KR100578071B1 - Emotion management and synthesis device and method for personal robot - Google Patents

Abstract

The present invention synthesizes its emotional state and desire state so that the personal robot shows emotional characteristics similar to humans based on various factors (stimulation, events, etc.) inside and outside for human-centered emotional human robot interaction. An emotional synthesizing apparatus for a personal robot capable of (update and manage) and a method thereof.

According to the present invention, there is provided an emotion value changing means for reducing an emotion value using a decay function if there is no factor affecting emotion, and increasing an emotion value using an influence function if there is an influence factor; After receiving the emotion values and reflecting the degree of mutual suppression and stimulation between each emotion, and receiving the reflected value and the current mood value of the robot, the Michaelis-Menten function is moved in parallel to the axis of the emotion values to activate each emotion. Emotion value updating means for updating the emotion value by adjusting the degree; And mood value updating means for receiving the updated emotion value and updating the mood value to affect the current mood in inverse proportion to the current time. There is provided a emotion manager of a robot comprising a.

Personal Robot, Emotion Recognition, Bayesian Learning, Need Management, Emotion Management

Description

Emotion manager and method of personal robot, emotional synthesis device and method using same {Emotion management and synthesis device and method for personal robot}             

1 is a conceptual block diagram of a personal robot emotion synthesizing apparatus according to an embodiment of the present invention,

FIG. 2 is a flow chart showing processing occurring in the event receiver 110 shown in FIG. 1,

3 is a flow diagram illustrating processing that occurs in the event handler 120 shown in FIG. 1,

FIG. 4 is a flow diagram illustrating processing that occurs in multiple desire managers 130 shown in FIG. 1,

FIG. 5 is a flowchart illustrating processing occurring in the plurality of emotion managers 140 shown in FIG. 1,

6 is a graph showing a result of the non-linear function of the emotional variable converted by the value of the mood variable according to an embodiment of the present invention,

7A is a conceptual diagram illustrating a case where a Bayesian learning method used by an embodiment of the present invention is applied to emotional recognition (recognition part) of the surrounding environment,

7B is a conceptual diagram illustrating a case where a Bayesian learning method used by an embodiment of the present invention is applied to emotional recognition (applied part) of the surrounding environment,

FIG. 8 is a graph showing a learning result of probabilistic models of six emotional responses of a user arbitrarily synthesized to two environment variables using a modified Bayesian learning method according to an embodiment of the present invention.

The present invention relates to an emotion manager and method of a personal robot, and to an apparatus and method for synthesizing emotions using the same, in particular, for the interaction of a human-centered emotional human robot, various factors (stimulation, event, etc.) The present invention provides a device and a method for synthesizing (updating and managing) an emotional state and a desire state in order to exhibit emotional characteristics similar to those of a human being.

Until now, patents and researches on emotion synthesis of personal robots have not been conducted.

Considering that non-verbal communication accounts for 65% of human interactions and 55% of facial expressions in the medium of message delivery, robots can be self-contained for natural human-centered interaction. It is very important to synthesize and express the emotions of people.

Conventional research on emotional synthesis of personal robots has been conducted mainly in the United States and Japan, such as Waseda University, OMRON, SONY, and MIT University, and most of the research results are based on the existing cognitive science, neurophysiology, and psychology. Based on the research results of human emotion system, we propose a conceptual composition of similar emotions.

However, these prior arts have a problem in that it is difficult to utilize them in general and general personal robot's emotional synthesis because the emotional state algorithm is not specifically presented, and the emotional synthesis method is designed mainly for a specific robot.

An object of the present invention for solving the problems of the prior art as described above is based on the object-oriented programming method and message exchange method in order to enable the robot emotion synthesis software can be easily integrated into the software platform of the general personal robot with extensibility It is to provide a robot emotion synthesizing apparatus and a method thereof.

In order to achieve the above object, according to the present invention, in the emotion manager of the robot for updating the value of the emotional variables based on the input event, the robot's current desire state and the mood of the robot, the input Emotion value changing means for reducing the emotion value by using a decay function if the values do not have a factor affecting emotion, and increasing the emotion value by using an influence function if there is an influence factor; After receiving the changed emotion value and reflecting the degree of mutual suppression and stimulation between each emotion, and receiving the reflected value and the current mood value of the robot, the Michaelis-Menten function is moved in parallel with the axis of the emotion value to determine Emotion value updating means for updating an emotion value by adjusting an activation degree; And mood value updating means for receiving the updated emotion value and updating the mood value to affect the current mood in inverse proportion to the current time. There is provided a emotion manager of a robot comprising a.

