JPH11143849A - Action generation device, action generation method and action generation program recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate an action extremely approximate to the action of a real organism by generating an autonomous action to a condition on the basis of concepts of learning, oblivion and consciousness. SOLUTION: An artificial organism model 1 is composed of an external information detection part 2, a knowledge set storage part 3, a recognition evaluation part 4, a condition storage part 5, an action plan part 6, and an operation output part 7. The knowledge set storage part 3 stores a relationship between recognition result of the external information which the external information detection part 2 detects in a process in which the artificial organism model 1 acts and a priori action target stored in advance in the knowledge set storage part 3 together with the degree of association and the degree of storage. The recognition evaluation 4 refers to, recognizes and evaluates storage contents of the knowledge set storage part 3 with regard to each of the external information. The condition storage part 5 stores evaluation result in the recognition evaluation part 4. The action plan part 6 decides an action to be executed according to evaluation results stored in the condition storage part 5 and outputs to the action output part 7.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a behavior generating apparatus for generating an autonomous behavior to be executed by a simulated creature such as a robot and an artificial creature model such as a virtual creature displayed on a display of a computer system. The present invention relates to an action generation method and an action generation program recording medium.

[0002]

2. Description of the Related Art As simulated creatures such as robots and virtual creatures displayed on displays of computer systems, etc.
Conventionally, there is an artificial creature model that behaves autonomously according to information input from the outside. Conventionally,
Many attempts have been made to make the behavior performed by this type of artificial creature model as close as possible to the behavior of a real creature.

[0003] Here, the real creatures acquire knowledge through learning based on experience, and by using the acquired knowledge according to the situation, gradually increase the number of situations in which they can act as a reaction, thereby increasing the variety of situations. Generate behavior patterns autonomously. In addition, when a real creature does not encounter a particular situation for a long period of time, it forgets to react to the situation, and a phenomenon called forgetting occurs such that the response to the situation is delayed or completely unresponsive. Furthermore, in a real creature, there is a consciousness that when there are a plurality of events to be responded at the same time, the creature acts only on the event having the highest value to respond to itself.

Accordingly, learning, forgetting, and consciousness are indispensable for generating the behavior of a real creature, and in order to approximate the behavior of an artificial creature model to the behavior of a real creature, the behavior to be performed with respect to the situation is required. The concept of learning, forgetting and consciousness is required in the generation process.

[0005]

However, in the conventional artificial creature model, the relationship between the external situation as a stimulus and the action as an output is fixedly determined in advance, and the action according to the situation is based on experience. What cannot be acquired as knowledge by learning is common. In addition, there has been proposed a configuration in which an action as a response pattern is generated based on a history of an external stimulus or an internal state, but also in this case, an appropriate response pattern is set for a situation not assumed in advance. Cannot be generated. In such a case, it is difficult to anticipate all possible situations in advance, and a large-capacity storage means is required to presume and store many situations, which is not practical.

Further, in the conventional artificial creature model, when generating a reaction pattern based on the past stimuli and internal state histories, the weight value of new history data is increased and the weight value of old history data is decreased. In some cases, the concept of forgetting is incorporated into the generation of reaction patterns.However, in such a method of generating a reaction pattern, a large-capacity storage is used to store history data on individual stimuli and internal states. There is a problem that requires measures.

[0007] Further, in the conventional artificial creature model, there is no technique for generating a response action to a stimulus using the concept of consciousness. As a result of trying to generate a reaction action, there is a problem that the processing time is lengthened and the execution of the reaction action is delayed, or a so-called deadlock state occurs because the priority of the reaction action cannot be determined.

An object of the present invention is to generate an action that is very close to the action of a real creature by generating an autonomous action for a situation based on the concepts of learning, forgetting, and consciousness. An object of the present invention is to provide a generation device, an action generation method, and an action generation program recording medium.

[0009]

The invention according to claims 1, 9 and 17 detects external information and acquires a relationship between the detected external information and each of a plurality of predetermined basic action targets. It is characterized in that it is stored as a goal and an action to be performed on external information is determined based on a basic action goal or an acquired action goal.

According to the invention described in claims 1, 9 and 17, a new action goal is learned from the relationship between the detected external information and a predetermined basic action goal as an acquisition action goal. The action to be performed on the external information is determined based on the action goal including Therefore, an action target for determining an action to be performed on external information is supplemented by learning, and an action to be performed on external information whose relationship with the action target is not predetermined is determined autonomously. Is done.

The invention described in claims 2, 10 and 18 stores a degree of storage indicating a period in which the acquired action target is to be stored for each of the plurality of external information, and individually stores each acquired action target based on the stored degree. It is characterized by erasing.

According to the second, tenth and eighteenth aspects of the present invention, new action targets acquired by learning are individually deleted based on the degree of memory. Therefore, a part of the newly acquired action goal is not erased, and the action goal for determining the action to be performed on the external information is not excessively stored.

The inventions described in claims 3, 11 and 19 are characterized in that each of the storage degrees is individually reduced with the passage of time.

[0014] In the inventions set forth in claims 3, 11 and 19, the acquired action targets are individually deleted according to the degree of memory that decreases with the passage of time. Therefore,
When determining an action to be performed on external information, a forgetting phenomenon that occurs in a real creature is reproduced.

[0015] The invention described in claims 4, 12 and 20 is characterized in that each of the storage degrees is individually increased or decreased according to the use state of the acquisition action target for each external information.

According to the fourth, twelfth and twentieth aspects of the present invention, the acquired action targets are individually deleted in accordance with the degree of memory that decreases in accordance with the state of use. Therefore, when an action to be performed on external information is determined, a forgetting phenomenon that occurs in a real creature according to the appearance frequency of each external information is more faithfully reproduced.

The inventions described in claims 5, 13 and 21 are characterized in that importance levels representing the priorities of a plurality of basic action goals are stored.

In the fifth, thirteenth, and twenty-first aspects of the invention, the priority of each of the plurality of basic action goals is represented as importance. Therefore, when determining the action to be performed on the external information, the detected external information is related to the plurality of basic action goals by referring to the importance assigned to each of the plurality of basic action goals. Also, a single action is determined for the external information.

The invention described in claims 6, 14 and 22 is characterized in that the relevance of each of the plurality of external information with the basic action goal is stored.

In the inventions described in claims 6, 14 and 22, each of a plurality of pieces of external information related to each basic action goal is assigned a degree of association with the basic action goal. Therefore, when determining the action to be performed on the external information, by referring to the degree of relevance given to each of the external information, a plurality of external information related to the same basic action goal was detected at the same time. In some cases, a single action for a single piece of external information is determined.

