achievement B (Jdc , R) is equal to or greater than a predetermined threshold), the transition probability S (R) is increased.
Conversely, although the selected association information R is applicable, the situation Jdc after the execution of the action Ac is not similar to the scheduled target situation Jd (the transition achievement degree B (Jdc, R) If it is less than the predetermined threshold), the transition probability S (R) is decreased, and at the same time, new association information (non-target action association information) R1 = [Jic, Ac, which represents the actual transition (Jic + Ac → Jdc) Jdc] is generated, and the transition probability information Si (R1) = [Ns (R1), Nt (R1), S (R1)] is represented as one success in one trial [1,1 , 1], and the transition (Jic + Ac → Jdc) is made into knowledge by storing these R1 and S (R1) in the association database unit 22.

Further, the selected association information R = [Ji, Ac, Jd] is not applicable in the situation Jic (situation conformity A (Jic, R)
Is less than the predetermined threshold), since the initial situation Ji is not originally similar to the actual situation Jic, the transition probability S (R), which is a measure indicating the correctness of the association information R, is not changed and the actual situation is not changed. Generates new association information (non-target action association information) R1 = [Jic, Ac, Jdc] representing the transition (Jic + Ac → Jdc) and its transition probability information
Si (R1) = [Ns (R1), Nt (R1), S (R1)] is set to [1,1,1], which indicates that one attempt was successful in one trial, and these R1 and S (R
By storing 1) in the association database unit 22, only the knowledge of this transition is performed.

As described above, when the degree of expectation of the situation improvement is low or cannot be applied, one of the related information R is selected from the related information whose effect is in danger, and the result is confirmed. Transition probability S (R) of information R
Is changed (by the reliability learning unit 25 to be described later), or new association information (non-target action association information) R1 for making the transition that has actually occurred into knowledge is generated (the non-target action learning unit 26 to be described later). The robot can learn new aspects of the behavior Ac. Then, when trying the applicable link information, as a result, the link information R with the increased transition probability S (R) is preferentially selected in a situation similar to the situation Jic, and the transition probability S ( The association information R with reduced R) becomes increasingly difficult to select in situations similar to situation Jic. That is, related information that is established under a certain situation (transition probability S increases) and unrelated information (transition probability S decreases) are revealed. In addition, as a result of executing the linkage information R appropriately selected as described above, regardless of the applicability / impossibility, new linkage information (non-target action linkage information) R1 that is established under the situation is clarified. That is, the knowledge of what kind of situation is brought about by the action is stored in the association database 22 in the form of the association information, both in the case where the situation can be improved (successful example) and in the case where the situation cannot be improved (failure example). Go.

The action search ability described above is particularly exhibited at the initial stage when the robot starts to operate, and at the time when the transition probabilities S of all applicable association information are still small. The same applies to a case where the frequency of occurrence of the initial situation described in the association information is low and sufficient trials have not been performed even if the use period has become long. Alternatively, the same applies to a case in which much of the association information held by the robot becomes inapplicable or cannot be established due to a drastic change in the environment around the robot such as a change in the place of use or the user. A conventional robot in which a selectable action is fixedly given in a specific situation cannot cope with such a drastic environmental change.

The action search ability works as follows. For example, consider a situation in which a robot scolded by a user tries to escape from this user, but is further chased and scolded. Robot is scolded. "Run away" is action information as applicable information in Jic.
Only the following link information R1 in Ac1 is linked database section 2
Suppose that it was stored in 2.

R1 = [Ji1: scolded Ac1: escaped Jd2: escaped] Naturally, the robot executes this association information R1, but this is suggested by the user being chased by the user. "Escaped" situation Jd2, this linked information R1
Is informed that it is useless here. Then, the transition establishment S (R) of the link information R1 decreases, and after repeating such a process several times, the transition probability S (R) of the link information R1 becomes extremely small, and the situation improvement expectation Ep (R ) Falls below the predetermined threshold value, the association information R1 can no longer be executed. Of course, at this time, the robot does not have any other powerful association information, so it cannot perform anything. Then, the robot will try other related information by the action search ability. Probably after some unnecessary screeching, the robot happens to execute the following association information R2 having "dancing" selected in a completely different initial situation "happy" in the action information Ac2.

R2 = [Ji2: happy Ac2: dancing Jd2: fun] The initial situations J1 and J2 of R1 and R2 are completely different. "Dancing" action
Ac2 is an action that would never be selected in a situation that would normally be “scolded”. However, it is assumed that as a result of the robot "dancing" due to the action search ability, the user has been angered and stopped pursuing. As a result, the robot learns for the first time by actually experiencing that the robot "does not be scolded" by performing the action "dancing" in the initial situation of "being scolded by the user". Then, this experience is added to the association database unit 22 as new association information (non-target action association information) R3 shown below,
Influence the future behavior of the robot.

Untargeted action association information R3 = [Ji3: scolded Ac3: dances Jd3: no longer scolded] Here, it is assumed that the signals detected by the various sensors are stored in the memory 17 in advance. . For example, if the user beats the robot's head and scolds it, the robot is recognized as a state in which a force greater than a predetermined value is detected by the pressure sensor of the robot head for less than a predetermined time, and the user recognizes that the robot Is scolded and scolded is recognized by the robot as a state in which a voice greater than a predetermined value is detected in the robot microphone.

As can be seen from this example, the actions that can be tried by the robot are limited to the actions stored in the association database unit 22 as the association information, and the actions that deviate therefrom cannot be executed. . The action search ability is merely an experience ability to know a new effect of a known action. (Behavior Duplicating Ability) With the above behavior searching ability, the robot can discover a new effect of a known behavior (for example, “dancing”), that is, an unpredictable transition from an actually occurring transition. As a result, the robot cannot increase the content of the action itself, but can increase the knowledge of using the action. However, the robot has not yet discovered any new behavior.

On the other hand, the behavior duplication ability executed by the other behavior learning unit 27 described later is based on the fact that the robot observes unknown behaviors performed by others around itself, and says, “I can do this in this situation. Is obtained, and this is stored in the association database unit 22 as new association information (other person's behavior association information) R, so that there is an ability to imitate the behavior to acquire the user's new behavior.

For example, when the user X is crying, another user Y touches or hugs the back of the crying user X, so that the crying user X can cry and smile. Assume that the robot was observing what was shown.

The robot detects this series of operations with the imaging camera and the microphone, and the user X “crys”.
Situation Ji is "backlashed" by user Y,
It recognizes that the state Ac changes to the state of “crying and showing a smile” by the action Ac of “embracing”. It should be noted that the memory 17 stores expressions such as a crying expression, a crying / not crying expression, and a smiling expression so that the robot can recognize the expression. Also,
Actions such as touching the back (behavior element: contacting the forelimb with the object + sliding the forelimb) and hugging (being the behavior element: holding the object in both forelimbs + acting element: staying for a while) Can be recognized as a chain of action segments stored in the memory 17 in advance.

Therefore, in order to bring the initial situation Ji in which somebody is "crying" to "stop crying" or "show a smile", the robot "touches the back" of the crying person. Or the act of cuddling
Knowing that it can be achieved by Ac, these Ji, Ac, Jd are converted into a new set of related information (other-action related information) R = [Ji, Ac, J
d], and adds the other person's action association information R to the association database unit 22 to increase the action itself.

The actions that the robot can duplicate by imitating are limited to actions that the robot can execute as a combination of action segments stored in the memory 17. It cannot be duplicated. The behavior replication ability is an experience ability to know a new effect of a behavior composed of a known or a new combination by a known behavior segment by observing the behavior of another person. (Situation adjustment ability) While aiming for situation Jd with the same goal situation, two different association information R1 = [Ji1, Ac1, Jd] and R2 = [Ji2, Ac2, Jd] starting from different initial situations Ji1 and Ji2 Due to the difference between the initial situations Ji1 and Ji2 and the difference between the actions Ac1 and Ac2, the transition probability S (R1) of R1 is high and the transition probability S (R2) of R2
Is low. In other words, it is easy to reach Jd from Ji1, but rarely to Jd from Ji2.

The situation adjusting ability executed by the situation adjusting behavior learning unit 28 described below means a situation Ji2 in which such a transition is difficult to succeed, and a situation Ji1 in which the transition is likely to succeed.
New association information (situation adjustment action association information) R3 = [Ji2, Ac3, Ji1] with the target situation of
And the ability to acquire supplementary preparatory actions that make it easier to reach Jd from Ji2.

This is because the link information R2 = [Ji2, Ac
Instead of trying to make a direct transition from Ji2 to Jd (2, Jd) (Ji2 + Ac2 → Jd), first, the situation-regulation action-related information R3 = [Ji
[2, Ac3, Ji1], which leads to a situation that is likely to fail from Ji2 to a situation that is likely to succeed Ji1, and then the related information R1 = [Ji1, Ac1, Jd]
Two-stage transition from Ji1 to final goal Jd (Ji2 + Ac3 → Ji1 + Ac1
→ The robot learns the behavior pattern of Jd).
At this time, the situation adjustment action association information R3 removes the condition (failure factor) specific to the situation Ji2 that is likely to fail in transition, realizes the condition (success factor) specific to the situation Ji1 that is likely to succeed, and uses the association information R1. This is association information indicating a preparatory action for setting a transition. If the preparatory action succeeds, the situation Ji1 that includes the success factor without including the failure factor is realized, so that the possibility of success in transition to the target situation Jd increases.

For example, assume that there are the following two pieces of association information R1 and R2.

Linked information R1 = [Ji1: The robot is alone on the sofa and misses Ac1: Search for a user on the spot Jd: I am happy to find the user] (S (R1) = 0.6) Linked information R2 = [Ji2: The robot is alone behind the sofa and lonely Ac2: Searches for the user on the spot Jd: I am happy to find the user] (S (R2) = 0.01) At this time, the transition probability S ( It is assumed that R1) is equal to or greater than a predetermined value and is greater than the transition probability S (R2) of the association information R2 by a predetermined value or more. This is because it is easier to find the user on the sofa than behind the sofa, and the difference in the transition probability S indicates this. The failure factor is being behind the sofa, and the success factor is being on the sofa. At this time, the robot generates the following new association information (situation adjustment action association information) R3 from the situation Ji2 to the situation Ji1. The transition probability information of this R3, Si (R3) = [Ns (R
3), Nt (R3), S (R3)] is [0,0,
0].

[0098] Situation control action relation information R3 = [Ji2: Robot is alone and lonely behind sofa Ac3: Robot climbs on sofa Ji1: Robot is alone on sofa and lonely] (S (R3) = 0.0 In this example, since the pleasant emotion strength C of the robot does not increase, the situation improvement effect E of this transition is 0. In this way, the situation adjustment ability is independent of the situation improvement effect E,
Preparatory action R3 is added to the action pattern of the robot. Situation improvement expectation Ep (R3) is insufficient (transition probability S (R3) is insufficient)
When the preparation action R3 is generated, R3 is selected based on the action search ability. It is likely that R3, which fits the same initial situation, will be tried, especially since R2 is useless.