In addition, the pre-processor for the emotional recognition of the surrounding environment so that the robot reacts to the surrounding environment similarly to the user; A desire manager which receives the results of the preprocessor and updates a value of a corresponding desire variable; An emotion manager that receives the input event, the desire variable updated by the desire manager and the current mood of the robot, and updates the values of the emotion variables; And a state output unit that considers the emotion represented by the emotion variable having the highest value among the emotion variables updated by the emotion manager as the emotion state of the robot. The emotion manager may reduce the emotion value by using a decay function if the input values do not affect the emotion, and use an influence function if there is an influence factor. Emotion value changing means for increasing the emotion value; After receiving the changed emotion value and reflecting the degree of mutual suppression and stimulation between each emotion, and receiving the reflected value and the current mood value of the robot, the Michaelis-Menten function is moved in parallel with the axis of the emotion value to determine Emotion value updating means for updating an emotion value by adjusting an activation degree; And mood value updating means for receiving the updated emotion value and updating the mood value to affect the current mood in inverse proportion to the current time. Emotional synthesis apparatus of a robot is provided comprising a.

In addition, the preprocessor preprocesses the modified Bayesian learning method.

In addition, there is provided an emotion management method and an emotion synthesis method of the robot which go through the above process.

Hereinafter, with reference to the accompanying drawings will be described in more detail the emotion synthesis apparatus and method of the personal robot according to an embodiment of the present invention.

The present invention proposes a robot emotion synthesis program structure based on an object-oriented programming method and a message exchange method so that robot emotion synthesis software can be easily integrated into a general personal robot software platform. In addition, the present invention proposes a specific update algorithm for the emotional and desire variables that make up the emotional state. In addition, we propose a method of synthesizing the emotional state so that the emotional response similar to the user with respect to the surrounding environment state using the modified Bayesian learning method.

1 is a conceptual block diagram of a personal robot emotion synthesizing apparatus according to an embodiment of the present invention, the personal robot emotional synthesizing apparatus includes a preprocessor 100, an event receiver 110, an event processor 120, and a plurality of desires. It is configured to include a manager 130, a plurality of emotion manager 140 and the state processor 150.

The preprocessor 100 performs emotional recognition on the surrounding environment by the robot to preprocess it mathematically into an event, and the event receiver 110 receives an event (stimulus) preprocessed by the preprocessor 100. In addition, the event processor 120 transmits the input events at a predetermined time interval to each corresponding desire manager 130 and emotion manager 140.

The plurality of desire managers 130 updates a value of a corresponding desire variable based on the events transmitted from the event processor 120, and the plurality of emotion managers 140 transmits an event transmitted from the event processor 120. After updating the values of the emotion variables based on the current desire state of the robot and the mood of the robot reflected by the updated needs variable, the robot returns to the event processor 120 and performs the corresponding emotion processing.

The state output unit 150 serves as an output unit that reflects the final result of the event processor 120.

In the present specification, the processing occurring in the preprocessor 100 will be described later.

FIG. 2 is a flowchart showing processing occurring in the event receiver 110 shown in FIG. 1. First, in step S201, the event receiver 110 receives an event from the preprocessor 100 and then steps. In S202, it is stored in the Input_Event Queue. In this case, the event is defined by a name and a number of parameters, which must be predefined in the program.

For example, the parameter "FACE, owner, happy" is an event sent from the face / expression recognition module and is sent when the owner recognizes that he is smiling.

FIG. 3 is a flowchart showing processing occurring in the event handler 120 shown in FIG. 1, which will be described in detail as follows.

First, in step S301, information stored in the Input_Event_Queue stored by the event receiver 110 at a predetermined time interval is stored in an Event_List, and then in step S302, each desire manager corresponding to events based on the events stored in the Event_List. The value of the desires is updated through 130.

Then, in step S303, the values of emotions are updated through the corresponding emotion manager 140 based on the updated desire values, the current mood value, and the events of Event_List.