The inventions described in claims 7, 15 and 23 are characterized in that the relevance is individually increased or decreased in accordance with the use state of the acquired action target for each external information.

In the inventions described in claims 7, 15 and 23, the degree of relevance given to each piece of external information is individually increased or decreased according to the use state of the acquired action target for the external information. Therefore, when determining the action to be performed on the external information, by referring to the degree of relevance given to each of the external information, a plurality of external information related to the same basic action goal was detected at the same time. Also in this case, a single action for a single piece of external information having the highest frequency of appearance is determined.

According to the present invention, any one of a plurality of detected external information is selected according to the degree of importance and the degree of association, and a basic action target or a basic action target is selected for the selected external information. The action to be executed is determined based on the acquired action goal.

In the invention described in claims 8, 16 and 24, when a plurality of pieces of external information are detected simultaneously, a single piece of external information is referred to by referring to the importance and the relevance. The action to be performed is determined. Therefore, when deciding the action to be performed on the external information, the conscious choice that occurs in the real creature is simulated.

[0025]

FIG. 1 is a diagram showing a configuration of an artificial creature model to which a behavior generating device according to an embodiment of the present invention is applied. The artificial creature model 1 is a simulated creature such as a robot that acts in the real world via an actuator,
Or, it is a virtual creature that acts in a virtual world in a display constituting a computer system. The artificial creature model 1 includes an external information detection unit 2, a knowledge setting storage unit 3, a recognition evaluation unit 4, a state storage unit 5, an action planning unit 6, and an operation output unit 7. The following is an artificial biological model 1
1 The external information detection unit 2 Knowledge setting storage unit 2-1 Target storage unit 2-2 Experience knowledge storage unit 3 Recognition evaluation unit 3-1 Recognition unit 3-2 Consciousness level calculation unit 3 3 Evaluation unit 4 State storage unit 4-1 External state storage unit 4-2 Internal state storage unit 4-2-1 Extraction processing due to internal state 4-2-2 Update processing of internal state 5 Action planning unit 6 Action output unit 6-1 Walking motion and direction change 6-2 Facial expression synthesis 6-3 Voice synthesis 7 Processing procedure of artificial creature model

1 External Information Detecting Unit The external information detecting unit 2 detects external information as a stimulus from an external environment including other living things such as human beings. The external information detection unit 2 is configured by an input device such as a CCD camera, a microphone, a vibration sensor, or a mouse.

2. Knowledge Setting Storage Unit The knowledge setting storage unit 3 includes, as shown in FIG. 2, a goal storage unit 11 for storing innate action goals and an empirical knowledge storage unit 12 for storing acquired action goals. Have been.

2-1 Goal Storage Unit The goal storage unit 11 stores a goal inherently included in the artificial creature model 1 or an action goal that simulates a desire as an innate action goal which is a basic action goal of the present invention. I do. The innate behavior target may be set in advance when the artificial creature model 1 is manufactured, and may be initially set by the user who is the owner.

The innate behavioral goals stored in the goal storage unit 11 include an acquisition behavioral goal for the artificial creature model 1 to acquire a specific stimulus and an object, and an artificial behavioral model 1 for avoiding the specific stimulus and the object. There is an avoidance action goal to be attempted.
Examples of the innate behavioral goals are, for example, Goal 1: “Acquire a tactile stimulus that is stroked” Goal 2: “Acquire a light stimulus” Goal 3: “Avoid tactile stimuli that are hit” Goal 4: “Avoid food and drink Goal 5: Acquire the owner's joy Goal 6: Avoid the anger of the owner Goal 7: Avoid the sadness of the owner Goals 1, 2, 4, and 5 , And goals 3, 6, and 7 are avoidance goals. By setting the acquisition target and the avoidance target in this manner, the expression of joy or anger can be output to the artificial creature model 1 depending on whether the acquisition target is achieved or confused. The display of relief or disgust can be output to the artificial creature model 1 depending on whether it is done or disputed.

The goal storage unit 11 stores the importance of each of the innate action goals. This importance defines a priority order among a plurality of innate action goals, and is fixedly or variably set for each of the innate action goals. When changing the set value of the importance, for example, it is conceivable to change the value over time using a periodic function such as a sine function in which the input is time and the output is the importance.

As described above, by changing the importance indicating the priority of each innate action target over time, the action performed by the artificial creature model 1 with respect to the same stimulus changes in a time-series manner. It can reproduce the "freaky" that appears in the behavior of other living things.

2-2 Experience Knowledge Storage Unit The experience knowledge storage unit 12 stores the recognition result of the external information detected by the external information detection unit 2 and the target storage unit as shown in FIG. 11 is stored as an acquired behavior goal, which is the acquired behavior goal of the present invention.

The experience knowledge storage unit 12 stores the degree of association and the degree of storage for each piece of external information. The relevance is
It takes a value in the range of “0” to “1” and represents the relevance between each piece of external information and the innate action goal, and has a positive relevance and a negative relevance.

External information with a positive degree of association contributes to achievement of the innate action goal. For example, in FIG.
The color 1 is given a positive relevance of “0.7” to the innate action goal 1 of “acquiring a tactile stimulus to be stroked”, and the artificial creature model 1 changes the color 1 when stroked. Indicates that detection was performed. The positive relevance is increased by “0.1” when the same external information is recognized when the innate action goal to which each piece of external information belongs is achieved, and the innate action goal to which each piece of external information belongs is increased. If the same external information is not recognized when it is achieved, “0.1” is reduced.

External information to which a negative degree of relevance is attached confuses achievement of the innate action goal. For example, in FIG.
The sound 3 is given a negative relevance of “0.4” to the innate action target 3 of “avoiding the tactile stimulus being hit”, and the artificial creature model 1 generates the sound 3 when hit. Indicates that detection was performed. The negative relevance is increased by “0.1” when the same external information is recognized when the innate behavioral goal to which each piece of external information belongs is confused, and the innate behavioral goal to which each piece of external information belongs is intensified. If the same external information is not recognized at this time, the value is reduced by "0.1".

The degree of storage takes a value in the range of "0" to "1", and is used to determine when to delete the relationship between each external information and the innate action target. That is, the storage degree when the external information is newly stored in the experience knowledge storage unit 12 is “1”.
As time elapses, it is decreased, and as the degree of association is increased, it is increased. Then, the value of the memory degree is “0”
, The external information is deleted from the stored contents of the experience knowledge storage unit 12. This memory can be used to reproduce the "forgetting" that appears in the behavior of real life creatures. That is, the response to the external information having a large memory is accelerated, and the response to the external information having a small memory is delayed.
Further, when the storage degree becomes "0", the memory does not respond to the external information. By determining the storage period of the external information based on the degree of storage, the total number of external information to be stored can be reduced, and the storage capacity of the memory can be reduced.