As described above, two pieces of association information that have the same target situation but different initial situations are stored in the association database 22. One of the transition probabilities is sufficiently high, and the other is the other. Is sufficiently low, the function of generating new association information (situation adjustment behavior association information) for adjusting the situation is the situation adjustment ability.

The actions that the robot can acquire as preparatory actions in this way are limited to actions that can be executed by the robot as a combination of action segments stored in the memory 17. If the factor is a condition that cannot be controlled by the robot, for example, the presence or behavior of a third party, it is not possible to generate a behavior that calls or eliminates this. Despite the restrictions described above, the situation adjustment ability functions as an experience ability for the robot to know the preparatory actions for preparing the conditions for the success of a certain action. (Action plan ability) The action plan ability executed by the chain action learning unit 29 described later continuously connects the situations from the association database unit 22 (one objective situation is almost the same as the other initial situation). ) Discover a chain of linked information (actions), and if such a chain can improve the situation,
This is the ability to generate new linkage information (chain action linkage information) representing this linkage and add it to the linkage database unit 22.

[0101] The purpose information Jd1 of the relation information R1 = [Ji1, Ac1, Jd1]
Is stored in the association database unit 22 such that the initial situation Ji2 of the association information R2 = [Ji2, Ac2, Jd2] is substantially the same (the situation similarity Sj (Jd1, Ji2) is equal to or greater than a predetermined threshold). For example, R2
The initial situation Ji3 of R3 is almost the same as the objective situation Jd2, R3
If it is found that the target situation Jd3 is linked to Rn so that it is almost the same as the initial situation Jd4 of R4, the initial or middle linkage information Rk (1 ≦ k <n) in the initial situation Jik If the value of the pleasant emotion intensity C (Jdp) in the target situation Jdp of the association information Rp (k <p ≦ n) linked to the end of Rk is greater than the value of the pleasant emotion intensity C (Jik), A transition (Jik + Ack + ...) that continuously executes information Rk, ..., Rp
+ Acp → Jdp) means that the situation can be improved. Therefore, the robot creates new association information (chain action association information) Rx = [Jik, Ack +… + Ac
p, Jdp] is generated and added to the association database unit 22. As a result, the behavior pattern of the robot that can improve the situation is expanded. At this time, the transition probability S (Rx) of the chain action association information Rx
Is given by the product of the transition probabilities of the linked information Rk,..., Rp. It should be noted that all the status-improvable chains composed of a plurality of linkage information are extracted from the chains R1,..., Rn and added to the linkage database unit 22.

For example, R3 acquired by the above-mentioned situation adjustment ability is more easily selected by the action planning ability. In this example, R3 is generated as a preparatory action by comparing R1 and R2 below. However, R3 has no effect in improving the situation, and it is very unlikely to be executed as it is.

Linked information R1 = [Ji1: The robot is alone on the sofa and lonely Ac1: Search for the user on the spot Jd: I am happy to find the user] Linked information R2 = [Ji2: The robot is behind the sofa Ac2: Looking for a user on the spot Jd: I am happy to find a user] Related Information R3 = [Ji2: A robot is alone behind a sofa Ac3: A robot climbs on a sofa Ji1: Robot Is lonely alone on the sofa] The action planning ability is a new link information (chain action link information) R4 = [Ji2, Ac3 + Ac] shown below that combines this R3 and R1
1, Jd]. This chain action association information R4 clearly has a situation improvement effect, and if the transition probability is increased by a number of trials, it is highly likely that it will be selected as disassociation information by the normal operation of the action search unit 23 thereafter. .

Chain action association information R4 = [Ji2: Robot is alone behind the sofa and lonely Ac4: Climb on the sofa (Ac3) and search for a user there (Ac1) Jd: Find a user It is clear from the above that the number of pieces of linked information linked by the action plan ability may be any number, and the situation may be deteriorated on the way.
In short, the initial state Ji of the first linked information Rk to be chained
It is only necessary to be able to extract, from the chain, a portion where the value of the pleasant emotion strength C (Jdp) in the target situation Jdp of the last association information Rp is larger than the value of the pleasant emotion strength C (Jik) at k.

Note that, due to the action planning ability, a plurality of pieces of association information describing transitions in which the respective initial situations are substantially the same and the respective goal situations are also substantially the same are stored in the association database unit 22. It does not matter. In this case, by the normal processing of the action search unit 23, the one with the smaller number of action segments included, that is, the one with the higher degree of execution W is preferentially selected. Of course, if the execution results in a decrease in the transition probability S, a complex action having more action units will appear instead. Therefore, it is important to create a lot of related information even if it seems useless at that time.

[0106] In some cases, chain action association information in which the initial state of the chain and the target state are almost the same (in a loop) is found, but as described above, a chain that does not improve the situation is not generated. However, the association information in which the initial situation and the objective situation are equal is not generated.

The reliability learning unit 25, the non-target behavior learning unit 26, the other behavior learning unit 27, the situation adjusting behavior learning unit 28, and the chain behavior learning unit 29 that operate in accordance with each of the above-mentioned four abilities will be described in detail. explain.

(A) Reliability Learning Unit 25 The reliability learning unit 25 evaluates the result of the action performed by the robot and performs an operation of increasing or decreasing the transition probability of the association information describing the action. That is, the reliability learning unit 25 is one of means for realizing the above-described action search ability.

The reliability learning unit 25 selects the relation information R = [Ji, Ac, Jd] by the robot in the situation Jic, and stores the information in the relation information R when the action Ac is completed and the situation Jdc is reached. Situation similarity S between the completed target situation Jd and the actual situation Jdc after completion
j (Jdc, Jd), that is, the transition achievement B (Jdc, R) is obtained, and if this is equal to or greater than a predetermined threshold, the transition is successful (association information R
Has been established), the transition probability information Si of this association information R
(R) = Increase the number of executions Nt (R) and the number of successes Ns (R) in [Ns (R), Nt (R), S (R)] by 1 respectively, and S (R) = Ns (R) / Nt (R) is stored in the memory 17 as a new transition probability value.

Conversely, if the transition achievement degree B (Jdc, R) is less than the predetermined threshold value, the transition has failed (association information R has not been established), and the transition probability information Si (R ) = [Ns
(R), Nt (R), S (R)] only the number of executions Nt (R) is increased by 1,
S (R) = Ns (R) / Nt (R) as a new transition probability value in memory 17
To memorize.

After updating (learning) such transition probabilities,
When the robot selects the link information applicable in the same situation as the situation Jic described above, there is a relationship with the situation improvement effect E, but the link information R with the transition probability S higher than before is easily selected. Become. Conversely, the association information R whose transition probability S is reduced becomes difficult to be selected, and as a result, does not appear as an action.

Accordingly, the robot selects the association information and performs an action, and determines the transition probability S based on the success or failure of the transition.
Is updated, association information having a high transition probability is selected little by little.

It should be noted that the reliability learning unit 25
The point is not to increase the pleasant emotion intensity C but to evaluate the action based on whether or not the association information is correct. This is different from reinforcement learning performed by conventional robots. Assuming that the behavior of the robot is to increase the pleasant emotion intensity C, the reinforcement learning increases or decreases the frequency of the behavior depending on the magnitude of the pleasant emotion intensity C (referred to as a reward signal) actually experienced after the transition. Reinforcement learning is the action Ac, the initial situation Ji to select it, and the frequency of action S
This is the method of knowing only R = [Ji, Ac], and does not evaluate the objective situation Jd before the action.

Therefore, in reinforcement learning, even if the transition according to the association information does not occur, as long as the transition can be made to the situation where the pleasant emotion intensity C is increased, the frequency of the expression is increased even if it is an unscheduled transition. Will be increased. As a result, in fact, the action that causes a transition to a wrong situation is reinforced solely in spite of the mistake.

On the other hand, in the present invention, the robot is caused to act similarly for the purpose of increasing the pleasant emotion strength C (this mechanism is carried out by the action search unit 23). The transition probability S of the association information is increased / decreased by evaluating whether or not a result is produced. As a result, it is possible to select an action after grasping the correctness of the association information itself, so that detailed control of the situation by the action becomes possible. According to the present invention, correct association information
As long as R = [Ji, Ac, Jd], the robot can create the target situation Jd. In comparison, reinforcement learning is like closing your eyes and throwing a stone.

(B) Non-target behavior learning unit 26 The non-target behavior learning unit 26 evaluates the result of the action performed by the robot, and indicates a correct transition when the association information describing the action is incorrect. An operation of generating new association information (non-target action association information) and adding it to the association database unit 22 is performed. That is, the non-target action learning unit 26 is one of means for realizing the above-described action search ability.

The non-target action learning unit 26 determines that the robot selects the association information R = [Ji, Ac, Jd] in the situation Jic,
When Ac is completed and reaches situation Jdc, situation similarity Sj (Jdc, J2) between target situation Jd stored in association information R and actual situation Jdc after completion, that is, transition achievement B (Jdc , R)
If this is less than a predetermined threshold, the transition has failed (association information R has not been established), and the transition (J
Generates new association information (non-target action association information) R1 = [Jic, Ac, Jdc] representing ic + Ac → Jdc)
Add to 2. In addition, the generated non-target action correlation information R1
If the related information similar to is already stored in the related database unit 22, it is not added. This is the related database section 22
Is not punctured by a number of similar pieces of association information.

Whether or not the linkage information similar to the generated non-target action linkage information R1 is already stored in the linkage database unit 22 is determined by performing the following search.

First, the newly added association information R1 = [Ji
c, Ac, Jdc] and the transition database Ac
It compares the state transition action Acn in each piece of link information Rn = [Jin, Acn, Jdn] stored in 22 and sequentially detects link information Rn in which both are the same. This is to detect only the link information having the same action as the action Ac of the link information R1 to be added.

Next, the situations Jic, Jdc in the relation information R1 = [Jic, Ac, Jdc] and the detected relation information Rn = [Jin, Acn, Jd]
n], the situation similarity Sj (Ji
c, Jin) and the situation similarity Sj (Jdc, Jdn) between the objective situation, respectively, when both similarities obtain a predetermined threshold or more, the action is the same as the association information R1 to be added,
It is determined that the related information already sufficiently similar in both the initial situation and the objective situation already exists, and the addition of the related information R1 to the related database unit 22 is cancelled.

Here, the difference between the present invention and the conventional reinforcement learning will be described.

In the conventional reinforcement learning, the frequency of occurrence of an action is adjusted depending on the success or failure of the action of the robot executing the selected action pattern (increase or decrease of the pleasant emotion intensity C). In other words, the transition that actually occurred as a result of the action was evaluated only with the pleasant emotion intensity C (reward signal), and only the frequency of occurrence was learned.

On the other hand, the non-target action learning unit of the present invention
26 can learn the transition (Jic + Ac → Jdc) that actually occurred as the additional information R1 = [Jic, Ac, Jdc] to be added. That is, the initial situation Jic that causes the action Ac to appear and the destination situation Jdc to be visited as a result can be made into knowledge according to reality.