In operation S304, after receiving result values of the plurality of desire managers 130 and emotion managers 140, the mood values are updated based on recent emotion values, and then, in step S305, all the information of Event_List is deleted. , Rest for a certain period of time.

Meanwhile, in the present embodiment, the event value processor 120 is designed to be in charge of the event processor 120, but this is merely for convenience and does not exclude processing by the emotion manager 140. Therefore, when describing the operation of the emotion manager 140 to be described later, it will be described in detail including this.

FIG. 4 is a flowchart illustrating processing occurring in the plurality of desire managers 130 shown in FIG. 1, wherein the plurality of desire managers 130 values values of corresponding desire variables based on the events transmitted from the event handler 120. Update to bar, which will be described in detail as follows.

Since the desire manager 130 is designed to be defined separately for each desire, a robot developer can easily add / delete / modify as needed.

On the other hand, according to the study of Lorenz and Tinbergen, if the human desire is not given a stimulus, the strength is increased, and it is desirable that the value of the desire is maintained within a certain range, respectively. It is known that there is an act that fulfills every desire.

The desire of a personal robot can be defined as the measures by which the personal robot must remain at a certain level as it interacts with humans. Each desire processor 130 according to the present invention is designed to reflect the results of the Lorentz and tinbergen above.

Therefore, each desire variable basically has a smaller value as it is satisfied and a larger value when it is not satisfied, and when the value of the desire variable exceeds a predetermined threshold (stable region), it affects the emotional state. Given.

Therefore, first, in step S401, if the events of Event_List affect desire n, the value of desire n is decreased, and in step S402, if there is no event affecting desire n in Event_List, the value of desire n is increased. .

Next, in step S403, the value of desire n is updated to reflect the mutual suppression and stimulus degree between desires.

On the other hand, the value of the desire variable is updated based on Equation 1 below.

는 시간 t의 욕구 i의 값이고,

는 이벤트 k가 욕구 i에 미치는 영향을 나타내며,

는 욕구 j와 욕구 i 상호간의 억제 및 자극 정도를 나타내는 값이고,

는 아무 자극도 없는 경우에 욕구 변수의 값을 증가시키는 함수이다. 또한,

,

및

등의 값은 로봇 개발자가 직접 설정함으로써, 로봇이 개발자가 의도하는대로 감성 반응을 할 수 있게끔 한다.here,

Is the value of desire i at time t,

Represents the effect of event k on desire i,

Is a value representing the degree of inhibition and stimulation between desire j and desire i,

Is a function that increases the value of the desire variable when there is no stimulus. Also,

,

And

The value of, etc. is set directly by the robot developer, allowing the robot to react emotionally as the developer intended.

5 is a flowchart illustrating processing occurring in the plurality of emotion managers 140 illustrated in FIG. 1, wherein the emotion manager 140 includes an event transmitted from the event processor 120 and a robot updated by the desire manager. The values of the emotional variables are updated based on the current desire state of and the current mood of the robot.

The emotion manager 140 is also designed to be defined separately for each emotion. On the other hand, the emotion manager proposed in the present invention is designed to reflect the characteristics of the human emotion system shown in Table 1 below generally known.

Characteristics of the Human Emotion System Response decay If not stimulated again, the intensity of sensitivity weakens over time. Repeated Strike If the stimulus is repeated, the intensity of the emotions increases. Non-linearity Human emotional systems are non-linear, but can be approximated linear in the input and output ranges of certain areas. Activation In order for emotion to be active, the intensity must be above a certain level, which is influenced by mood. Saturation Sensitivity intensity converges at some level and no longer increases.

The emotion manager updates the emotion value in three steps.

(1) Step 1: Emotional value change by desire state and events

In this step, each emotional variable is changed by Equation 2 below.

는 시간 t의 감정 i의 강도값이고,

는 쇠퇴 함수(Decay Function)로서, 자극이 없는 경우에 감정 변수의 값을 낮추는 함수이며,

는 이벤트 k가 감정 i에 미치는 영향 함수(Influence Function)이고,

는 욕구 j가 감정 i에 미치는 영향 함수이다.From here,

Is the intensity of emotion i at time t,

Is a decay function, which is a function that lowers the value of an emotional variable in the absence of stimulation.

Is the influence function of event k on emotion i,

Is an influence function of desire j on emotion i.

In addition, each function can be defined by the robot developer as needed.