The contents stored in the target storage unit 11 and the experience knowledge storage unit 12 are updated based on an evaluation output list output from the recognition evaluation unit 4 to the state storage unit 5 described later. As shown in FIG. 8, the evaluation output list includes columns for a goal, a state, an object, a direction, novelty, importance, a degree of association, an associated direction, and a degree of conformity.

In the goal column, any of the innate action goals stored in the goal storage unit 11 is described. The status column describes whether the innate action goal described in the goal column is in an achieved, confused, likely to achieve, or confused state. In the target column, the type of recognized single or plural external information is described. In the direction column, the coordinate value of the center of gravity of the external information on the imaging screen of the CCD camera constituting the external information detection unit 3 is described.

The novelty column describes whether or not the relationship between the innate action goal described in the goal column and the external information described in the target column is new. The importance column describes the importance of the innate action goal described in the goal column. The relevance column describes the relevance of the external information described in the target column. In the case of new external information, an intermediate value “0.5” is described. In the related direction column, "positive" is described when the goal is achieved, and "negative" is described when the goal is confused. The relevance column describes the relevance at the time of recognizing the external information described in the target column.

The evaluation output list shown in FIG. The experience knowledge storage unit 12 refers to the evaluation output list stored in the state storage unit 5 and, if the description content of the state column is “achieved” or “confused”, is described in the goal column. Update the memory contents belonging to the innate action goal.

That is, when the description content of the novelty column is "YES", the external information described in the target column is newly stored with "0.5" described in the relevance column. And
If the description content of the novelty column is “NO”, the relevance and storage of the external information described in the target column are set to “0.
In addition, among the external information belonging to the innate action goal described in the target column, the relevance of the external information not described in the target column is reduced by “0.1”.

3. Recognition Evaluation Unit The recognition evaluation unit 4 includes, as shown in FIG.
A recognizing unit 31 for recognizing the detected external information, for determining whether to ignore the external information based on the relevance of the recognized external information and the importance of the innate action target to which the external information belongs. Awareness level calculator 3 that calculates the consciousness level
2, and an evaluation unit 33 for evaluating whether or not each external information is conscious based on the calculated consciousness level.

3-1 Recognition Unit The recognition unit 31 recognizes the external information detected by the external information detection unit 2. The recognition unit 31 is, for example, a vibration recognition unit, a light recognition unit, a color recognition unit, an individual face recognition unit, an owner emotion recognition unit, a voice recognition unit, a mouse operation recognition, depending on the input device configuring the external information detection unit 2. And a virtual world recognition unit.

The vibration recognizing unit uses the input from the vibration sensor of the external information detecting unit 2 to recognize whether the owner strokes or hits the artificial creature model 1. That is, when the owner strokes the artificial creature model 1, the amplitude of the vibration detected by the vibration sensor is relatively small, and the number of signals per unit time increases. The owner is an artificial creature model 1
Is hit, the amplitude of the vibration detected by the vibration sensor is relatively large, and the number of signals per unit time decreases. The vibration recognition unit outputs, based on the amplitude data and the frequency data input from the vibration sensor, for example, by fuzzy inference using a fuzzy rule shown in FIG. 6, whether or not the owner is stroking or hitting. This external information is directly related to the innate behavioral goal, and when the stroke is recognized, the innate behavioral goal of "obtaining a tactile stimulus to be stroked" is achieved, and when it is struck, The innate action goal of "avoiding the tactile stimulus being hit" has been confused.

The light recognizing unit determines whether or not the average brightness of a pixel group composed of a predetermined number or more of pixels in the field of view of the CCD camera of the external information detecting unit 3 exceeds a threshold value. When the average lightness exceeds the threshold value in this determination, the light recognition unit outputs the number of pixels of the pixel group and the coordinate value of the pixel group, which are the degrees of matching of the recognition results. This external information is directly related to the innate behavioral goal, and when the light recognition unit recognizes the light, "acquire a light stimulus"
That is to say, the innate action goal has been achieved.

The color recognition section calculates a difference between a predetermined number of frames in a continuous image taken by the CCD camera of the external information detection section 3 when any of the innate action goals is achieved or confused. , A portion having a small difference is removed as a background. Also, the skin color portion in the image is removed by considering it as a human face or arm, and the remaining portion is determined as clothes worn by humans, and the color having the largest number of pixels in this portion is detected and detected. The intensity values of RGB, which are the three primary colors of the additive color mixture, are extracted for the extracted colors.

The color recognizing unit performs the RGB extraction from the color of the clothes.
From each intensity value of, for each of RGB
For example, a membership function, which is a triangular wave function shown in FIG. 7, is created and registered, and the fitness in the registered membership function is obtained for each of the RGB intensity values extracted from the newly detected color. When the minimum value of the three conformances is equal to or greater than a predetermined threshold, the conformity is output. When the minimum value is less than the threshold, “0” is output as the conformity. When a new membership function is created and registered, "1" is output as the matching degree. The color recognizing unit outputs the coordinates of the center of gravity of the image determined to be clothing, along with the degree of matching.

Note that the color recognizing unit sets a plurality of colors whose RGB intensity values are within the width Mw of each membership function to be similar colors, and shifts the central value Mc to form one set of membership values. Use functions. As a result, the number of membership functions to be registered can be reduced, and the required storage capacity can be reduced.

By the processing of the color recognizing unit, the artificial creature model 1 reacting to a specific color can be created based on the history of the detection result of the external information. For example, if a person wearing red clothing hits the artificial creature model 1, the innate action goal of "avoiding the tactile stimulus being hit" was confused, and a membership function for recognizing red was created. Is done. In addition, the external knowledge storage unit 12 stores red external information to which a negative association degree with respect to the innate action goal of “avoiding a tactile stimulus being hit” is added. Next, when the person wearing the red clothing approaches the artificial creature model 1, the red clothing is recognized as external information in the color recognition unit, and is given to the red for the innate action goal of "avoid tactile stimulus being hit". Based on the negative degree of association, the artificial creature model 1 can take an action such as expressing a fear expression or moving in a direction away from the creature.

In the case where a blue object interrupts between the artificial creature model 1 and the light source while the artificial creature model 1 is moving in the direction of the light source to receive the light stimulus, the light of the light source is interrupted. The innate behavioral goal of "obtaining stimulus" was confused by the blue object, and a membership function for recognizing blue color was created. Is stored in the empirical knowledge storage unit 12. Next, when blue is recognized as external information, the artificial creature model 1 generates an expression of anger based on the negative degree of association given to blue for the innate action goal of “acquiring light stimulation”. It can take actions such as showing and attacking.