The present invention and the conventional reinforcement learning are the same in that the frequency of occurrence of a behavior having a high effect of improving the situation is increased and the frequency of occurrence of a low effect of the improvement is reduced, but the actual strength of reinforcement learning is the same. Since we do not learn the initial situation, `` This time (Ji) '' in the knowledge that `` This time (Ji) should do this action (Ac) '' is fixed, and that fixed With this knowledge, only the fact that it is easier to get a reward if you take this action is acquired by learning the expression frequency S. That is, R = [Ji,
Ac] is fixed and will not be learned. Furthermore, since the objective situation Jd is not made into knowledge, it is not possible to learn an action pattern to cope with a subtle discrepancy in the situation.

On the other hand, the present invention is different from reinforcement learning in that an action pattern for coping with a subtle discrepancy in a situation can be learned. This is necessary in order to build a robot that can accumulate and operate knowledge about what situation an action creates. When knowledge is established (transition according to the related information occurs)
There is no problem with continuing to apply this knowledge. On the other hand, when the knowledge is not established (the transition according to the association information does not occur), it is necessary to newly construct a possible knowledge different from this knowledge. The failure may be due to subtle differences in the initial situation, or to uncertainties that could not be observed as situational information. Therefore, in such a case, the non-target ethology main unit 26 additionally builds up knowledge including the situation before and after, so that the present invention can exert a situation adaptation ability beyond reinforcement learning (does not make the situation into knowledge). Become like

As described above, in addition to the reliability learning unit 25 learning the ease of transition to the target posterior situation, it is determined that there is a transition to a situation different from the target posterior situation. The behavior learning unit 26 learns. Unlike reinforcement learning, this mechanism in which applicable and reachable situations can be subdivided and accumulated (as association information) gives the system higher learning adaptability than before.

The non-target behavior learning unit 26 converts the known initial state to which a known action can be applied and a new initial state to which a known action can transition into knowledge. Things. As a result, the system will be able to apply the known behavior with a new effect, that is, obtain a new behavior pattern (in which case this behavior can be performed), but increase the number of known behaviors It is not possible. That is, the non-target action learning unit 26 operates to expand the situation information in the association information. That is,
The learning performed here does not mean acquiring new behavior information.

By the cooperation of the action search section 23, the reliability learning section 25, and the non-target action learning section 26, the robot can exhibit the "action search ability" described above. That is, even when the robot does not store an action having a sufficient situation improvement effect, the action of the action search unit 23 allows the user to try some action. Such trials are accumulated through the operation of the reliability learning unit 25 and the non-target behavior learning unit 26 as knowledge that influences future behavior regardless of the success or failure of the situation improvement.

In particular, when the transition according to the link information is not performed, new link information is constructed. Therefore, even if a situation different from the knowledge is reached, it is converted into knowledge as new link information. The situation adaptability of this robot is improved.

In order for the robot to exert the above-described action search ability, it is necessary to first store some types of action patterns (association information) in the association database unit 22. This memorized action pattern is substantial
The more (the larger the number), the more situations the robot can or will be able to handle.

However, if the situation is not improved by performing any action, this robot simply keeps performing unnecessary actions endlessly. This is the limit of the situational ability of a robot without the ability to enhance the behavior itself. In order for the robot to exhibit more situational abilities, a learning function that increases the situation transition behavior itself is necessary.

(C) Other-action learning unit 27 The other-action learning unit 27 causes the association database unit 22 to store an action performed by another user such as a user as new association information (other-action association information). The robot performs an operation that is one of the behavior patterns of the robot. That is, the other party behavior learning unit 27 is a unit for realizing the behavior duplication ability described above.

More specifically, the situation input unit 21 recognizes the voice, action (attitude), facial expression, etc. of another person using a microphone or an imaging camera provided in the robot, and determines how the situation of the other person is before and after this action. Is estimated. The situation of others includes internal state parameters such as emotions that need to be estimated, in addition to external state parameters that can be immediately detected by observation. The estimation of the internal state parameters is performed based on the input voice, action (attitude), and expression of the other person, and dictionary information for recognizing them is stored in the memory 17 in advance. The contents of this dictionary information are information for detecting facial expressions, gestures and other gestures (attitudes), speech vocabulary, etc. from captured images and captured voices, and identifying internal state parameters that can be read from them. It is. As a result, for example, if it is observed that an image of an angry expression changes from an image of an angry expression due to a certain action, it is estimated that the emotional state of the other person has changed from anger to joy. The estimated result is stored in the status information sequence.

When the other party's action learning unit 27 detects the other party's action Ac in the situation information sequence by the situation input unit 21, it operates in the following two cases.

(1) When the Action Ac is Performed on a Robot Normally, the situation input unit 21 treats the emotional state or behavior of another person as an external state of the robot, and the emotional state or behavior of the robot is It is summarized in the status information as an internal state. Since the action Ac was performed by the other person with respect to the robot, if the relationship between the ego is turned upside down, it should be possible to reconsider the behavior of the robot with respect to the other person.

The external state parameters observed as the surroundings of the robot can be decomposed into two parts: (a) a part relating to an actor, and (b) an environment part excluding the other person. (a) is the expression, behavior, and attitude of the actor, and the emotions and desires of the other person that the robot can estimate from them. (B) is an environment part common to both the robot and the other person, such as surrounding trees and objects. Further, the internal state parameter of the robot is (c) an internal state parameter indicating the emotion or desire of the robot.

At this time, in addition to the above (b) and (c), the external state parameter viewed from the other person is such that a robot actually exists, but this is replaced by the presence of some other person. It is synthesized from symbol data. Further, the internal state parameter of the other person is the above (a).

[0138] Therefore, the other party action learning unit 27 converts the situation information strings Ji and Jd indicating the situation before and after the action Ac into the above (a), (b),
By disassembling and reorganizing into (c), the situation before and after the action Ac is captured from the perspective of others as Ji 'and Jd', and Ji 'is the initial situation and Jd' is the objective situation. , Ac as new contextual information (other-related information) R = [Ji ', A
c, Jd '] is generated and added to the association database unit 22.
It should be noted that if link information similar to the generated other-person action link information R has already been stored in the link database section 22, no information is added. It should be noted that the search for whether or not there is linkage information similar to the other person's behavior linkage information R in the linkage database unit 22 is similar to the search performed by the non-target behavior learning unit 26.

(2) When Action Ac is Performed Between Others The emotion input state and action of the other person X and the other person Y are treated as external states of the robot by the situation input unit 21, and are summarized in the situation information. I have. Assuming that the action Ac is performed by the other person Y on the other person X, the external state parameters of the robot include (a) a part relating to the other person Y as the actor, and (b) the other person as the target person. It can be decomposed into three parts: a part related to X and (c) an environment part excluding parts related to X and Y. (a) shows the expression, behavior, and attitude of the actor Y, and the emotions and desires of the actor Y that can be estimated by the robot from the expressions, actions, and attitudes. (b) is the target
The facial expressions, actions, and attitudes of X, and the emotions and desires of the subject X that the robot can estimate from them. (C) is an environment part common to both others X and Y, for example, surrounding trees and objects.

At this time, in addition to the above (b) and (c), the external state parameter viewed from the actor Y includes the fact that the target person X actually exists. Is synthesized from the symbol data replaced with. Also, actor Y
Is the above-mentioned (a).

Therefore, the other party behavior learning unit 27 divides each of the situation information strings Ji and Jd indicating the situation before and after the action Ac into the above (a), (b),
By disassembling and reorganizing into (c), the situation before and after the action Ac is captured from the perspective of others as Ji 'and Jd', and Ji 'is the initial situation and Jd' is the objective situation. , Ac as new contextual information (other-related information) R = [Ji ', A
c, Jd '] is generated and added to the association database unit 22.
It should be noted that if link information similar to the generated other-person action link information R has already been stored in the link database section 22, no information is added. It should be noted that the search for whether or not there is linkage information similar to the other person's behavior linkage information R in the linkage database unit 22 is similar to the search performed by the non-target behavior learning unit 26.

It should be noted that if link information similar to the generated other-action link information R is already stored in the link database section 22, no addition is made. It should be noted that the search for whether or not there is linkage information similar to the other person's behavior linkage information R in the linkage database unit 22 is similar to the search performed by the non-target behavior learning unit 26.

[0143] The other-action learning unit 27 allows the robot to exhibit "the ability to learn from others", that is, the "action-copying ability".

For example, when the user Y touches the back of the crying user X, the robot X observes the site where the crying user X cryed and showed a smile, so that the robot can use the user Y Such interpersonal behavior can be learned. And
This learned behavior appears at another opportunity, such as when the crying user Z is detected.

At this time, in order for such an interpersonal action to be selected, the action needs to increase the pleasant emotion strength C of the robot. This involves the good and bad feelings the robot has about others.

For example, when the robot has a favorable impression on a person, the plight of the person is detected by the situation input unit 21 and acts so as to impair the pleasant emotion strength C of the robot. Therefore, the robot is required to take action to improve the plight intensity C of this person,
As a result, interpersonal behaviors that are effective in improving the situation of others learned earlier appear.

Conversely, when the robot does not like a person, the plight of the person does not impair or increase the pleasant emotion strength C of the robot. Behaviors that do not develop or that further increase poverty develop.

Positive and negative feelings for others are formed through the previous relationship between the robot and others. For a person who loves the robot, a label (information) of "this person is a good person" is attached to the dictionary information for detecting the person. Conversely, a person who is bullying a robot is labeled "this person is a bad person" in the dictionary information for detecting the person.
Each time a person is detected, it is determined whether the person is good or bad, and the effect of the observed external state on the emotional state of the robot changes according to the result. These processes are performed by the status input unit 21.

As described above, in the present invention, the other party behavior learning unit
27 contributes to the acquisition of interpersonal actions by the robot by imitating the actions of others, and the situation input section 21
Changes the pleasant emotion strength C of the robot based on the judgment of good or bad with respect to the detected other person, and the behavior search unit 23 expresses the interpersonal behavior learned based on the pleasant emotion strength C.

With the above-described configuration, the robot can exhibit "action searching ability" and "action copying ability". Specifically, (I) adaptation of the frequency of occurrence of behavior, (I
It is possible to (I) newly acquire an initial situation, which is an action expression condition, (III) newly acquire an objective situation, which is an effect of the action, and (IIII) newly acquire a situation transition action itself, which is the content of the action. In particular, (I
Items I) to (IIII) are knowledge that cannot be acquired by conventional robots.

However, since the robot can only act by being restricted by the effect of improving the situation of each action with such abilities alone, the robot is finally improved by combining a plurality of actions while making the situation worse. May not be able to demonstrate the planning of reaching the situation. The situation adjustment behavior learning unit 28 and the chain behavior learning unit 29 described below correspond to this point, so that the robot can perform a planned behavior.

(D) Situation adjustment behavior learning unit 28 When the situation adjustment behavior learning unit 28 cannot transition from a certain initial situation to a certain target situation, the other relation information that can transition from the other initial situation to the target situation. In order to enable the application, the transition for creating the initial situation is stored in the association database unit 22 as new association information (situation adjustment behavior association information), and the operation is performed as one of the robot behavior patterns. That is, the situation adjustment behavior learning unit 28 is means for realizing the situation adjustment ability described above.