(2) Step 2: Apply Mutual Stimulation and Suppression

This step considers mutual stimulation and suppression between emotions. In general, there is a mutual restraint between 'joy' and 'sorrow', and a mutual stimulus relationship between 'anger' and 'sorrow'. In addition, various stimulus / suppression relationships can be defined according to the needs of the robot developer. In this step, the current final emotional variable is determined by Equation 3 below.

는 감정 k가 감정 i에 미치는 영향을 정의하는 함수이다.here,

Is a function that defines the effect of emotion k on emotion i.

(3) Step 3: Conversion Using Nonlinear Functions

를 아래의 [수학식 3]을 사용하여 변환한다.In order to apply the characteristics of non-linearity, saturation, and activation to the robot emotion synthesis

Is converted using Equation 3 below.

The function is based on the Michaelis-Menten formula, which is known to be very useful for modeling the biological behavior of cells. Here, a and b are constants, mood (t) represents the mood of the robot, and k is a value for adjusting the curvature.

The mood of the robot is defined based on the recent emotional state, and a positive value means good mood, and a negative value means bad mood. On the other hand, the equation for defining the mood of the robot is shown in Equation 5 below.

는 감성 j가 로봇의 기분에 영향을 미치는 정도를 정의한 함수이다. '분노', '혐오', '슬픔' 등의 부정적인 감정은 기분에 부정적인 영향을 미치며, '기쁨'은 기분에 긍정적인 영향을 미친다.here,

Is a function that defines the degree to which emotion j affects the mood of the robot. Negative emotions such as anger, disgust, and sadness have a negative effect on mood, and joy has a positive effect on mood.

는 i 번째 전의 감성 상태가 현재의 기분에 영향을 미치는 정도를 정의한 함수인데, i가 커질수록 그 값이 작아지도록 함으로써, 최근의 감성 상태가 현재의 기분에 더 많은 영향을 줄 수 있도록 정의되어 있다.

Is a function that defines the degree to which the emotional state before the i th affects the current mood. The larger the value of i is, the smaller the value is, so that the recent emotional state can have more influence on the current mood. .

는 기분이 감정 i에 미치는 영향을 정의한 함수로서, 부정적인 감정에 대해서는 기분이 좋을 때 활성화가 억제되도록 하며, 긍정적인 감정에 대해서는 기분이 나쁠 때 활성화가 억제되도록 한다.

Is a function that defines the effect of mood on emotion i. For negative emotions, activation is suppressed when feeling good and activation is suppressed when feeling bad.

At this time, the mood value is updated in such a way that the emotion value is inversely proportional to the current time to affect the current mood.

The above technical contents are arranged according to the flow, as follows.

First, in step S501, if the events and desire variables stored in the Event_List have an influence on emotion n, the value of emotion n is increased, and in step S502, if there is no factor influencing the value of emotion n, Decrease the value.

Next, in step S503, the value of emotion n is updated to reflect the degree of mutual suppression and stimulation between emotions, and in step S504, the Michaelis-Menten function is based on the relationship between the robot's mood value and emotion n. Move parallel to.

In step S505, the emotional n value is updated using the Michaelis-Menten function so that the emotional expression of the robot exhibits the characteristics of non-linearity, saturation and activation, which are characteristics of the human emotional system.

Then, in step S506, the mood value is updated so that the emotion value affects the current mood in inverse proportion to the current time.

At this time, the parallel movement of the Michaelis-Menten function follows the following four cases.

(Case 1) If the mood value is greater than zero (good mood), and emotion n is a positive emotion, the activation of emotion n is promoted by shifting the Michaelis-Menten function by (-mood value).

(Case 2) If the mood value is greater than zero and the emotion n is a negative emotion, the movement of the emotion n is suppressed by moving in parallel by (+ mood value).

(Case 3) If the mood value is less than 0 (bad mood), and the emotion n is a positive emotion, it is moved by (-mood value) in parallel, thereby suppressing the activation of the emotion n.

(Case 4) If the mood value is less than 0 and the emotion n is a negative emotion, it is moved by (+ mood value) in parallel to promote activation of the emotion n.

6 is a graph showing a result of the non-linear function of the emotional variable converted by the value of the mood variable according to an embodiment of the present invention. As can be seen from the graph, it can be seen that the nonlinear graph of the emotional variable on the left side is converted into the graph on the right side.

