JP2002205289A - Action control method for robot device, program, recording medium and robot device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To allow a robot device to identify an interactive object and take desirable action depending on thereto. SOLUTION: The robot device comprises an action learning means for making the robot device learn input information concerning the interactive object and then learn highly evaluated action with a RNN 100 and an action control means for identifying the interactive object from the input information concerning the interactive object in accordance with the learning result and presenting the action having a high evaluation value learned by the action learning means depending on the identified interactive object with an inverse RNN (RNN-1).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a robot apparatus, an operation control method of the robot apparatus for controlling the operation of the robot apparatus, and more particularly, to a robot apparatus capable of learning the operation, an operation control method of the robot apparatus, and a program. And a recording medium.

[0002]

2. Description of the Related Art In recent years, there has been provided a robot device whose external shape is formed by imitating an animal such as a dog or a cat. Some of such robot devices operate autonomously according to external information or internal conditions.

[0003]

By the way, as in humans, animals sometimes try to attract the other party. For example, it increases one's motivation. Also, if the other party sees such an action to attract attention, it cannot be sick.

[0004] For example, in the relationship between the owner and the pet, the owner is glad if he sees the action that the pet is trying to pay attention to to be praised. For example, personality improves.
Among the robot devices, there is also a robot device whose character changes in a dialogue of a user or the like. In the case of such a robot device, it is possible to influence an operation result to be praised on the character, and thus it can be said that entertainment is further increased. .

[0005] The operation to be praised differs depending on the other party (for example, the owner). That is,
For some owners, the so-called "sitting" movement or posture is a target for praise, or the so-called "lying"
Movement or posture is a target for praise.

Accordingly, the present invention has been made in view of the above situation, and has an operation control method, a program, and a recording apparatus for a robot apparatus capable of identifying a conversation partner and performing an operation commended accordingly. It is an object to provide a medium and a robot device.

[0007]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, a motion control method for a robot apparatus according to the present invention includes the steps of: identifying a conversation partner; A learning step of learning the robot device, and identifying the conversation partner based on the learning result in the learning process from the information input with respect to the conversation partner, and learning the highly evaluated operation by the robot device according to the identified conversation partner. And an operation control step of exposing.

In such an operation control method for a robot device, the conversation partner is identified, the robot device learns an operation in which the identified conversation partner evaluates highly, and a learning result is obtained from information input with respect to the conversation partner. And the robot apparatus expresses the learned high evaluation operation based on the identified conversation partner.

According to such an operation control method for a robot device, the robot device learns the conversation partner and the operation that the conversation partner evaluates highly, and based on the learning result,
Express highly evaluated actions according to the conversation partner.

[0010] Further, in order to solve the above-mentioned problems, the program according to the present invention includes a learning step of identifying a dialogue partner and causing the robot apparatus to learn an operation in which the identified dialogue partner evaluates highly. Executing a motion control step of identifying a conversation partner based on the learning result in the learning process from information input with respect to the conversation partner, and expressing a learned operation with a high evaluation by the robot apparatus according to the identified conversation partner. Let it.

[0011] Such a program identifies a conversation partner and performs an operation in which the identified conversation partner evaluates highly.
The robot device is made to learn, the conversation partner is identified based on the learning result from the information input with respect to the conversation partner, and the learned motion with high evaluation is expressed by the robot device according to the identified conversation partner.

A robot apparatus whose operation is controlled by such a program learns a conversation partner and an operation that the conversation partner evaluates highly, and, based on the learning result, performs a movement with a high evaluation according to the conversation partner. Express.

[0013] Further, in order to solve the above-mentioned problems, the recording medium according to the present invention includes a learning step of identifying a dialogue partner and causing the robot apparatus to learn an operation of highly evaluating the identified dialogue partner. An operation control step of identifying a dialogue partner based on learning results in a learning process from information input with respect to the dialogue partner, and causing a robot apparatus to express a learned highly evaluated operation according to the identified dialogue partner. The program to be executed is recorded.

Such a recording medium allows the robot apparatus to learn an operation in which the conversation partner is identified and gives a high evaluation to the identified conversation partner, and performs the conversation based on the learning result from the information input with respect to the conversation partner. The other party is identified, and the learned motion with a high evaluation is expressed by the robot apparatus according to the identified dialogue partner.

A robot device whose operation is controlled by a program recorded on such a recording medium learns a conversation partner and an operation that the conversation partner evaluates highly, and based on the learning result, gives the conversation partner. An operation with a high evaluation is expressed accordingly.

Further, in order to solve the above-mentioned problems, the robot apparatus according to the present invention has a learning means for identifying a conversation partner and learning an operation in which the identified conversation partner gives a high evaluation, and an input for the conversation partner. Operation control means for identifying a conversation partner based on the learning result of the learning means based on the obtained information, and expressing a learned operation with a high evaluation according to the identified conversation partner.