The situation adjustment behavior learning unit 28 is activated every time the link information is added to the link database unit 22, and the target state similar to the target state Jd1 of the added link information R1 = [Ji1, Ac1, Jd1]. The related information R2 having Jd2 = [Ji2, Ac2, Jd2] (the situation similarity Sj (Jd1, Jd2) is equal to or more than a predetermined threshold) is searched from the association database unit 22. As a result of the search, when the association information R2 is detected, the situation adjustment action learning unit 28 determines the transition probability S (R1)
And S (R2), the transition probability S (R1) of the association information R1 is equal to or more than a predetermined threshold, and the transition probability S (R2) of the association information R2 is lower than S (R1) by a predetermined threshold or more. An operation of generating new association information (situation adjustment behavior association information) R3 from the initial situation Ji2 to the initial situation Ji1 of R1 and storing it in the association database unit 22 is performed. Alternatively, conversely, the transition probability S (R2) of the association information R2 is equal to or greater than a predetermined threshold, and the transition probability S (R1) of the association information R1 is S (R2).
If it is lower than the predetermined threshold value, an operation of generating new association information (situation adjustment behavior association information) R3 from the initial situation Ji1 of R1 to the initial situation Ji2 of R2 and storing it in the association database unit 22 is performed.

It should be noted that the generated situation control action association information R3
If the related information similar to is already stored in the related database unit 22, it is not added. Further, the search for whether or not the linkage information similar to the situation adjustment behavior linkage information R3 exists in the linkage database unit 22 is the same as the search performed by the non-target behavior learning unit 26.

More specifically, a plurality (for example, two) of pieces of related information R1 = [Ji1, Ac1, Jd] and R2 =
[Ji2, Ac2, Jd] (association information with any one added)
Among them, it can be determined from the transition probabilities of the two that the frequency of the one piece of linked information R1 is high and the frequency of the other piece of linked information R2 is low. The reason that one can reach the target situation Jd but the other cannot reach is due to the difference between the initial situation as the starting point and the actions that can be taken at that time. Therefore, the initial conditions Ji1 and Ji2 of the link information R1 that succeeds in transition and the link information R2 that does not succeed are defined as an additional condition Jk that is commonly included in the two pieces of associated information, and an additional condition Js that is specific to the former (success). Separate into the incidental condition Jf peculiar to the latter (failure). This Js is a success factor, and Jf is a failure factor. Ji1 = Jk + Js, J
i2 = Jk + Jf.

The situation adjustment action is an action for changing the situation from the initial situation Ji2 = (Jk + Jf) to the target situation Ji1 = (Jk + Js).

For example, consider the following two pieces of association information having the same purpose situation.

Related information R1 = [Ji1: The robot is alone on the sofa and misses (No. 1
Ac1: Search for the user on the spot Jd: I am glad to find the user] Related information R2 = [Ji2: The robot is alone behind the sofa and misses (No.
Ac2: Find a user on the spot Jd: I am happy to find a user] At this time, the common condition Jk is that "I am lonely and alone"
"I am happy to find the user" is the target situation, the success factor Js is "on the sofa", and the failure factor Jf is "behind the sofa". The processing performed by the situation adjustment behavior learning unit 28 is Jf
To make Js, specifically the action of “climbing on the sofa”
Ac3 is generated, and new association information (situation regulation action association information) including this action Ac3 is R3 = [Ji2 = Jk + Jf, Ac3, Ji1 = Jk
+ Js].

Situation adjustment action association information R3 = [Ji2: Robot is alone and lonely behind sofa Ac3: Climb on sofa Ji1: Robot is alone and lonely on sofa] This new situation adjustment action association According to the information R3, the robot performs an action of "when I am alone behind the sofa, I climb on the sofa". In addition, the fact that the robot climbs something must be stored in the memory 17 as a robot's action element. However, the robot has no action of climbing on the sofa, and of course, for the first time, the behavior pattern of "If you are alone behind the sofa and you are lonely, climb on the sofa" is acquired.

(E) Chained Action Learning Unit 29 The chained action learning unit 29 is capable of continuously performing a plurality of pieces of linked information (actions) to improve the situation when individual linked information (actions) cannot improve the situation. New linkage information that finds a chain of information and describes the chain (chain behavior linkage information)
Is generated and stored in the association database unit 22, and an operation as one of the behavior patterns of the robot is performed. That is, the chain action learning unit 29 is means for realizing the action plan ability described above.

The chained action learning unit 29 includes a non-target action learning unit 26.
Each time new association information is added to the association database unit 22 by the action of the other person's behavior learning unit 27 or the situation adjustment behavior learning unit 28, it is started, and the contents of the association information stored in the association database unit 22 are investigated. A plurality of pieces of related information R1 = [Ji1, Ac1) in which the purpose situation of one piece of related information is similar to the initial situation of the other piece of related information, that is, three or more situations are continuously transitioned in a chain of two or more actions. , Jd1], R2 = [Ji2, Ac2, Jd2] (situation similarity Sj (Jd1, Ji2) is equal to or greater than a predetermined threshold), R3 = [Ji3, Ac3,
Jd3] (Situation similarity Sj (Jd2, Ji3) is equal to or greater than a predetermined threshold) ... Rn =
Search all the chains consisting of [Jin, Acn, Jdn] (where n is a natural number of 4 or more). Here, the link information R1 is the first link information, R2 is the second link information, and R3 is the third link information. In this case, there are two links. A chain is a different chain if at least one element of the link information is different, and is a different chain if the order of the element link information is different even at one place. This search is limited to a search for a chain in which the added association information is any of the above R1 to Rn. When a chain is detected as a result of the search, the chain action learning unit 29 includes, in the extracted chain, the association information Rk.
(1 ≦ k <n) Initial situation Jik's pleasant emotion intensity C (Jik)
Pleasant emotion strength C (Jd of the target situation Jdp of the association information Rp (k <p ≦ n)
All the partial chains (including the above-mentioned chain itself) with improved p) are extracted, and each of them is extracted as new association information (chain action association information) Rx = [Jik, Ack + ... + Acp, Jdp]. Remember at 22.

Note that if link information similar to the generated chain action link information Rx has already been stored in the link database section 22, it is not added. In addition, the search for whether or not linkage information similar to the chained behavior linkage information Rx exists in the linkage database unit 22 is similar to the search performed by the non-target behavior learning unit 26.

Further, it is also possible to create and store new chained action linkage information from the chained action linkage information by the above-described procedure.

Further, by the above-mentioned processing, a certain related information R10
The related information R11 having the initial situation Ji11 similar to the target situation Jd10, the related information R12 having the initial situation Ji12 similar to the objective situation Jd11 of the related information R11, and the objective situation Jd12 of the related information R12 are similar. Linkage information with initial status Ji13
When R13 is present, even if the pleasant emotion intensity C of the target situations Jd11 and Jd12 of R11 and R12 is lower than the initial situation Ji10 of the association information R10, the pleasant emotion strength C of the objective situation Jd13 of the association information R13 is large. For example, the link information R10 to R13 is stored in the link database unit 22 as new chain action link information. In this case, the number of chains is 2 (n = 2), and the number of pieces of association information is 3 (n + 1).

As described above, in the case of generating chained action association information by combining a plurality of pieces of association information, even if the intensity of the pleasant emotion C temporarily decreases when an intermediate action is executed, the intensity of the pleasant emotion is eventually reduced. Assume that C has increased.

With such a configuration, the robot exhibits a situation adjusting ability and an action planning ability. Specifically, not only actions that directly improve the situation, but also preparatory actions that lead to indirect improvement of the situation without directly improving the situation, It is possible to generate and acquire linked information describing a series of actions that lead to situation improvement.

As described above, the situation adjustment behavior learning unit 28 and the chain behavior learning unit 29 are linked to the association information database unit 22.
Is monitored at the time of system startup, and is started each time new link information is added. In addition, it is started again immediately after the link information is completely added by itself, and is started recursively until the link information to be added cannot be found.

Having described the configuration as described above and the capability of each component to execute, the action forming method will now be described with reference to the flowchart of the action forming method in FIG. It is assumed that, at the time of operation of the robot, in addition to the individual linkage information, the situation adjustment behavior linkage information and the chain behavior linkage information have already been stored in the linkage database unit 22. If not, the situation adjustment behavior learning unit 28 and the chain behavior learning unit
29 is activated, and by its operation, the situation adjustment behavior linkage information and the chain behavior linkage information for all the linkage information are added.

(1) The situation input unit 21 acquires situation information around the robot (S1). The status information includes, for example, the presence or absence of a detected object (user or the like), the expression or state of the detected object, the detected object, and the like, which are detected by the distance sensor 11, the pressure-sensitive sensor 14, the imaging camera 10, and the microphone 12. The sound emitted and the action taken by the detected object. The detected status information is edited into the status information sequence obtained so far, and the behavior expression loop, learning loop 1,
It is output to each of the learning loop 2 and the learning loop 3.

First, the behavior expression loop will be described.

(2) The action search unit 23 extracts the association information (candidate association information) having an initial situation matching the latest situation information sequence (current situation Jic) from the association database unit 22 stored in the memory 17. (S2). At the time of this extraction, the situation matching degree A of each piece of link information is calculated and evaluated, and only link information that has obtained a predetermined threshold or more is extracted as candidate link information. At this time, the score information Sc (R) including the situation improvement effect E, the transition probability S, and the ease of execution W of the candidate association information is also extracted. The extracted candidate association information is temporarily stored in the memory 17 together with the score information.

(3) Next, the CPU 1 determines whether or not association information (candidate association information) that matches the input status information Jic has been extracted.
It is determined at 6 (S3).

If at least one candidate association information has been extracted, the process proceeds to S4, and if no candidate association information has been extracted at all, one association is randomly detected from the association database unit 22 in order to conduct an action search. The information is selected as disassociation information, and the process proceeds to S5.

(4) If at least one candidate association information has been extracted, solution association information is determined from each extracted candidate association information (S4).

The expected situation improvement degree Ep = situation improvement effect E × transition probability S × executability W of each piece of candidate association information is calculated, and the CPU 16 determines whether or not it has a value equal to or greater than a predetermined threshold.
As a result of the determination, the candidate association information for which the degree of expectation of situation improvement is less than the predetermined threshold is deleted from the temporary storage of the memory 17. Then, among the remaining candidate linkage information, the one having the highest degree of situation improvement expectation is set as dissolution linkage information, and the process proceeds to S5. If there is no remaining candidate association information, one association information is randomly selected as the association information from the association database unit 22 in order to perform an action search, and the process proceeds to S5.

(5) Next, the CPU 16 outputs the action segment information described in the action information to the action output unit 24 in accordance with the timing information described in the action information of the solution association information (S5).

The action output unit 24 starts operating according to the received action unit information. The operation is that the robot is driven, for example, the LED (9) blinks, a sound is emitted from the speaker 13, the robot 1 itself moves or deforms, and the like.

After executing the dissociation information, the process returns to S1.