Through this process, the level of activation of each emotion can be controlled.

When the value of the emotional variable is updated, the emotion represented by the emotional variable having the highest value among the emotional variables is regarded as the emotional state of the robot.

The processing that takes place in the preprocessor 100 will now be described.

Emotional recognition of the surrounding environment is necessary to induce the robot to react emotionally to the surrounding environment similarly to the user. The robot responds to the surrounding environment similarly to the emotional response of the user, thereby controlling the surrounding environment to be suitable for the user, and contributes to making the user feel homogeneous and friendly by displaying a similar response to the user.

There are many factors that influence human emotional reactions and the process is complicated, but there is still no high level of research in this field, and there is a limit to the information that robots can detect. It is impossible to learn emotional responses to the environment. Therefore, the present invention aims to assume the emotional response of the user to the surrounding environment as a probabilistic model and to learn it.

Basically, the Bayesian Learning method can be used to accurately learn the original model when there is enough learning data, but it takes a lot of time for the robot to gather enough learning data online. In the present invention, it is assumed that the user's response to the neighboring areas in the environment state space formed by the environment variables will be similar, and modified to Bayesian learning method represented by Equation 6 below.

In the present invention, P 'is defined as a probability model of the emotional response of the user.

The meaning of Equation 6 is as follows.

First, since the goal is to find the probability of the user's emotional response at A = (a1, a2, ..., an) in the environmental state space, we will base the probability of the user's emotional response at each location around A. Obtained using Bayesian learning, then multiply the result by the weight.

Meanwhile, in the present invention, the weight for each point is set in inverse proportion to the distance to A, and the sum of the weights is set to 1. (Normalization) In addition, in the present invention, the probability of the emotional response of the user at A is determined. Instead of specifying the set of surrounding points in detail, the developer decides through experiments considering the size of the state space and the number of emotional responses of the user.

According to this method, in the problem of obtaining the probability of the emotional response of the user at a point in the environmental state space, using the method, the same or more learning data with less learning data than the basic Bayesian learning method is used. High accuracy results can be obtained.

FIG. 7A is a conceptual diagram illustrating a case where a Bayesian learning method used in an embodiment of the present invention is applied to emotional recognition (recognition) of a surrounding environment.

The robot assumes that the surrounding environment information (temperature, brightness, noise, etc.) at the time when the facial expression of the user is recognized is the cause that causes the user to express the expression (recognized user's facial expression and surrounding environment information (environmental state space) Corresponds to a point on the image.)) Generates training data.

At regular intervals, the robot updates the probability of the user's emotional response at each point in the environmental state space using the Bayesian learning method and the learning data generated up to that point. This information becomes the user's emotional response model.

7B is a conceptual diagram illustrating a case where the Bayesian learning method used in one embodiment of the present invention is applied to emotional recognition (applied part) of the surrounding environment.

In the application part, when the surrounding environment information is input through the sensor, the probability of the emotional response of the user at the point on the environment state space corresponding thereto is calculated using the algorithm (P ',' user's emotional response model 'is used). ), It is used to update the emotional state of the robot.

On the other hand, even when there is a sufficient amount of learning data, if only the assumption is maintained, results similar to basic Bayesian learning are provided.

FIG. 8 is a graph showing a learning result of probabilistic models of six emotional responses of a user arbitrarily synthesized to two environment variables using a modified Bayesian learning method according to an embodiment of the present invention.

As can be seen from FIG. 8, the method proposed by the present invention has higher accuracy as the number of learning data increases than the other two methods (basic Bayesian learning, Naive-Bayes Classfier).

While the invention has been described above based on the preferred embodiments thereof, these embodiments are intended to illustrate rather than limit the invention. It will be apparent to those skilled in the art that various changes, modifications, or adjustments to the above embodiments can be made without departing from the spirit of the invention. Therefore, the protection scope of the present invention will be limited by the appended claims, and should be construed as including all such changes, modifications or adjustments.

As described above, according to the present invention, due to the integrated structure of the object-oriented program structure in consideration of the scalability and the message delivery method in consideration of the universality, there is an effect that it is relatively easy to implement the emotional synthesis of the personal robot for various purposes.