The personal face recognizing unit registers or recognizes a human personal face present in the field of view of the CCD camera of the external information detecting unit 3 when the innate behavioral goal is achieved or confused. CC
As a method of extracting a personal face from an image captured by a D-camera, for example, “face recognition by computer” (Shigeru Akamatsu, Transactions of the Institute of Electronics, Information and Communication Engineers, D-II, Vol.J80-D-II, No.
8, pp.2031-2046,1997) and "Unlocking Control System by Face Image Matching" (Genki Doi, Ken Chen, Kunihiro Chihara et al., IEICE Transactions, D-II, Vol.J80-D-II) , No.8, pp.2203-220
8,1997) can be used.

The individual face recognizing unit outputs the degree of conformity of the recognition result, and outputs a degree of conformity “0” when no face matching the recognized face is registered. Also, the degree of conformity when newly registering is “1”. If the fitness exceeds a predetermined threshold, the coordinate value of the recognized individual face on the screen captured by the CCD camera is output.

By the processing of the individual face identification unit, the artificial creature model 1 that responds to a specific individual face can be created based on the history of the detection result of the external information. For example, when a person in the field of view of the CCD camera of the external information detection unit 3 strokes the artificial creature model 1, the innate action goal of "acquiring a stroked sensory stimulus" has been achieved, and the person The personal face is extracted and registered, and the personal face to which a positive degree of relevance to the innate action goal of “acquiring a stroked sensory stimulus” is added to the experience knowledge storage unit 1
2 is stored. Next, when the same individual face is extracted from the image captured by the CCD camera, based on the positive degree of association given to the individual face with respect to the innate action target of “obtaining a stroked sensory stimulus” The artificial creature model 1 can take an action such as showing an expression of joy or moving in a direction approaching the person.

The owner's emotion recognition unit recognizes the owner's emotion based on the facial expression of the owner and the tone of the voice, and outputs the recognized owner's emotion and fitness. By the processing of the owner's emotion recognition unit, "acquire the owner's joy",
It is possible to recognize whether innate behavioral goals such as "avoid owner's anger" and "avoid owner's sadness" have been achieved or distracted, and the artificial creature model 1
It is possible to recognize the effects of objects and stimuli existing near the owner on the emotions of the owner. For example, if the owner recognizes the joy of the owner when the specific object is present, and then recognizes the sadness of the owner, the artificial creature model 1 performs an action such as moving the object closer to the owner. Can be taken.

As a method of recognizing facial expressions, for example, “Recognition of facial expressions of a plurality of persons from a moving image by HMM using continuous output probability density distribution” (Naohiro Otsuka, Jun Ohtani,
Ryohei Nakatsu, IEICE Transactions, D-II, Vol.J80-DI
I, No. 8, pp. 2129-2137, 1997) and the like. Also, as a method of recognizing emotion from human voice,
For example, the configuration of an emotion recognition device disclosed in Japanese Patent Application Laid-Open No. 5-12023 can be used. Further, as a method of recognizing emotions from facial expressions and voices, for example,
"Examination of emotions of interlocutor using face image and voice"
(Hiroshi Sukekawa, Hirotoshi Iwano, Katsuhiko Shirai, IEICE Technical Report, PRU94-109, 1885) can be used.

The voice recognition unit recognizes and registers the voice input from the microphone of the external information detection unit 3 when the innate action goal is achieved or confused. As a speech recognition method, an LPC cepstrum coefficient obtained by linear prediction analysis can be used as a speech feature amount. For example,
The feature value when the twentieth-order linear prediction analysis is performed is a 20-dimensional vector. A membership function for these features is created in the same manner as the membership function in the above-described processing of the color recognition unit.

The voice recognition unit, like the color recognition unit described above,
Output the fitness after the threshold processing. Also, the matching degree of newly registered voice is “1”. This voice recognition unit is
A microphone is provided, and the direction of the sound source is recognized by determining the center position of the intensity of the sound collected by each microphone.

By the processing of the voice recognition unit, the artificial creature model 1 that responds to a specific voice can be created based on the history of the detection result of the external information. For example, if a person hits the artificial creature model 1 while uttering a specific sound, the innate action goal of "avoiding the tactile stimulus being hit" is confused, and the member for recognizing the sound is confused. A ship function is created, and a voice with a negative degree of association with the innate action goal of “avoiding a tactile stimulus being hit” is stored in the experience knowledge storage unit 12 as external information. Next, when the same voice is recognized, the artificial creature model 1 expresses the expression of anger because the voice has a negative association degree with respect to the innate action goal of “avoiding a tactile stimulus being hit”. It is possible to take an action such as indicating or moving away from the sound source.

The mouse operation recognition unit recognizes the operation of the mouse by the owner in an artificial creature model that behaves in a virtual world represented by an image on the display screen of the computer system. For example, an act of hitting the artificial creature model 1 by operating the mouse to move the cursor in the display to the display position of the artificial creature model 1 and clicking the mouse button can be expressed. Further, after the cursor in the display is moved to the display position of the artificial creature model 1, the act of stroking the artificial creature model 1 can be expressed by reciprocating the cursor a plurality of times while clicking the mouse button.

The mouse operation recognizing unit outputs the fitness of whether the artificial creature model 1 is stroked or hit by the operation of the mouse, and outputs the fitness “0” when there is no operation. For example, for the act of stroking, the number of reciprocating movements of the cursor is counted with the maximum value being 10 reciprocations, and the degree of conformity normalized in the range of “0” to “1” is output. For the act of tapping, the number of clicks of the mouse button is counted with the maximum value of 10 times,
The degree of conformity normalized in the range of “0” to “1” is output.

By the processing of the mouse operation recognition unit, it is possible to create the artificial creature model 1 that responds to mouse operation based on the history of the detection result of the external information. For example, if a tapping action is performed by operating a mouse, the innate action goal of "avoiding the tactile stimulus being hit" is confused, and the innate action of "avoiding the tactile stimulus being hit" is confused. The mouse operation provided with the negative degree of association with the target is stored in the experience knowledge storage unit 12.
Next, when the cursor moves in a direction approaching the artificial creature model 1, the artificial creature model 1 can take an action such as showing an expression of anger or moving in a direction away from the cursor.

The virtual world recognizing section recognizes an object in the virtual world formed in the screen of the display or the like in the computer system and outputs the degree of conformity of the shape of the object and the coordinate values of the object.