Such a robot device identifies a conversation partner, learns an operation in which the identified conversation partner gives a high evaluation, and identifies a conversation partner based on a learning result from information input about the conversation partner. Then, the learned highly evaluated operation is expressed according to the identified conversation partner. Therefore, the robot device learns the conversation partner and the operation in which the conversation partner evaluates high, and expresses the operation with high evaluation according to the conversation partner based on the learning result.

[0018]

Embodiments of the present invention will be described below in detail with reference to the drawings. This embodiment is an autonomous robot apparatus that autonomously acts according to a surrounding environment (external factors) and an internal state (internal factors). According to the application of the present invention, the robot device can identify a user through learning and can express an operation praised accordingly.

In the embodiment, first, the configuration of the robot device will be described, and thereafter, the application portion of the present invention in the robot device will be described in detail.

(1) Configuration of Robot Apparatus According to the Present Embodiment As shown in FIG. 1, a so-called pet robot having a shape imitating a "dog" is provided, and leg units 3A, 3A, 3B, 3C, and 3D are connected, and a head unit 4 and a tail unit 5 are connected to the front end and the rear end of the body unit 2, respectively.

As shown in FIG. 2, a CPU (Central Processing Unit) 10 and a DRA
M (Dynamic Random Access Memory) 11, Flash ROM (Read 0nly Memory) 12, PC (Personal Co.)
A controller 16 formed by interconnecting a card interface circuit 13 and a signal processing circuit 14 via an internal bus 15 and a battery 17 as a power source of the robot apparatus 1 are housed therein. . The body unit 2 also houses an angular velocity sensor 18 and an acceleration sensor 19 for detecting the acceleration of the direction and movement of the robot device 1.

The head unit 4 includes a CCD (Charge Coupled Device) camera 20 for capturing an image of an external situation, and a pressure applied by a physical action such as "stroke" or "hit" from the user. , A distance sensor 22 for measuring a distance to an object located ahead, a microphone 23 for collecting external sounds, and a speaker for outputting a sound such as a cry 24 and LEDs (Light Emitting Diodes (not shown)) corresponding to the “eyes” of the robot apparatus 1 are arranged at predetermined positions.

Further, the joints of the leg units 3A to 3D, the leg units 3A to 3D, and the trunk unit 2
The actuators 25 _{1 for the} degrees of freedom are respectively provided at the connection portions of the head unit 4 and the trunk unit 2, and at the connection portion of the tail 5 A of the tail unit 5.
To 25 _n and potentiometers 26 _{1 to} 26 _n are provided. For example, each of the actuators 25 _{1 to} 25 _n has a servomotor. By driving the servo motor, the leg units 3A to 3D are controlled, and transition to a target posture or operation is made.

The angular velocity sensor 18, acceleration sensor 19, touch sensor 21, distance sensor 22, microphone 23, speaker 24 and each potentiometer 2
6 _1-26 various sensors and LED and the actuator ₂₅ 1 to 25 _n, such as _n is connected to the signal processing circuit 14 of the control unit 16 via a corresponding hub ₂₇ 1 ~ 27 _n, CCD camera 20 and the battery 1
7 are directly connected to the signal processing circuit 14, respectively.

The signal processing circuit 14 sequentially takes in the sensor data, image data, and audio data supplied from the above-described sensors, and sequentially stores them at predetermined positions in the DRAM 11 via the internal bus 15. In addition, the signal processing circuit 14 sequentially takes in remaining battery power data indicating the remaining battery power supplied from the battery 17 and stores the data at a predetermined position in the DRAM 11.

The sensor data, image data, voice data, and remaining battery data stored in the DRAM 11 in this manner are thereafter transmitted to the robot device 1 by the CPU 10.
It is used when controlling the operation of.

In practice, when the power of the robot apparatus 1 is turned on, the CPU 10 executes the control program stored in the memory card 28 or the flash ROM 12 inserted in the PC card slot (not shown) of the body unit 2 into a P program.
The data is read out via the C card interface circuit 13 or directly and stored in the DRAM 11.

The CPU 10 then determines the status of itself and its surroundings and the usage based on the sensor data, image data, audio data, and remaining battery data sequentially stored in the DRAM 11 from the signal processing circuit 14 as described above. Judge the instruction from the person and the presence or absence of the action.

Furthermore, CPU 10 is configured to determine a subsequent action based on the control program that is stored in the determination result and DRAM 11, by driving the actuator ₂₅ 1 to 25 _n required based on the determination result, the head The unit 4 can be swung up, down, left and right, the tail 5A of the tail unit 5 can be moved, and each leg unit 3A can be moved.
3D is driven to perform an action such as walking.

At this time, the CPU 10 generates audio data as required and supplies the generated audio data to the speaker 24 as an audio signal via the signal processing circuit 14 so that the audio based on the audio signal is output to the outside. LE above
D is turned on, off or blinked.

As described above, the robot apparatus 1 can autonomously act in accordance with the situation of itself and the surroundings, and instructions and actions from the user.