In addition, by designing and providing a mood model of the robot and a specific emotional state updating algorithm, it is possible to implement a practical and useful emotional synthesis function similar to a human, and also, through the emotional recognition of the surrounding environment, the personal robot has its own user. There is an effect that can lead to an emotionally similar reaction.

In conclusion, the present invention can support human-centered friendly human robot interaction, which can bring about high value-adding of personal robots and the growth of the personal robot market.

Claims (21)

In the robot emotion manager that updates the values of the emotional variables based on the input event, the robot's current desire state, and the robot's mood, If the input values do not affect the emotion, the emotion value is reduced by using a decay function, and if there is an influence factor, the emotion value is increased by using the influence function. Way; After receiving the changed emotion value from the emotion value changing means, reflecting mutual suppression and stimulus degree between each emotion, and receiving the reflected emotion value and the current mood value of the robot, the Michaelis-Menten function is applied to the axis of the emotion value. Emotion value updating means for updating the emotion value by adjusting the degree of activation of each emotion by moving in parallel with the current mood value; And By receiving the updated emotion value from the emotion value updating means, the emotion values affecting the mood of the robot affect the mood of the robot in inverse proportion to the elapsed time, reflecting the updated emotion value of the robot. Mood value updating means for updating a current mood value; Emotion manager of the robot comprising a. The method of claim 1, And said emotion value changing means changes each emotion value by the following [Equation 1]. [Equation 1]

From here,

Is the intensity of emotion i at time t,

Is a decay function, which is a function that lowers the value of an emotional variable in the absence of stimulation.

Is the influence function of event k on emotion i,

Is an influence function of desire j on emotion i. The method of claim 1, The emotion value updating means of the emotion controller of the robot, characterized in that to reflect the degree of mutual suppression and stimulation between each emotion by the following [Equation 2]. [Equation 2]

here,

Is a function that defines the effect of emotion k on emotion i. The method according to claim 1 or 3, The emotion value updating means is the emotion manager of the robot, characterized in that for finally updating the emotion value by the following equation (3). [Equation 3]

Here, a and b are constants, mood (t) represents the mood of the robot, and k is a value for adjusting the curvature. The method of claim 1, And said mood value updating means determines the mood value by the following [Equation 4]. [Equation 4]

here,

Is a function that defines how emotional j affects the robot's mood,

Is a function that defines how the emotional state before the i th affects the current mood,

Is a function that defines the effect of mood on emotion i. In the robot emotion management method of updating the value of the emotional variables based on the input event, the robot's current desire state and the mood of the robot, A first step of decreasing the emotion value using a decay function if the input values do not affect the emotion, and increasing the emotion value using an influence function if there is a factor influencing the input values ; After receiving the emotion value changed in the first step and reflecting the degree of mutual suppression and stimulation between each emotion, and receiving the reflected emotion value and the current mood value of the robot, the Michaelis-Menten function is applied to the axis of the emotion value. A second step of updating the emotion value by adjusting the degree of activation of each emotion by moving in parallel by the current mood value; And The current emotion value of the robot is reflected by reflecting the updated emotion value to receive the updated emotion value in the second step, so that the emotion values affecting the mood of the robot affect the mood of the robot in inverse proportion to the elapsed time. A third step of updating the mood value; Emotional management method of a robot comprising a. The method of claim 6, The second step is a emotion management method of the robot, characterized in that for updating the emotion value by the following equation (5). [Equation 5]

Here, a and b are constants, mood (t) represents the mood of the robot, and k is a value for adjusting the curvature. The method according to claim 6 or 7, Feeling value of the robot is emotion management method of the robot, characterized in that determined by the following [Equation 6]. [Equation 6]

here,

Is a function that defines how emotional j affects the robot's mood,

Is a function that defines how the emotional state before the i th affects the current mood,