3-2 Consciousness Level Calculator The consciousness level calculator 32 simulates the concept of consciousness of a real creature in the artificial creature model 1. That is, a real creature does not behave consciously of all stimuli and objects input from sensory organs, and does not pay attention to objects not directly related to itself. This means that in real life, the processing speed of the brain and the capacity of short-term memory are limited, and if attention is paid to all objects, processing of important objects such as life-threatening will be delayed and endangered This is because there are cases.

In the artificial creature model 1 as well, there is a limit in the calculation processing speed and the storage capacity. If processing is performed for all the external information recognized by the recognition unit, it takes a long time to take an action as a response to a stimulus. It may take time, and it may not be possible to determine which of the processes for a plurality of targets has priority. Therefore, the consciousness level calculation unit calculates how much each of the plurality of external information is related to the action goal of the artificial creature model 1 and determines the priority order for each of the plurality of external information, thereby obtaining the reaction behavior. In addition to shortening the time taken to take a deadlock, a deadlock is prevented.

The consciousness level calculation unit 32 calculates the consciousness level of each recognized external information based on the plurality of recognition results input from the recognition unit and the contents stored in the experience knowledge storage unit 12 as follows: Is calculated by

(Consciousness level) = (degree of conformity) × (degree of association)
× (memory) × (importance) Here, the relevance, the relevance, the memory, and the importance are all determined in the above-described processing.

The consciousness level of the virtual object is
It is calculated by the following equation.

(Consciousness level) = (distance from artificial creature model to virtual object) × (relevance) × (memory) × (importance) where new stimuli not stored in experience knowledge storage unit 12 ( The consciousness level for external information or a virtual object) is “1”. In addition, the degree of association and the degree of storage of the stimulus directly related to the innate behavioral goal stored in the goal storage unit 11 are “1”. Further, when one stimulus belongs to a plurality of innate action goals, the consciousness level is calculated for each of the innate action goals, and the maximum value of the calculated consciousness level is set as the consciousness level of the stimulus.

The consciousness level calculator 32 performs threshold processing on the consciousness levels of all recognized stimuli,
A stimulus having a consciousness level equal to or higher than a predetermined threshold is output as a conscious object.

3-3 Evaluation Unit The evaluation unit 33 evaluates which innate action goal the conscious object output from the consciousness level calculation unit 32 relates to, and evaluates the evaluation result as shown in FIG. Output as output list.

There are three types of output methods of the evaluation output list depending on the conscious object. That is, when the conscious target is a target directly related to the innate behavioral goal, an already registered target, and an unregistered target, the output method of the evaluation output list is different. different.

When the conscious object is an object directly related to the innate action goal, the number of the innate action goal to which the conscious object is directly related, And the type of avoidance target, and the status column describes the type of achievement or confused. In the target column, a target directly related to the innate behavioral target is described. If the artificial creature model 1 is a robot existing in the real world, vibration 1 (stroke),
Either vibration 2 (slap), light stimulation or user emotion is described, and in the case of a virtual creature existing in the virtual world, food and drink are further included. In the direction column, the coordinate value of the light source when the description content of the target column is a light stimulus, and in the case of a virtual creature, the coordinate value of the object is described. "NO" in the novelty column
However, in the importance column, the importance corresponding to the content described in the target column is described. Also, "1" is displayed in the relevance column.
However, in the related direction column, the positive /
In the negative column, the relevance column describes the relevance of the recognition result.

When the conscious object is an object already registered in the experience knowledge storage unit 12 and is an object directly related to the target, the consciousness target is displayed in the target column of the evaluation output list. The number of the innate goal that the subject is directly related to,
In addition, the type of whether the target is the acquisition target or the avoidance target is described, and the type of whether the target is achieved or confused is described in the state column. If the target is not directly related to the target, the target column indicates the number of the innate action target to which the target belongs, and the type of whether the target is an acquisition target or an avoidance target, and the status column indicates whether the degree of association is positive. / "It seems to be achieved" according to negative
/ "Less likely to be confused" is described respectively. If the registered conscious object belongs to more than one congenital action goal, the most relevant congenital action goal is described, and the relevance for multiple congenital action goals is also equal. Describes the innate action goal with the greatest memory, and if the memories of a plurality of innate action goals are equal, the innate action goal with the smaller goal number is described.

As an example, when there is a conscious object directly related to the innate behavioral goal (for example, when struck) and when another conscious object is color 1 (see FIG. 5) In this case, referring to FIG. 5, there is already the knowledge that the color 1 is related to the innate action target 1 (obtains a tactile stimulus to be stroked). Of the evaluation output list, “3 (avoidance)” is displayed in the target column, “confused” in the status column, and “color 1” in the target column of the evaluation output list. ], The barycentric coordinate value of color 1 in the direction column,
"YES" is shown in the novelty column, the importance of the innate action goal 3 is shown in the importance column, the relevance "0.5" of the new external information is shown in the relevance column, and "relevance" is shown in the "relation direction" column. “Negative” is described in the relevance column as the relevance in the recognition result for color 1.

On the other hand, when there is no conscious object directly related to the innate action goal and only the color 1 exists as the conscious object, referring to FIG.
(Acquisition of a tactile stimulus to be stroked) has a positive relevance, so that recognizing color 1 is interpreted as "achieving congenital action goal 1". Based on this interpretation, “1 (acquisition)” is displayed in the target column of the evaluation output list, “achieved” in the status column, “color 1” in the target column, and color 1 in the direction column.
Of the center of gravity, "NO" in the novelty column, the importance of the innate action goal 1 (0.8 in the example of FIG. 5) in the importance column,
In the relevance column, the relevance of the new external information (0.8 in the example of FIG. 5), “positive” in the relevance direction column, and the relevance in the recognition result of color 1 in the relevance column, Each is described.

When the conscious object is an unregistered object and there is an innate action goal directly related to the conscious object, the target column of the evaluation output list contains the conscious object. The number and type of the innate action goal directly related to the target are described in the status column, and the status of achievement / conflict is described in the status column. In the target column, the type and reference number of the conscious target recognized this time are described. For example, if the color 5 is already registered, the color 6 is described. In the direction column, the coordinate value of the conscious object recognized this time, and in the novelty column, “YES”
However, in the importance column, the importance of the innate action goal described in the goal column is shown, and in the relevance column, the relevance “0.
"5", "positive" / "negative" in the relevant direction column according to the achievement / conflict described in the status column, and the relevance column in the recognition result of the conscious object recognized this time in the relevance column. Are described respectively.

If the conscious object is an unregistered object and there is no innate action goal directly related to the conscious object, the evaluation output list is not created.