Next, the learning loop 1 (reliability learning section 25, non-target action learning section 26) will be described.

(6) First, the action search unit 23 to the action output unit 24
CPU 16 determines whether or not the operation is being performed in accordance with the dissociation information output to (S6).

If the operation has been performed, the process proceeds to S7. If the operation has not been performed, the process returns to S1. When returning to S1, the robot is not performing any operation.

(7) If the robot is operating, it is determined whether or not the operation according to the solution association information has been completed.
It is determined at U16 (S7).

If the processing has been completed, the flow proceeds to S8, and if not, the processing returns to S1. When returning to S1, the robot is operating.

(8) If the operation has been completed, the CPU 16
Calculates the degree of achievement B (Jdc, Jd) between the target situation Jd described in the solution relation information and the situation Jdc after the actual operation, and
It is determined whether or not Jd has been reached (S8). Transition achievement B (Jdc, J
If d) is equal to or greater than the predetermined threshold, the transition based on the selected solution association information is regarded as successful, and if less than the predetermined threshold, it is determined that the transition has failed, and the content of the transition probability information Si is updated as appropriate.
By performing such an operation, reliability learning is performed on the solution association information.

(9) Next, when it is determined that the process has failed in S8, the CPU 16 determines that a transition has been made to a status different from the target status of the disassociation information, and New association information (non-target action association information) in which the situation is replaced with the actual situation (including the score information Sc) is generated by the CPU 16 (S9). The generated link information is stored unless the same link information exists in the link database unit 22.

Return to S1.

Subsequently, the learning loop 2 (other action learning unit 27)
Will be described.

(10) First, the situation input unit 21 detects whether or not the detected object (particularly the user) is taking any action (S10).

If the action of the detected object can be detected, S11
The process proceeds to step S1, and if not detected, proceeds to step S1. When proceeding to S1, the detected object is not in the detection range, the detected object is not performing an action, or the detected object is performing an action but the robot cannot recognize this action. State.

(11) If the detection is successful, the CPU 16 determines whether or not the action of the detected object has been completed based on the status information sequence detected by the status input unit 21 (S11).

When the action has been completed, the process proceeds to S12,
If not, the process proceeds to S1. When going to S1,
The robot keeps detecting the action of the detected object until the action of the detected object ends.

(12) When the action is completed, the CPU 16
The initial situation Ji of the detected object, the situation transition action Ac performed subsequently, and the target situation Jd achieved by the situation transition action Ac are extracted from the situation information sequence, and these are extracted as a set of related information ( (S12). The edited association information is stored unless the same association information exists in the association database unit 22.

By the learning loop 2, the robot can take in the behavior of the detected object as its own new motion pattern. Proceed to S1.

Subsequently, the learning loop 3 (situation adjustment behavior learning unit)
28, the chain action learning unit 29) will be described.

(13) First, the CPU 16 checks whether or not there is newly added association information in the association database unit 22 (S13).

If new association information exists, the process proceeds to S14, and if not, the process proceeds to S1.

To detect new association information, the number of pieces of association information at the time of the previous detection is stored in the memory 17, and this number is compared with the number of pieces of association information at the time of the current detection. And the indexes (identification symbols such as R0001) of all the related information at the time of the previous detection are stored in the memory 17, and the linked information is compared with the index at the time of the current detection one by one. There is a way to judge.

(14) At the time of system startup or when new association information has been added, the situation adjustment behavior learning unit
By 28, new association information (situation adjustment behavior association information) is generated based on the newly added association information, and is stored if the same association information does not exist in the association database unit 22 (S14).

(15) Next, in the chain action learning section 29, new link information (chain action link information) is generated based on the newly added link information, and the similar link information is stored in the link database section 22. If it does not exist, it is stored (S15). S1
Proceed to.

Through the steps (1) to (15), the behavior of the robot is formed / operated.

Specifically, when the robot does not know a valid action, it can try several known actions and discover and learn new effects by using the action search ability. With the ability to replicate behaviors, discover and learn about new behaviors by observing the behavior of others, or 3.
Finding and learning actions that can change a situation that cannot be dealt with into a situation that can be dealt with by using the situation adjustment ability, and 4.Continuously executing multiple actions that were previously known as separate actions using the action planning ability Four new learnings, such as discovering and learning new effects, are performed in almost parallel processing to increase operation patterns.

In the first embodiment as described above,
The robot itself autonomously learns and increases the movement pattern and its usage over time, so that the robot can be continuously used without fading the user's interest. In addition, the actions performed by the robot become more accurate and advanced, and the user can feel that the robot grows.

Further, unlike the related art, a substantially single operation is not performed for a specific input.

Further, it is not necessary for the user to newly input / store an operation pattern, so that the operability can be improved and the robot can be made as close as possible to a living thing with the passage of time.

[0205] Further, by learning the behavior pattern from the detected object, it is possible to learn a new behavior that is beyond the scope of one's own behavioral experience and use it as one of its own behavioral patterns.

[0206] Further, since the success factor and the failure factor of the action are extracted, and the association information that removes the failure factor and captures the success factor is formed, it is possible to cope with the situation with a higher success rate as the user gets more experienced.

Also, by combining a plurality of action patterns,
By using a series of action patterns, it is possible to reach a situation that cannot be achieved by individual actions.

Further, it is possible to realize a learning and growth ability in the behavioral aspect of the robot at an unprecedentedly high level, and to provide a more insatiable robot which is worthwhile for the user to use (foster).

Next, the configuration of a second embodiment in which the pseudo-biological system according to the present invention is implemented by a robot will be described with reference to FIG.

[0210] In the following embodiments, the same components are denoted by the same reference numerals, and redundant description will be omitted.

The feature of the second embodiment is a method of forming a robot action using a computer-readable storage medium storing a program for operating the robot.

FIG. 5 is a block diagram for explaining the second embodiment, and the storage medium 60 stores a program for implementing the first embodiment.

(1) The program in the storage medium 60 is read by a personal computer 61 as an electronic device and stored in the memory 62. The read program is transmitted to the robot 1 by wireless communication, and the control unit in the robot 1 15 is stored in the memory 17.

In this case, the personal computer 61 includes an insertion section 63 into which the storage medium 60 is inserted, a CPU 64 for performing reading / storing / arithmetic operations, and a memory 6 for storing the read program.
2. A transmission unit 65 for transmitting the read program to the robot 1 and an input unit 66 such as a keyboard and a mouse are provided. Further, the robot 1 is provided with a receiver 67 for receiving the transmitted operation signal on a part of the exterior, and the operation signal received by the receiver 67 is stored in the memory 17 in the controller 15 and starts operation.

(2) Alternatively, the storage medium may be inserted into the program reading section 68 provided in the robot 1 to directly operate the program by storing the program in the memory 17 in the control section 15 in the robot 1.

As described above, according to (1) and (2), the robot
After the program is stored in the memory 17 in 1, the same operation as in the first embodiment is performed.

In the second embodiment, the type of the program stored in the same robot can be changed. For example, an active robot program can be used for an active operation such as running well. If the robot program is a spoiled robot program, it becomes a robot that positively performs gestures and squeals that the user is willing to give, and a robot that the user likes can be realized. Therefore, the robot only needs to purchase one robot, and the user can obtain multiple enjoyments simply by changing the type of the program.

The present invention is not limited to the above embodiments, and it goes without saying that various modifications can be made without departing from the spirit of the present invention. For example, it is possible to provide a temperature sensor for measuring the temperature around the robot on the movable part or the main body of the robot. In this way, the robot can handle the ambient temperature as an external situation.

The storage medium of the present invention includes:
Magnetic disk, floppy (registered trademark) disk, hard disk, optical disk (CD-ROM, CD-R, DVD, etc.),
As long as the storage medium can store a computer program and can be read by a computer, such as a magneto-optical disk (MO or the like) or a semiconductor memory, the storage form and the type of the physical medium may be any form.

Also, based on instructions of a computer program installed in the computer from the storage medium, an operating system (OS) running on the computer, MW such as database management software, network, etc.
(Middleware) or the like may execute a part of each process for implementing the present embodiment.

Further, the storage medium in the present invention is not limited to a medium independent of a computer, but may be a storage medium storing or temporarily storing a computer program for the purpose of transmission over a LAN or the Internet, or transmitted over a LAN or the Internet. A storage medium in which the computer program downloaded is stored and temporarily stored is also included.

Further, the number of storage media is not limited to one, and the case where the processing in the present embodiment is executed from a plurality of media is also included in the storage media of the present invention, and the configuration of the media may be any configuration. Good.

Note that the computer according to the present invention executes each process in the present embodiment based on a program stored in a storage medium.
Any configuration such as a single device or a system in which a plurality of devices are connected to a network may be used.

The computer in the present invention is
It is not limited to a personal computer, but also includes an arithmetic processing unit, a microcomputer, and the like included in an information processing device, and collectively refers to devices and devices that can realize the functions of the present invention by a program.

Further, the action forming method is stored as a computer program.

The pseudo-biological system according to the present invention includes, in addition to a robot as a real object, an anthropomorphic or pseudo-biological character by using a computer graphics (CG) or the like to animate the character on a display screen of a computer or the like. It also includes software applications that can be virtualized. In this case, instead of controlling the motor of the robot (1) to move each joint to a desired angle, the action output unit (24) displays an animation in which each joint of the character is moved to a desired angle by coordinate transformation. Display on the screen. Since the system itself is not a robot but a computer, etc., it is possible to execute the action formation method according to the present invention, although there are some restrictions on situation information and executable actions that can be obtained compared to actual robots It is.

The information input to the non-target action learning unit may be either the status information directly input from the status input unit or the information (unsatisfied association information) input from the reliability learning unit. It does not matter.

[0228]

According to the present invention as described above, the simulated creature itself creates and adds a new motion pattern, so that the simulated creature does not get tired of the user.

[Brief description of the drawings]

FIG. 1 is a perspective view of a first embodiment in which a pseudo-creature according to the present invention is implemented by a robot.

FIG. 2 is a block diagram of a first embodiment in which the pseudo creature according to the present invention is implemented by a robot.

FIG. 3 is a block diagram for explaining an operation of the first embodiment in which the pseudo-creature according to the present invention is implemented by a robot.

FIG. 4 is a flowchart illustrating a behavior forming method according to the first embodiment in which the pseudo creature according to the present invention is implemented by a robot.

FIG. 5 is a block diagram for explaining a second embodiment in which the pseudo creature according to the present invention is implemented by a robot.