Is a function that defines the effect of mood on emotion i. The method of claim 6, The second step is Non-linearity (human emotion system is non-linear, but approximated linearly in the input / output range of a specific area), Activation (emotion is activated when the intensity of the reference or the like is above a certain level), The intensity of emotion converges to a specific value and does not increase any more. A preprocessor for emotionally recognizing the surrounding environment such that the robot reacts emotionally to the surrounding environment similarly to the user; A desire manager which receives the results of the preprocessor and updates a value of a corresponding desire variable; An emotion manager that receives the input event, the desire variable updated by the desire manager and the current mood of the robot, and updates the values of the emotion variables; And A state output unit that considers the emotion represented by the emotion variable having the highest value among the emotion variables updated by the emotion manager as the emotion state of the robot; Including, The emotion manager, If the input values do not affect the emotion, the emotion value is reduced by using a decay function, and if there is an influence factor, the emotion value is increased by using the influence function. Way; After receiving the changed emotion value from the emotion value changing means, reflecting mutual suppression and stimulus degree between each emotion, and receiving the reflected emotion value and the current mood value of the robot, the Michaelis-Menten function is applied to the axis of the emotion value. Emotion value updating means for updating the emotion value by adjusting the degree of activation of each emotion by moving in parallel with the current mood value; And By receiving the updated emotion value from the emotion value updating means, the emotion values affecting the mood of the robot affect the mood of the robot in inverse proportion to the elapsed time, reflecting the updated emotion value of the robot. Mood value updating means for updating a current mood value; Emotional synthesizing apparatus for a robot comprising a. The method of claim 10, The preprocessor applies a probability model modified from a basic Bayesian learning method under the assumption that the user's response to neighboring regions in the environment state space generated by environment variables will be similar. Emotional synthesis device. The method of claim 11, The preprocessor obtains the probability of the emotional response of the user at each location around an arbitrary point in the environmental state space by using the basic Bayesian learning method, and multiplies the result by weight to add the results to preprocess events. Emotional synthesis device of robot. The method of claim 12, The apparatus for synthesizing the robot, characterized in that the weight for each position is set in inverse proportion to the distance to the arbitrary point, and the total of the weights is normalized to be 1. The method of claim 10, The desire manager decreases the desire value when the events input from the preprocessor affect the desire, and increases the desire value if it does not affect the desire value, and then reflects the mutual suppression and stimulus levels between the desires. Emotional synthesis device of the robot, characterized in that for updating. The method of claim 14, The desire manager is a robot emotion synthesis device, characterized in that for determining the desire value using the following equation (7). [Equation 7]

here,

Is the value of desire i at time t,

Represents the effect of event k on desire i,

Is a value representing the degree of inhibition and stimulation between desire j and desire i,

Is a function that increases the value of the desire variable when there is no stimulus. A first step of emotionally recognizing the surrounding environment such that the robot responds emotionally to the surrounding environment similarly to the user; A second step of receiving the results of the first step and updating a value of a corresponding desire variable; If the input value and the updated desire variables do not have a factor affecting emotion, the emotion value is reduced using a decay function, and if there is a factor that affects the emotion value using an influence function A third step of increasing; After receiving the emotion value changed in the third step and reflecting the degree of mutual suppression and stimulation between each emotion, and receiving the reflected emotion value and the current mood value of the robot, the Michaelis-Menten function is applied to the axis of the emotion value. A fourth step of updating the emotion value by adjusting the activation degree of each emotion by moving in parallel with the current mood value; And The current emotion value of the robot is reflected by receiving the updated emotion value in the fourth step, so that the emotion values affecting the mood of the robot affect the mood of the robot in inverse proportion to the elapsed time. A fifth step of updating the mood value; Emotional synthesis method of a robot comprising a. The method of claim 16, The first step is to apply a probabilistic model modified from a basic Bayesian learning method under the assumption that the user's response to neighboring regions in the environment state space generated by the environment variables will be similar. Emotional synthesis method of robot. The method of claim 17, The first step is to calculate the probability of the emotional response of the user at each location around an arbitrary point in the environmental state space using the basic Bayesian learning method, and then multiply the result by weight to preprocess the events. Emotion synthesis method of robot. The method of claim 18, The weight for each position is set in inverse proportion to the distance to the arbitrary point, and the sum of the weight is normalized (Normalization) so that the sum of the weights (Normalization). The method of claim 19, The modified Bayesian learning method is the emotion synthesis method of the robot, characterized by the following [Equation 8]. [Equation 8]

Here, P 'is a probabilistic model of the emotional response of the user. The method of claim 16, The desire manager is a method for synthesizing the robot, characterized in that to determine the desire value using the following equation (9). [Equation 9]

here,

Is the value of desire i at time t,

Represents the effect of event k on desire i,

Is a value representing the degree of inhibition and stimulation between desire j and desire i,

Is a function that increases the value of the desire variable when there is no stimulus.

Images