When the artificial creature model 1 exists in the virtual world, a virtual world evaluation unit is provided together with the evaluation unit 33. The virtual world evaluation unit evaluates the state of the virtual world and
Create the evaluation output list shown in. Here, the state of the virtual world means the state of the virtual object existing in the virtual world and the behavior of the artificial creature model of the virtual world. For example, “the artificial creature model B has given food to itself” ( Example 1), "Artificial creature C hit himself" (Example 2), "Artificial creature C approaches himself" (Example 3), and the like.

The evaluation output list created by the virtual world evaluation unit in the cases of Examples 1 to 3 will be described below.

In the case of Example 1, upon recognizing that “artificial creature model B has given food to itself”, the innate action target 4 of “obtaining food and drink” is obtained.
Is determined, “4 (acquired)” is displayed in the target column of the evaluation output list, “achieved” in the status column, “artificial creature model B” is displayed in the target column, and artificial The coordinate value of the biological model B, "YES" or "NO" in the novelty column according to whether or not the artificial biological model B has not been registered, and the innate behavior which is the description content of the target column in the importance column. The importance of the goal 4 is set to “0.5” according to “YES” / “NO” in the novelty column in the relevance column, or the relevance stored in the experience knowledge storage unit 12 is set in the relevant direction column. Is described as "correct", and "1" is described in the fitness column.

In the case of Example 2, upon recognizing that "artificial creature C hit itself", it is determined that the innate action goal 3 of "avoid touching tactile stimulation" has been confused, and the target column of the evaluation output list "3
(Avoidance) "," confused "in the status column," artificial creature model C "in the target column, and artificial creature model C in the direction column.
In the novelty column, “YES” or “NO” is displayed in the novelty column according to whether or not the artificial creature model C is not registered. In the importance column, the innate action target 3 which is the description content of the target column is displayed. The importance is
In the association degree column, "0.5" according to "YES" / "NO" in the novelty column or the association degree stored in the experience knowledge storage unit 12, and in the association direction column, "negative" matches. "1" in the degree column
Are described respectively.

When Example 3 Occurs After Example 2 After recognizing that "artificial creature C hit yourself" after recognizing that "artificial creature C hit yourself", the target column of the evaluation output list is displayed. Is "3 (avoiding)", "Confuses" in the status column, and "artificial creature model C" in the target column.
In the direction column, the coordinate value of the artificial creature model C, in the novelty column, “NO”, in the importance column, the importance of the innate action goal 3, and in the relevance column, is stored the experience knowledge storage unit 12. Is described in the related direction column, and "1" is described in the matching degree column. Based on the evaluation output list described in this manner, the artificial creature model 1 takes an anxiety that the artificial creature model C will be hit by the artificial creature model C, or moves in a direction away from the artificial creature model C. .

4. State Storage Unit The state storage unit 5, as shown in FIG.
1 and an internal state storage unit 42.

4-1 External State Storage Unit The external state storage unit 41 temporarily stores the evaluation output list created by the recognition evaluation unit 4. The evaluation output list stored in the external state storage unit 41 is stored in the internal state storage unit 4.
2. Knowledge setting storage unit 3, action planning unit 6, and operation output unit 7.
Referenced by

4-2 Internal State Storage Unit The internal state storage unit 42 simulates, in the artificial creature model 1, emotions corresponding to the emotions and moods of human beings, which are real creatures. That is, emotions such as joy, anger, sadness, disgust, and fear are reproduced. Here, emotion refers to a short-term mental state at a high alertness level, and mood refers to a long-term mental state at a low alertness level. The generation of emotions
It changes depending on the situation where the human being is. For example,
According to Theory of Computational Theory (by Akira Fumi), emotions of pleasure occur when goals are achieved, and emotions of sadness occur when goals are confounded. Other emotions also occur and change depending on the state of goals, preferences, and other factors. Mood persists at low levels of alertness with joy and sadness.

Based on this idea, the internal state storage unit 42 determines the state of the emotion of the artificial creature model 1 according to the contents of the evaluation output list output from the recognition evaluation unit.
It is updated and stored by the extraction process due to the internal state and the update process of the internal state.

4-2-1 Extraction Process Due to Internal State As a cause for updating the internal state, the degree of achievement of the target,
Set the degree of dispute. In addition, the degree of expectation when the goal is likely to be achieved and the degree of anxiety when the goal is likely to be confused are set. The degree of achievement, degree of dispute, degree of expectation, and degree of anxiety are determined using fuzzy inference. FIG. 9 shows an example of a fuzzy rule used for the fuzzy inference. As for the degree of expectation and the degree of anxiety, a value one step before is used as an input of fuzzy inference. Individuality can be given to the artificial creature model by individually setting parameters such as fuzzy rules and membership functions used for fuzzy inference.

4-2-2 Internal State Update Process Since the internal state is to simulate the emotions and moods of human beings, the amount of increase in the emotional intensity is calculated by fuzzy inference. FIG. 10 shows an example of fuzzy rules used for fuzzy inference. By individually setting parameters such as fuzzy rules and membership functions, individuality can be given to the artificial creature model.

Each emotion intensity Ei can be calculated by the following equation.

[0090]

(Equation 1)

The attenuation amount γ starts to increase in the positive direction when the emotion intensity becomes a positive value from “0”, and returns to “0” when the emotion intensity becomes “0” or less. When the emotion intensity is equal to or higher than a predetermined threshold, the emotion and intensity having the maximum emotion intensity are output to the action planning unit 22 and the operation output unit 6.

By storing the calculated emotion intensity as a long-term internal state until the value becomes “0” due to the attenuation γ, the feeling of human mood as a real creature is simulated in the artificial creature model 1. Reproduce it. In addition,
By individually setting the degree of influence Wj, individuality can be given to the artificial creature model.

5 Action Planning Unit The action planning unit 6 plans an action to be executed by the artificial creature model 1 based on the contents stored in the knowledge setting storage unit 3, the evaluation output list, and the emotion. When planning an action, for example, the action rules shown in FIG. 11 are referred to. If a plurality of behavior rules apply, the behavior rule for the goal with the highest importance is selected. The output of the action plan unit 6 includes an action name such as “approach”, “away”, “intimidating”, and the action target.

6 Action Output Unit The action output unit 7 outputs the action of the artificial creature model 1 based on the output of the action planning unit 6 and the evaluation output list and emotion stored in the state storage unit 5. The actions output by the action output unit 7 include, for example, a walking action, a change of direction, synthesis of facial expressions, and voice synthesis.