[Explanation of symbols]

 1 Robot 2 Head 3 Right arm 4 Left arm 5 Right foot 6 Left foot 7 Tail 8 Body 9 LED 10 Imaging camera 11 Capacitive distance sensor 12 Microphone 13 Speaker 14 Pressure sensor 15 Control unit 16 CPU 17 Memory 21 Situation input unit 22 Linkage database unit 23 Behavior search unit 24 Behavior output unit 25 Reliability learning unit 26 Non-target behavior learning unit 27 Others structure learning unit 28 Situation control behavior learning unit 29 Chain behavior learning unit

Claims (28)

[Claims] 1. A simulated creature device that realizes a simulated creature that moves according to a desired motion pattern, wherein external circumstance input means for detecting a circumstance around the simulated creature as an external parameter value. An internal state holding unit that holds a pseudo emotion of the pseudo creature as an internal parameter value; an action for causing the pseudo creature to perform the operation pattern; an external parameter value and an internal parameter value of the pseudo creature before the action; As a set of association information including the external parameter value and the internal parameter value of the pseudo-creature after this action, and the transition probability of this action, a plurality of association database means for storing a plurality of pieces of association information, And selecting the operation pattern to be performed by the simulated creature from the association information of the association database means. A motion search means, and a behavior output means for exercising based on the selected association information, wherein the pseudo-creature apparatus is provided inside the pseudo-creature apparatus and is detected after executing the selected association information. Calculating the similarity between the external parameter value and the retained internal parameter value and the external parameter value and the internal parameter value of the selected association information, and when the similarity is equal to or greater than a predetermined value, A reliability learning unit that increases the transition probability of the selected association information and decreases the transition probability of the selected association information when the similarity is less than a predetermined value. A simulated biological device characterized by the above-mentioned. 2. A simulated creature device for realizing a simulated creature that moves according to a desired motion pattern, wherein external circumstance input means for detecting a circumstance around the simulated creature as an external parameter value. An internal state holding unit that holds a pseudo emotion of the pseudo creature as an internal parameter value; an action for causing the pseudo creature to perform the operation pattern; an external parameter value and an internal parameter value of the pseudo creature before the action; As a set of association information including the external parameter value and the internal parameter value of the pseudo-creature after this action, and the transition probability of this action, a plurality of association database means for storing a plurality of pieces of association information, And selecting the operation pattern to be performed by the simulated creature from the association information of the association database means. A motion search means, and a behavior output means for exercising based on the selected association information, wherein the pseudo-creature apparatus is provided inside the pseudo-creature apparatus and is detected after executing the selected association information. Calculating the similarity between the external parameter value and the retained internal parameter value, and the external parameter value and the internal parameter value of the selected association information, and when the similarity is less than a predetermined value, The external parameter value and the internal parameter value before executing the selected association information, the behavior of the selected association information, and the external parameter value and the internal parameter value after the behavior of the pseudo-creature. If the generated association information is not stored in the association database means, it is stored as the association information. Pseudo organism apparatus being characterized in that; and a nontarget action learning means. 3. A simulated creature device for realizing a simulated creature that moves according to a desired motion pattern, wherein external circumstance input means for detecting a circumstance around the simulated creature as an external parameter value. An internal state holding unit that holds a pseudo emotion of the pseudo creature as an internal parameter value; an action for causing the pseudo creature to perform the operation pattern; an external parameter value and an internal parameter value of the pseudo creature before the action; As a set of association information including the external parameter value and the internal parameter value of the pseudo-creature after this action, and the transition probability of this action, a plurality of association database means for storing a plurality of pieces of association information, And selecting the operation pattern to be performed by the simulated creature from the association information of the association database means. In a simulated biological device having: motion search means; and action output means for exercising based on the selected association information.The detected object is detected by the external situation input means, and the detected object performs an action. In this case, the action performed by the object, the external parameter value of the object extracted from the external parameter value detected before and after the action, and the external parameter value detected before and after the action And an internal parameter value of the detected object estimated from the following, as a set of related information, and generating the related information, if the stored related information is not stored in the related database means; A simulated biological device, comprising: 4. A simulated creature that realizes a simulated creature that moves according to a desired motion pattern, wherein external circumstance input means for detecting a circumstance around the simulated creature as an external parameter value. An internal state holding unit that holds a pseudo emotion of the pseudo creature as an internal parameter value; an action for causing the pseudo creature to perform the operation pattern; an external parameter value and an internal parameter value of the pseudo creature before the action; As a set of association information including the external parameter value and the internal parameter value of the pseudo-creature after this action, and the transition probability of this action, a plurality of association database means for storing a plurality of pieces of association information, And selecting the operation pattern to be performed by the simulated creature from the association information of the association database means. A motion search means; and a behavior output means for exercising based on the selected association information. In the pseudo-biological device, based on the association information stored in the association database means, the external parameter value after the action and the internal Extracting a plurality of pieces of the association information having parameter values that are substantially the same as each other, and among the extracted pieces of the association information, a transition probability is lower than the other pieces of the association information from which the transition information is extracted before the action of the association information; The external parameter value and the internal parameter value are first
And the external parameter value and the internal parameter value before the action of the other related information are used as a second parameter value.
The first parameter value, the action of changing the first parameter value to the second parameter value, and the external parameter value and the internal parameter value after the action of the other related information And a situation adjusting action learning means for generating a set of association information, and storing the generated association information if the generated association information is not stored in the association database means. 5. A simulated creature device for realizing a simulated creature that moves according to a desired motion pattern, wherein external circumstance input means for detecting a circumstance around the simulated creature as an external parameter value. An internal state holding means for holding, as an internal parameter value, a simulated emotion having at least a pleasant sensation of the simulated creature; an action for causing the simulated creature to perform the operation pattern; and an external parameter value of the simulated creature before the action. And an internal parameter value, an external parameter value and an internal parameter value of the pseudo-creature after this action, a transition probability of this action, as a set of linked information, as a set of linked information, The operation pattern corresponding to the internal parameter value and performed by the pseudo-creature is stored in the association database means. A pseudo-biological device having: a behavior search means for selecting from the linkage information; and a behavior output means for exercising based on the selected linkage information, wherein the first one of the linkage information stored in the linkage database means A second information having an external parameter value and an internal parameter value of the simulated creature before the action that substantially matches the external parameter value and the internal parameter value of the simulated creature after the action of the first linked information. Associated information, and extracting, the pleasant feeling in the external parameter value and the internal parameter value of the pseudo-creature before the action of the first linked information,
The external parameter value and the internal parameter of the simulated creature before the action of the first link information when the external parameter value and the internal parameter value of the simulated creature after the action of the second link information are lower than the pleasant emotion. A value, an action from the first link information to the second link information,
An external parameter value and an internal parameter value of the pseudo creature after the action of the second link information are generated as a set of link information, and the generated link information is not stored in the link database means. A simulated biological device, comprising: a chain action learning means for storing as linked information. 6. A simulated creature device for realizing a simulated creature that moves according to a desired operation pattern, wherein external circumstance input means for detecting a circumstance around the simulated creature as an external parameter value. An internal state holding means for holding, as an internal parameter value, a simulated emotion having at least a pleasant sensation of the simulated creature; an action for causing the simulated creature to perform the operation pattern; and an external parameter value of the simulated creature before the action. And an internal parameter value, an external parameter value and an internal parameter value of the pseudo-creature after this action, a transition probability of this action, as a set of linked information, as a set of linked information, The operation pattern corresponding to the internal parameter value and performed by the pseudo-creature is stored in the association database means. A pseudo-biological device having an action search means for selecting from the link information and an action output means for exercising based on the selected link information, wherein the first one of the link information stored in the link database means; A second association having associated information and an external parameter value and an internal parameter value of the pseudo-creature before the action, which substantially coincides with an external parameter value and an internal parameter value of the pseudo-life after the action of the first link information; Information, and third association information having an external parameter value and an internal parameter value of the simulated creature before the action that substantially match the external parameter value and the internal parameter value of the simulated creature after the action of the second association information. And the pleasant emotion in the external parameter value and the internal parameter value of the pseudo-creature before the action of the first association information,
The external parameter value and the internal parameter of the simulated creature before the action of the first link information when the external parameter value and the internal parameter value of the simulated creature after the action of the third link information are lower than the pleasant emotion. A value and an action and a second action of the second link information from the first link information
The action from the link information to the third link information and the external parameter value and the internal parameter value of the pseudo creature after the action of the third link information are generated as a set of link information. A chain behavior learning means for storing the link information as link information if the link information is not stored in the link database means. 7. A simulated creature device for realizing a simulated creature that moves according to a desired motion pattern, wherein external circumstance input means for detecting a circumstance around the simulated creature as an external parameter value. An internal state holding means for holding, as an internal parameter value, a simulated emotion having at least a pleasant sensation of the simulated creature; an action for causing the simulated creature to perform the operation pattern; and an external parameter value of the simulated creature before the action. And an internal parameter value, an external parameter value and an internal parameter value of the pseudo-creature after this action, a transition probability of this action, as a set of linked information, as a set of linked information, The operation pattern corresponding to the internal parameter value and performed by the pseudo-creature is stored in the association database means. A behavior search means for selecting from the linkage information; and an activity output means for exercising based on the selected linkage information. A pseudo-biological device, comprising: a first one of the linkage information stored in the linkage database means. A second information having an external parameter value and an internal parameter value of the simulated creature before the action that substantially matches the external parameter value and the internal parameter value of the simulated creature after the action of the first linked information. N (where n is a natural number of 2 or more) with the linkage information as one chain (n +
1) extracting the pieces of the related information, wherein the pleasant feeling in the external parameter value and the internal parameter value of the pseudo creature before the action of the first related information is:
When the external parameter value and the internal parameter value of the pseudo creature after the action of the second linked information of the n-th chain are lower than the pleasant emotion, the outside of the pseudo creature before the action of the first linked information. The parameter value and the internal parameter value and (n + 1) actions from the first link information to the (n + 1) link information are arranged in the (n + 1) link information. And the external parameter value and the internal parameter value of the pseudo creature after the action of the (n + 1) th association information are generated as a set of association information, and the generated association is generated. A chain behavior learning means for storing information as link information if the information is not stored in the link database means; 8. A simulated creature apparatus which realizes a simulated creature moving by a desired motion pattern and has an association database means for storing the operation pattern as association information including a plurality of parameters, wherein the situation around the simulated organism is provided. Detecting the pseudo-creature as an external parameter value, and holding the pseudo-emotion of the pseudo-creature as an internal parameter value, and corresponding to the external parameter value and the internal parameter value, and performing the operation pattern to be performed by the pseudo-creature. An action search step of selecting from the link information of the link database means; an action output step of exercising based on the selected link information; and the external parameter value detected after executing the selected link information. The retained internal parameter values and the external parameters of the selected association information And calculating the similarity between the internal parameter value and the similarity. If the similarity is equal to or greater than a predetermined value, the transition probability of the selected association information is increased, and the similarity is less than a predetermined value. A reliability learning step of reducing the transition probability of the selected association information. 9. A simulated creature device which realizes a simulated creature moving by a desired motion pattern and has an association database means for storing the operation pattern as association information including a plurality of parameters, wherein Detecting the pseudo-creature as an external parameter value, and holding the pseudo-emotion of the pseudo-creature as an internal parameter value, and corresponding to the external parameter value and the internal parameter value, and performing the operation pattern to be performed by the pseudo-creature. An action search step of selecting from the link information of the link database means; an action output step of exercising based on the selected link information; and the external parameter value detected after executing the selected link information. The retained internal parameter values and the external parameters of the selected association information And calculating the similarity between the internal parameter value and the external parameter value and the internal parameter value before executing the selected association information, if the similarity is less than a predetermined value. Generated as a set of the action of the link information and the external parameter value and the internal parameter value after the action of the pseudo-creature, and the generated link information is stored in the link database means. And a non-target behavior learning step of storing as the association information if the behavior is not present. 