6-1 Walking Motion and Direction Change When the artificial creature model 1 is a simulated creature existing in the real world, the walking motion and the direction change are realized by, for example, an actuator using a servomotor. The traveling direction of the walking motion is determined based on the output of the behavior planning unit 6 and the coordinate values of the evaluation output list. For example, when the action plan is “to approach the target” and the target is on the left side of the field of view of the CCD camera, the traveling direction is changed to the left side, and
When it is on the right side of the visual field of the CD camera, the artificial creature model 1 can be brought closer to the target by changing the traveling direction to the right and proceeding. When the action plan is “escape from the target”, the traveling direction is changed to the right when the target is on the left side of the field of view of the CCD camera, and the traveling direction is set when the target is on the right side of the field of view of the CCD camera. Is changed to the left, and the artificial creature model 1 can be kept away from the target.

When the target is a sound source, the artificial creature model 1
Determines the traveling direction based on information on whether the sound source is present on the left or right side of the current direction. When the action plan is “threatening”, the face expression and voice of “anger” are output while repeating forward and backward movements in the direction of the target.

The direction change can be realized by changing the outputs of the left and right servomotors. That is, when turning to the left, the output of the right servomotor is made larger than the output of the left servomotor, and when turning to the right, the output of the left servomotor is made higher than the output of the right servomotor. Also increase.

When the artificial creature model 1 is a virtual creature existing in a virtual world, the action is expressed using, for example, computer animation software and displayed on a display screen.

6-2 Face Expression Synthesis Face expressions are synthesized based on the output from the action planning section 6 and the emotions stored in the state storage section 5. Also, the degree of facial expression to be synthesized is adjusted according to the emotion intensity. For example, when the action plan is “intimidating”, the facial expression of anger is synthesized, and when the emotion intensity of “joy” exceeds a predetermined threshold in the state storage unit 5, the facial expression of joy is synthesized.

When the artificial creature model 1 is a simulated creature existing in the real world, for example, a “face robot” (Fumio Hara, Hiroshi Kobayashi, Akira Tange, Hiroki Endo, Japan Society of Mechanical Engineers 1st JS)
ME Robotome Symposia Proceedings, pp.77-84, 1996)
Thus, a facial expression is created by the operation of the actuator.

When the artificial creature model 1 is a virtual creature existing in the virtual world, for example, face part images such as eyes, mouths, eyebrows, etc. are prepared according to the emotion and the emotion intensity, and the computer animation is prepared. Synthesized using software,
Display on the display screen.

6-3 Speech Synthesis Speech is synthesized based on the output from the action planning unit 6 and the emotion stored in the state storage unit 5. In addition, the volume of the synthesized voice is adjusted according to the emotion intensity. For example, when the action plan is “escape”, a scream is synthesized, and when the emotion intensity of “joy” stored in the state storage unit 5 exceeds a predetermined threshold, a voice of joy is synthesized. The synthesized voice is output via a speaker.

7. Processing Procedure of Artificial Organism Model The processing procedure of the artificial creature model 1 configured as described above is shown in the flowchart of FIG. Artificial creature model 1
Until the external information detection unit 2 detects the external information, the movement and the stop are randomly repeated by the actions of the action planning unit 6 and the operation output unit 7 (s1, s2). When the external information detection unit 2 detects the external information, the recognition unit 31 of the recognition evaluation unit 4
Performs the process of recognizing the external information detected by the external information detection unit 2 (s3), and calculates the consciousness level in the consciousness level calculation unit 32 for the recognized external information (s3).
4).

Next, of the recognized external information, external information whose consciousness level is equal to or higher than a predetermined threshold is extracted as a conscious object by the evaluation unit 33 (s5), and the extracted conscious object is evaluated by the evaluation unit 33. An output list is created (s6). Further, based on the created evaluation output list, the internal state stored in the internal state storage unit 42 is updated (s7), and the content stored in the experience knowledge storage unit 12 of the knowledge setting storage unit 3 is updated (s8).

Thereafter, an action plan is created in the action plan section 6 based on the evaluation output list and the contents stored in the experience knowledge storage section 12 (s9), and the action is output by the action output section 7 based on the created action plan. Output (s10). The artificial creature model 1 repeatedly executes the processing of s1 to s10 while the power is on.

By the above processing, the artificial creature model 1 according to this embodiment simulates the concepts of learning, forgetting, and consciousness used in the process of generating actions in a real creature, and autonomously executes the external information. And perform behaviors that closely resemble those of real living things.

In the artificial creature model 1 described above, the knowledge setting storage unit 3, the recognition evaluation unit 4, the state storage unit 5, and the action planning unit 6 can be configured in a microcomputer having a memory. Can be supplied to the microcomputer as a processing program recorded in a recording medium such as a CD-ROM.

[0108]

According to the first, ninth, and seventeenth aspects of the present invention, a new action goal is learned as an acquired action goal from the relationship between the detected external information and a predetermined basic action goal. By determining the action to be performed on external information based on the action goal including the acquired action goal, the action goal for determining the action to be performed on the external information can be supplemented by learning. In addition, it is possible to autonomously determine an action to be performed for unforeseeable external information, and to reduce the number of action targets to be set in advance.

According to the second, tenth and eighteenth aspects of the present invention, a part of the newly acquired action goal is erased by individually erasing the new action goal acquired by learning based on the memory level. Without excessively storing an action goal for determining an action to be performed with respect to external information, thereby preventing the action determination process from becoming complicated and reducing the storage capacity of the action goal. Can be.

According to the third, nineteenth, and nineteenth aspects of the present invention, the actions to be performed on the external information can be determined by individually deleting the acquired action targets in accordance with the degree of memory that decreases with time. When deciding, it is possible to reproduce a forgetting phenomenon that occurs in a real creature, and to generate an action similar to a real creature.

According to the fourth, twelfth and twentieth aspects of the present invention, the action to be performed on the external information can be determined by individually erasing the acquired action targets in accordance with the degree of memory decreasing according to the use state. At the time of the determination, the forgetting phenomenon that occurs in a real creature can be reproduced more faithfully according to the appearance frequency of each piece of external information, and an action closer to the real creature can be generated.

According to the fifth, thirteenth, and twenty-first aspects of the present invention, when deciding an action to be performed on external information, the importance assigned to each of the plurality of basic action goals is referred to. By this, even when the detected external information is related to a plurality of basic action goals, a single action can be determined for the external information, and a deadlock state due to a plurality of conflicting actions determined. Can be reliably prevented.

According to the invention described in claims 6, 14 and 22, when deciding an action to be performed on external information, the relevance assigned to each of the external information is referred to to determine the same action. Even when multiple external information related to the basic action goal of the above is detected at the same time, it is possible to determine a single action for a single external information, and deadlock due to the determination of multiple conflicting actions The state can be reliably prevented.