10. A simulated creature device that realizes a simulated creature that moves according to a desired motion pattern and has an association database means that stores the operation pattern as association information including a plurality of parameters. Detecting the pseudo-creature as an external parameter value, and holding the pseudo-emotion of the pseudo-creature as an internal parameter value, and corresponding to the external parameter value and the internal parameter value, and performing the operation pattern to be performed by the pseudo-creature. An action search step of selecting from the association information of the association database means; an action output step of exercising based on the selected association information; and a detected object is detected by the external situation input means, and the detected object is When an action is performed, the action performed by the detected object and the outside detected before and after the action are performed. An external parameter value of the detected object extracted from the parameter value and an internal parameter value of the detected object estimated from the external parameter value detected before and after the action are generated as a set of association information. A learning process of the imitation creature in the imitation creature device, further comprising the step of storing the generated association information if the generated association information is not stored in the association database means. 11. A simulated creature device that realizes a simulated creature that moves by a desired motion pattern and has an association database means that stores the operation pattern as association information including a plurality of parameters, Detecting the pseudo-creature as an external parameter value, and holding the pseudo-emotion of the pseudo-creature as an internal parameter value, and corresponding to the external parameter value and the internal parameter value, and performing the operation pattern to be performed by the pseudo-creature. An action search step of selecting from the link information of the link database means, an action output step of exercising based on the selected link information, and the external action after the action from the link information stored in the link database means. A plurality of the series having parameter values and internal parameter values having substantially the same parameter value. Extracting information, in the extracted the associated information, the external parameter value and the internal parameter value before the action of the associated information below are other of said associated information near the transition probabilities are extracted first
And the external parameter value and the internal parameter value before the action of the other related information are used as a second parameter value.
And the first parameter value is the second parameter value.
And an external parameter value and an internal parameter value after the action of the other related information are generated as a set of related information, and the generated related information is stored in the related database means. And a situation adjusting action learning step of storing the information as linkage information if the action is not stored in the pseudo-creature apparatus. 12. A simulated creature apparatus which realizes a simulated creature moving by a desired motion pattern and has an association database means for storing the operation pattern as association information including a plurality of parameters, wherein Detecting as a parameter value an external parameter value; holding a pseudo emotion having at least a pleasant feeling of the pseudo-creature as an internal parameter value; performing on the pseudo-creature corresponding to the external parameter value and the internal parameter value. An action search step of selecting the action pattern to be performed from the association information of the association database means; an action output step of exercising based on the selected association information; and the action information stored in the association database means. , First association information, and external parameters of the pseudo-creature after the action of the first association information. Extracting second association information having an external parameter value and an internal parameter value of the pseudo creature before the action that substantially matches the meter value and the internal parameter value, and extracting the pseudo information before the action of the first association information. The pleasant feeling in the external parameter value and the internal parameter value of the living thing,
The external parameter value and the internal parameter of the simulated creature before the action of the first link information when the external parameter value and the internal parameter value of the simulated creature after the action of the second link information are lower than the pleasant emotion. A value, an action from the first link information to the second link information,
An external parameter value and an internal parameter value of the pseudo creature after the action of the second link information are generated as a set of link information, and the generated link information is not stored in the link database means. And a chain action learning step of storing as link information. 13. A simulated creature device that realizes a simulated creature that moves by a desired motion pattern and has an association database means that stores the operation pattern as association information including a plurality of parameters. Detecting as a parameter value an external parameter value; holding a pseudo emotion having at least a pleasant feeling of the pseudo-creature as an internal parameter value; performing on the pseudo-creature corresponding to the external parameter value and the internal parameter value. An action search step of selecting the action pattern to be performed from the association information of the association database means; an action output step of exercising based on the selected association information; and the action information stored in the association database means. , First association information, and external parameters of the pseudo-creature after the action of the first association information. Second association information having an external parameter value and an internal parameter value of the simulated creature before the action that substantially matches the meter value and the internal parameter value, and an external parameter of the simulated creature after the action of the second association information Extracting the third association information having the external parameter value and the internal parameter value of the simulated creature before the action that substantially matches the value and the internal parameter value, and extracting the simulated creature before the action of the first association information. The pleasant emotion in the external parameter value and the internal parameter value,
The external parameter value and the internal parameter of the simulated creature before the action of the first link information when the external parameter value and the internal parameter value of the simulated creature after the action of the third link information are lower than the pleasant emotion. A value and an action and a second action of the second link information from the first link information
The action from the link information to the third link information and the external parameter value and the internal parameter value of the pseudo creature after the action of the third link information are generated as a set of link information. A chain action learning step of storing the linked information as link information if the link information is not stored in the link database means. 14. A simulated creature device that realizes a simulated creature that moves according to a desired motion pattern and has an association database means for storing the operation pattern as association information including a plurality of parameters, wherein Detecting as a parameter value an external parameter value; holding a pseudo emotion having at least a pleasant feeling of the pseudo-creature as an internal parameter value; performing on the pseudo-creature corresponding to the external parameter value and the internal parameter value. An action retrieval step of selecting the action pattern to be performed from the association information of the association database means; an action output step of exercising based on the selected association information; and an action output step of the association information stored in the association database means. , First association information, and external parameters of the pseudo-creature after the action of the first association information. An n (where n is a natural number of 2 or more) chain is defined as a chain including the second association information having the external parameter value and the internal parameter value of the simulated creature before the action, which substantially matches the meter value and the internal parameter value. (N +
1) extracting the pieces of the related information, wherein the pleasant feeling in the external parameter value and the internal parameter value of the pseudo creature before the action of the first related information is:
When the external parameter value and the internal parameter value of the pseudo creature after the action of the second linked information of the n-th chain are lower than the pleasant emotion, the outside of the pseudo creature before the action of the first linked information. The parameter value and the internal parameter value and (n + 1) actions from the first link information to the (n + 1) link information are arranged in the (n + 1) link information. And the external parameter value and the internal parameter value of the pseudo creature after the action of the (n + 1) th association information are generated as a set of association information, and the generated association is generated. A chain action learning step of storing information as link information if the information is not stored in the link database means. 15. A computer-readable program for operating a simulated creature moving with a desired motion pattern and operating a simulated creature apparatus having an association database means for storing the operation pattern as association information including a plurality of parameters. A storage medium that stores the pseudo-creature's surroundings as an external parameter value, and stores a pseudo-emotion of the pseudo-creature as an internal parameter value, and corresponds to the external parameter value and the internal parameter value. Then, the operation pattern to be performed by the simulated creature is selected from the association information of the association database means, exercise is performed based on the selected association information, and the motion is detected after the selected association information is executed. The external parameter value and the retained internal parameter value Calculating a similarity between the external parameter value and the internal parameter value of the obtained association information, and when the similarity is equal to or greater than a predetermined value, increasing the transition probability of the selected association information. A computer-readable storage medium storing a program for reducing the transition probability of the selected association information when the similarity is less than a predetermined value; 16. A computer readable program for operating a simulated creature that moves according to a desired operation pattern and operating a simulated creature apparatus having an association database means for storing the operation pattern as association information including a plurality of parameters. A storage medium that stores the pseudo-creature's surroundings as an external parameter value, and stores a pseudo-emotion of the pseudo-creature as an internal parameter value, and corresponds to the external parameter value and the internal parameter value. Then, the operation pattern to be performed by the simulated creature is selected from the association information of the association database means, exercise is performed based on the selected association information, and the motion is detected after the selected association information is executed. The external parameter value and the retained internal parameter value Calculating a similarity between the external parameter value and the internal parameter value of the obtained association information. If the similarity is less than a predetermined value, the external information before the execution of the selected association information is calculated. A parameter value and an internal parameter value, an action of the selected association information,
If the external parameter value and the internal parameter value after the behavior of the simulated creature are generated as a set of link information, and if the generated link information is not stored in the link database means, the link information is used as the link information. A computer-readable storage medium storing a program to be stored. 17. A computer readable program for operating a simulated creature that moves according to a desired operation pattern and operating a simulated creature apparatus having an association database means for storing the operation pattern as association information including a plurality of parameters. A storage medium that stores the pseudo-creature's surroundings as an external parameter value, and stores a pseudo-emotion of the pseudo-creature as an internal parameter value, and corresponds to the external parameter value and the internal parameter value. Then, the operation pattern to be performed by the simulated creature is selected from the association information of the association database means, and exercise is performed based on the selected association information, and the detected object is detected by the external situation input means. When the detected object performs an action, the action performed by the detected object is The external parameter value of the detected object extracted from the external parameter value detected before and after this action, and the internal parameter value of the detected object estimated from the external parameter value detected before and after this action , As a set of association information, and a computer-readable storage medium storing a program for storing the generated association information unless the generated association information is stored in the association database means. 18. A computer readable program for operating a simulated creature that moves according to a desired operation pattern and operating a simulated creature apparatus having an association database means for storing the operation pattern as association information including a plurality of parameters. A storage medium that stores the pseudo-creature's surroundings as an external parameter value, and stores a pseudo-emotion of the pseudo-creature as an internal parameter value, and corresponds to the external parameter value and the internal parameter value. Then, the operation pattern to be performed by the simulated creature is selected from the association information of the association database means, the exercise is performed based on the selected association information, and an action is performed from the association information stored in the association database means. The subsequent external parameter value and internal parameter value are substantially the same. A plurality of the association information having a parameter value are extracted, and, among the extracted association information, the external parameter value and the internal of the association information before the action of the association information whose transition probability is lower than the other associated information extracted are also lower. Parameter value first
And the external parameter value and the internal parameter value before the action of the other related information are used as a second parameter value.
And the first parameter value is the second parameter value.
And an external parameter value and an internal parameter value after the action of the other related information are generated as a set of related information, and the generated related information is stored in the related database means. A computer-readable storage medium storing a program to be stored as association information if the program is not stored in the storage medium. 19. A computer-readable program for operating a pseudo-creature that moves according to a desired motion pattern and operating a pseudo-creature device having an association database means for storing the operation pattern as association information including a plurality of parameters. The storage medium stored in the, the situation around the pseudo-creature is detected as an external parameter value, pseudo-feelings having at least a pleasant feeling of the pseudo-creature are held as internal parameter values, the external parameter value and the external parameter value Corresponding to the internal parameter value, the operation pattern to be performed by the simulated creature is selected from the association information of the association database means, the exercise is performed based on the selected association information, and the motion pattern is stored in the association database means. Of the linkage information,
The second related information having an external parameter value and an internal parameter value of the pseudo creature before the action, which substantially matches the external parameter value and the internal parameter value of the pseudo creature after the action of the first linked information. And the associated information of, the pleasant feeling in the external parameter value and the internal parameter value of the pseudo-creature before the action of the first linked information,
The external parameter value and the internal parameter of the simulated creature before the action of the first link information when the external parameter value and the internal parameter value of the simulated creature after the action of the second link information are lower than the pleasant emotion. A value, an action from the first link information to the second link information,
An external parameter value and an internal parameter value of the pseudo creature after the action of the second link information are generated as a set of link information, and the generated link information is not stored in the link database means. A computer-readable storage medium storing a program to be stored as association information. 20. A computer-readable program for operating a pseudo-creature that moves according to a desired motion pattern and operating a pseudo-creature device having an association database means for storing the operation pattern as association information including a plurality of parameters. The storage medium stored in the, the situation around the pseudo-creature is detected as an external parameter value, pseudo-feelings having at least a pleasant feeling of the pseudo-creature are held as internal parameter values, the external parameter value and the external parameter value Corresponding to the internal parameter value, the operation pattern to be performed by the simulated creature is selected from the association information of the association database means, the exercise is performed based on the selected association information, and the motion pattern is stored in the association database means. Of the linkage information,
The second related information having an external parameter value and an internal parameter value of the pseudo creature before the action, which substantially matches the external parameter value and the internal parameter value of the pseudo creature after the action of the first linked information. A third information having an external parameter value and an internal parameter value of the pseudo creature before the action, which substantially coincides with the external parameter value and the internal parameter value of the pseudo creature after the action of the second link information.
Of the pseudo-creatures before and after the action of the first link information in the external parameter value and the internal parameter value,
The external parameter value and the internal parameter of the simulated creature before the action of the first link information when the external parameter value and the internal parameter value of the simulated creature after the action of the third link information are lower than the pleasant emotion. A value and an action and a second action of the second link information from the first link information
The action from the link information to the third link information, and the external parameter value and the internal parameter value of the pseudo creature after the action of the third link information are generated as a set of link information. A computer-readable storage medium storing a program for storing the linked information if the linked information is not stored in the linked database means. 21. A computer-readable program for operating a pseudo-creature that moves according to a desired motion pattern and operating a pseudo-creature device having an association database means for storing the operation pattern as association information including a plurality of parameters. The storage medium stored in the, the situation around the pseudo-creature is detected as an external parameter value, pseudo-feelings having at least a pleasant feeling of the pseudo-creature are held as internal parameter values, the external parameter value and the external parameter value Corresponding to the internal parameter value, the operation pattern to be performed by the simulated creature is selected from the association information of the association database means, the exercise is performed based on the selected association information, and the motion pattern is stored in the association database means. Of the linkage information,
The second related information having an external parameter value and an internal parameter value of the pseudo creature before the action, which substantially matches the external parameter value and the internal parameter value of the pseudo creature after the action of the first linked information. (N + 1 is a natural number of 2 or more), and (n + 1) pieces of the linked information are extracted. The outside of the pseudo-creature before the action of the first linked information The pleasant emotion in the parameter value and the internal parameter value is:
When the external parameter value and the internal parameter value of the pseudo creature after the action of the second linked information of the n-th chain are lower than the pleasant emotion, the outside of the pseudo creature before the action of the first linked information. The parameter value and the internal parameter value and (n + 1) actions from the first link information to the (n + 1) link information are arranged in the (n + 1) link information. The action performed in the given order and the external parameter value and the internal parameter value of the pseudo creature after the action of the (n + 1) th link information are generated as a set of link information, and the generated link is generated. If the information is not stored in the association database means, a computer-readable storage medium storing a program for storing the information as the association information. 22. A computer program for operating a pseudo-creature that moves according to a desired motion pattern and operating a pseudo-creature device having an association database means for storing the operation pattern as association information including a plurality of parameters, A function of detecting a situation around the simulated creature as an external parameter value, a function of retaining a simulated emotion of the simulated creature as an internal parameter value, and a function performed on the simulated creature in response to the external parameter value and the internal parameter value. A function of selecting the operation pattern to be performed from the association information of the association database means, a function of exercising based on the selected association information, and an external function detected after the selected association information is executed. The parameter value and the retained internal parameter value, and the selected A function of calculating a similarity between the external parameter value and the internal parameter value of the association information; and, when the similarity is equal to or more than a predetermined value, increasing the transition probability of the selected association information, If the similarity is less than a predetermined value, a function of reducing the transition probability of the selected association information. 23. A computer program for operating a pseudo-creature that moves by a desired motion pattern and operating a pseudo-creature device having an association database means for storing the operation pattern as association information including a plurality of parameters, A function of detecting a situation around the simulated creature as an external parameter value, a function of retaining a simulated emotion of the simulated creature as an internal parameter value, and a function performed on the simulated creature in response to the external parameter value and the internal parameter value. A function of selecting the operation pattern to be performed from the association information of the association database means, a function of exercising based on the selected association information, and an external function detected after the selected association information is executed. The parameter value and the retained internal parameter value, and the selected A function of calculating a similarity between the external parameter value and the internal parameter value of the association information; and, if the similarity is less than a predetermined value, the external parameter before the selected association information is executed. Values and internal parameter values, the action of the selected association information,
The external parameter value and the internal parameter value after the behavior of the simulated creature are generated as a set of link information. If the generated link information is not stored in the link database means, the link information is stored as link information. Computer program that realizes the function of 24. A computer program for operating a pseudo-creature that moves by a desired motion pattern and operating a pseudo-creature apparatus having an association database means for storing the operation pattern as association information including a plurality of parameters. A function of detecting a situation around the simulated creature as an external parameter value, a function of retaining a simulated emotion of the simulated creature as an internal parameter value, and corresponding to the external parameter value and the internal parameter value, performed on the simulated creature. A function of selecting the operation pattern to be performed from the association information of the association database means, a function of exercising based on the selected association information, and a detection object detected by the external situation input means. Performs an action, the action performed by the detected object and An external parameter value of the detected object extracted from the external parameter value detected before and after the movement, and an internal parameter value of the detected object estimated from the external parameter value detected before and after the action, A computer program for realizing a function of generating a set of association information, and storing the generated association information if the generated association information is not stored in the association database means. 25. A computer program for operating a pseudo-creature that moves according to a desired motion pattern and operating a pseudo-creature device having an association database means for storing the operation pattern as association information including a plurality of parameters, A function of detecting a situation around the simulated creature as an external parameter value, a function of retaining a simulated emotion of the simulated creature as an internal parameter value, and corresponding to the external parameter value and the internal parameter value, performed on the simulated creature. A function of selecting the operation pattern to be performed from the association information of the association database means, a function of exercising based on the selected association information, and the association information stored in the association database means. Parameters with almost the same external parameter value and internal parameter value Extracting the plurality of pieces of association information having data values, and among the extracted pieces of association information, the external parameter values and the internal values of the association information whose transition probabilities are lower than those of the other pieces of association information from which the transition probability is extracted. Parameter value first
And the external parameter value and the internal parameter value before the action of the other related information are used as a second parameter value.
And the first parameter value is the second parameter value.
And an external parameter value and an internal parameter value after the action of the other related information are generated as a set of related information, and the generated related information is stored in the related database means. A computer program for realizing the function of storing the information as association information if the information is not stored in the computer program. 26. A computer program for operating a pseudo-creature that moves by a desired motion pattern and operating a pseudo-creature device having an association database means for storing the operation pattern as association information including a plurality of parameters, A function of detecting a situation around the simulated creature as an external parameter value, a function of holding a simulated emotion having at least a pleasant feeling of the simulated creature as an internal parameter value, and corresponding to the external parameter value and the internal parameter value, A function of selecting the operation pattern to be performed by the simulated creature from the link information of the link database means, a function of exercising based on the selected link information, and a function of the link information stored in the link database means. Of which, the first related information and the action of the first related information Second association information having an external parameter value and an internal parameter value of the simulated creature before the action that substantially matches the external parameter value and the internal parameter value of the simulated creature after, and the first association The pleasant feeling in the external parameter value and the internal parameter value of the simulated creature before the action of the information,
The external parameter value and the internal parameter of the simulated creature before the action of the first link information when the external parameter value and the internal parameter value of the simulated creature after the action of the second link information are lower than the pleasant emotion. A value, an action from the first link information to the second link information,
An external parameter value and an internal parameter value of the pseudo creature after the action of the second link information are generated as a set of link information, and the generated link information must be stored in the link database means. A computer program for realizing a function of storing the information as association information. 27. An association database means for operating a movable creature by operating the movable portion according to a desired operation pattern and storing the operation pattern for operating the movable portion as association information including a plurality of parameters. A function of detecting a situation around the pseudo-creature as an external parameter value; a function of retaining a pseudo-emotion having at least a pleasant feeling of the pseudo-creature as an internal parameter value; and the external parameter value. And a function of selecting the operation pattern to be performed by the simulated creature from the association information of the association database means in accordance with the internal parameter value, a function of exercising based on the selected association information, and an association database. First association information among the association information stored in the means; The second association information having an external parameter value and an internal parameter value of the simulated creature before the action that substantially matches the external parameter value and the internal parameter value of the simulated creature after the action of the first association information, Extracting third association information having an external parameter value and an internal parameter value of the simulated creature before the action, which substantially matches the external parameter value and the internal parameter value of the simulated creature after the action of the second association information. The pleasant feeling in the external parameter value and the internal parameter value of the pseudo creature before the action of the first association information,
The external parameter value and the internal parameter of the simulated creature before the action of the first link information when the external parameter value and the internal parameter value of the simulated creature after the action of the third link information are lower than the pleasant emotion. A value and an action and a second action of the second link information from the first link information
The action from the link information to the third link information, and the external parameter value and the internal parameter value of the pseudo creature after the action of the third link information are generated as a set of link information. And a function of storing the linked information as the linked information if the linked information is not stored in the link database means. 28. A computer program for operating a pseudo-creature that moves according to a desired motion pattern, and operating a pseudo-creature device having an association database means for storing the operation pattern as association information including a plurality of parameters. A function of detecting a situation around the simulated creature as an external parameter value, a function of holding a simulated emotion having at least a pleasant feeling of the simulated creature as an internal parameter value, and corresponding to the external parameter value and the internal parameter value, A function of selecting the operation pattern to be performed by the simulated creature from the link information of the link database means, a function of exercising based on the selected link information, and a function of the link information stored in the link database means. Of which, the first related information and the action of the first related information The second linked information having the external parameter value and the internal parameter value of the simulated creature before the action, which substantially matches the external parameter value and the internal parameter value of the simulated creature after, is defined as n (where n is (N + 2) chain (n +
1) extracting the associated information, the pleasant feeling in the external parameter value and the internal parameter value of the pseudo creature before the action of the first associated information,
When the external parameter value and the internal parameter value of the pseudo creature after the action of the second linked information of the n-th chain are lower than the pleasant emotion, the outside of the pseudo creature before the action of the first linked information. The parameter value and the internal parameter value and (n + 1) actions from the first link information to the (n + 1) link information are arranged in the (n + 1) link information. And performing the action performed in the given order, and the external parameter value and the internal parameter value of the pseudo creature after the action of the (n + 1) th link information as a set of link information. A computer program for realizing a function of storing information as association information if the information is not stored in the association database means.