According to the seventh, fifteenth, and twenty-third aspects of the present invention, when determining an action to be performed on external information, the relevance assigned to each of the external information is referred to to determine the same. Even if multiple external information items related to the basic action goal are detected at the same time, a single action for the single external information item with the highest frequency of appearance can be determined, and multiple conflicting actions are determined. As a result, a deadlock state due to this can be reliably prevented.

According to the invention described in claims 8, 16 and 24, when a plurality of pieces of external information are detected simultaneously, a single piece of external information is referred to by referring to the importance and the relevance. By deciding the action to be performed in response to external information, it is possible to simulate conscious choices that occur in the real creature when deciding the action to be performed for external information, and to approximate the real creature. Action can be generated.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of an artificial creature model to which a behavior generation device according to an embodiment of the present invention is applied.

FIG. 2 is a diagram showing a configuration of a knowledge setting storage unit of the artificial creature model.

FIG. 3 is a diagram showing a configuration of a recognition evaluation unit of the artificial creature model.

FIG. 4 is a diagram showing a configuration of a state storage unit of the artificial creature model.

FIG. 5 is a diagram showing storage contents of the knowledge setting storage unit.

FIG. 6 is a diagram showing a membership function used for fuzzy inference for vibration recognition in the recognition evaluation unit.

FIG. 7 is a diagram showing a membership function used for color recognition in the recognition evaluation unit.

FIG. 8 is a diagram showing an evaluation output list created by the recognition evaluation unit.

FIG. 9 is a diagram illustrating a fuzzy rule used in an internal state-based extraction process in an internal state storage unit of the state storage unit.

FIG. 10 is a diagram showing a fuzzy rule used for an internal state update process in the internal state storage unit of the state storage unit.

FIG. 11 is a diagram showing an action rule used for determining an action plan in an action planning unit of the artificial creature model.

FIG. 12 is a flowchart showing a processing procedure in the artificial creature model.

[Explanation of symbols]

 1-artificial creature model 2-external information detection unit 3-knowledge setting storage unit 4-recognition evaluation unit 5-state storage unit 6-action planning unit 7-action output unit 11-target storage unit 12-experience knowledge storage unit 31- Recognition unit 32-Awareness level calculation unit 33-Evaluation unit 41-External state storage unit 42-Internal state storage unit

Claims (24)

[Claims] 1. An external information detecting section for detecting external information, a basic action target storing section for storing a plurality of predetermined basic action goals, and an external information detected by the external information detecting section and a basic action target storing section. An acquisition action target storage unit that stores, as an acquisition action goal, a relationship with each basic action goal stored by
Provide an action to be performed on the external information detected by the external information detection unit based on the basic action goal stored in the basic action goal storage unit or the acquisition action goal stored in the acquisition action goal storage unit. An action generation device characterized by determining. 2. The behavior generation device according to claim 1, wherein the acquisition behavior target storage unit stores, for each of the plurality of pieces of external information, a storage degree indicating a period in which the acquisition behavior target is to be stored. 3. The behavior generating apparatus according to claim 2, wherein the acquired behavior target storage unit individually decreases each of the degrees of storage with time. 4. The behavior generating apparatus according to claim 2, wherein the acquisition behavior target storage unit individually increases or decreases each of the degrees of storage according to a use state of the acquisition behavior target for each piece of external information. 5. The action generation device according to claim 1, wherein the basic action goal storage unit stores importance indicating the priority of each of the plurality of basic action goals. 6. The behavior generating apparatus according to claim 1, wherein the acquired behavior target storage unit stores, for each of a plurality of pieces of external information, a degree of association with a basic behavior goal. 7. The behavior generating apparatus according to claim 6, wherein the acquisition behavior target storage unit individually increases or decreases each of the relevance in accordance with a use state of the acquisition behavior target for each piece of external information. 8. At least one of a plurality of pieces of external information detected by the external information detection unit according to the importance stored in the basic action target storage unit and the relevance stored in the acquisition action target storage unit. The action generation device according to claim 6 or 7, wherein the action to be executed is determined based on the selected external information and the basic action goal or the acquired action goal. 9. Detecting external information, storing a relationship between the detected external information and each of a plurality of predetermined basic action goals as an acquisition action goal, and storing the external information based on the basic action goal or the acquisition action goal. An action generation method characterized by determining an action to be performed on a program. 10. A storage degree indicating a period in which an acquisition action target is to be stored for each of a plurality of external information, and each acquisition action target is individually erased based on the storage degree.
The behavior generation method according to the above. 11. The action generation method according to claim 10, wherein each of the degrees of memory is individually reduced with time. 12. The action generation method according to claim 10, wherein each of the storage degrees is individually increased or decreased according to the use state of the acquired action target for each piece of external information. 13. The action generation method according to claim 9, wherein importance levels indicating priorities of the plurality of basic action goals are stored. 14. The action generation method according to claim 9, wherein the degree of association with the basic action goal is stored for each of the plurality of external information. 15. The action generation method according to claim 14, wherein each of the degrees of association is individually increased or decreased according to the use state of the acquired action target for each piece of external information. 16. A method for selecting one of a plurality of detected external information according to the importance and the relevance, and determining an action to be executed for the selected external information based on a basic action goal or an acquisition action goal. Claim 14 or 1 to be determined
6. The behavior generating method according to 5. 17. A process for detecting external information, a process for storing a relationship between the detected external information and each of a plurality of predetermined basic behavior goals as an acquisition behavior goal, and And a process of determining an action to be performed on the external information based on the action recording program. 18. A program comprising: a process of storing a storage degree indicating a period in which an acquisition action target is to be stored for each of a plurality of external information; and a process of individually deleting each acquisition action target based on the storage degree. The action generation program recording medium according to claim 17, wherein 19. The action generation program recording medium according to claim 18, wherein a program including a process of individually decreasing each of the storage degrees with the passage of time is recorded. 20. An action generation program recording medium according to claim 18, wherein a program including a process for individually increasing or decreasing each of the storage degrees in accordance with the use state of the acquired action target for each piece of external information is recorded. 21. The action generation program recording medium according to claim 17, wherein a program including a process of storing importance indicating respective priorities of a plurality of basic action goals is recorded. 22. An action generation program recording medium according to claim 17, wherein a program including a process of storing a degree of association with a basic action goal for each of a plurality of pieces of external information is recorded. 23. The action generation program recording medium according to claim 22, wherein a program including a process of individually increasing or decreasing each of the relevance in accordance with the use state of the acquired action target for each piece of external information is recorded. 24. A process for selecting one of a plurality of detected external information according to the importance and the relevance, and executing the selected external information based on a basic action goal or an acquisition action goal. Processing to determine an action;
The action generating program recording medium according to claim 22, wherein the program includes a program including the following.