(2) Software Configuration of Control Program Here, the software configuration of the above-described control program in the robot apparatus 1 is as shown in FIG. In FIG. 3, a device driver layer 30 is located at the lowest layer of the control program, and includes a plurality of device drivers.
It comprises a device driver set 31 composed of drivers. In this case, each device driver
An object that is permitted to directly access hardware used in a normal computer, such as a CCD camera 20 (FIG. 2) and a timer, and performs processing in response to an interrupt from the corresponding hardware.

The robotic server object 32 is located at the lowest layer of the device driver layer 30 and includes, for example, the various sensors and actuators 2 described above.
A virtual robot 33 comprising a software group that provides a 5 _to 253 interface for accessing hardware such as _n, a bus word manager 34 made of a software suite for managing the power supply switching, the other various device A device driver manager 35, which is a group of software for managing drivers, and a designed driver, which is a group of software for managing the mechanism of the robot device 1,
And a robot 36.

The manager object 37 includes an object manager 38 and a service manager 39.
It is composed of Object Manager 38
Is a software group that manages activation and termination of each software group included in the robotic server object 32, the middleware layer 40, and the application layer 41, and includes a service manager 39.
Are a group of software for managing the connection of each object based on the connection information between the objects described in the connection file stored in the memory card 28 (FIG. 2).

The middleware layer 40 is located on the upper layer of the robotic server object 32 and is composed of a group of software for providing basic functions of the robot apparatus 1 such as image processing and sound processing. I have. Further, the application layer 41 is located above the middleware layer 40, and
It is composed of a software group for determining an action of the robot apparatus 1 based on a processing result processed by each software group constituting the layer 40.

FIG. 4 shows specific software configurations of the middleware layer 40 and the application layer 41.

As shown in FIG. 4, the middle wear layer 40 is for noise detection, temperature detection, brightness detection, scale recognition, distance detection, posture detection, touch sensor,
Each signal processing module 50 for motion detection and color recognition
58 and input semantics converter module 5
9 and an output semantics converter module 68 and signal processing modules 61 to 67 for attitude management, tracking, motion reproduction, walking, fallback recovery, LED lighting and sound reproduction. And an output system 69.

Each signal processing module 50 to 50 of the recognition system 60
Numeral 58 fetches corresponding data among the sensor data, image data, and audio data read from the DRAM 11 (FIG. 2) by the virtual robot 33 of the robotic server object 32, and performs predetermined processing based on the data. Is given to the input semantics converter module 59. Here, for example, the virtual robot 33 is configured as a part that exchanges or converts signals according to a predetermined communication protocol.

The input semantics converter module 59 detects "noisy", "hot", "bright", "ball detected", and "fallover" based on the processing results given from each of the signal processing modules 50 to 58. Self and surrounding conditions such as `` has been stroked '', `` stroked '', `` hitted '', `` he heard the domes '', `` detected a moving object '' or `` detected an obstacle '', and the user , And outputs the recognition result to the application layer 41 (FIG. 2).

As shown in FIG. 5, the application layer 41 includes five modules: a behavior model library 70, a behavior switching module 71, a learning module 72, an emotion model 73, and an instinct model 74.

As shown in FIG. 6, the behavior model library 70 includes “when the remaining battery power is low”, “returns to fall”, “when avoiding obstacles”, and “when expressing emotions”. , And independent action models 70 ₁ to 70 _n are respectively provided corresponding to some pre-selected condition items such as “when a ball is detected”.

The behavior models 70 ₁ to 70 _n
Are stored in the emotion model 73 as described later, as necessary, when a recognition result is given from the input semantics converter module 59 or when a certain period of time has passed since the last recognition result was given. The next action is determined with reference to the corresponding emotion parameter value and the corresponding desire parameter value held in the instinct model 74, and the determination result is output to the action switching module 71.

In the case of this embodiment, each of the behavior models 70 ₁ to 70 _n uses the following method to determine the next behavior.
One node (state) NODE _{0 to} N as shown in FIG.
The transition probability _P 1 to P _n which is set respectively arc _ARC 1 _{~ARC n1} connecting between whether to transition from ODE _n to any other node NODE ₀ ~NODE _n each node NODE ₀ ~NODE _n An algorithm called a finite stochastic automaton that determines stochastically on the basis is used.

More specifically, each of the behavior models 70 ₁ to 70 _1
n has a state transition table 80 as shown in FIG. 8 for each of the nodes NODE _{0 to} NODE _n corresponding to the nodes NODE ₀ to NODE _n forming their own behavior models 70 ₁ to 70 _n , respectively. ing.

In the state transition table 80, the node NO
Input events (recognition results) as transition conditions in DE _{0 to} NODE _n are listed in order of priority in the row of “input event name”, and further conditions for the transition conditions are described in the rows of “data name” and “data range”. It is described in the corresponding column.

Therefore, in the node NODE ₁₀₀ represented by the state transition table 80 in FIG.
LL) ”, the“ size (SIZ) of the ball given together with the recognition result is given.
E) is in the range of “0 to 1000”, or when a recognition result of “obstacle detected (OBSTACLE)” is given, the “distance (DISTANCE)” to the obstacle given together with the recognition result is given. )) Is in the range of “0 to 100”, which is a condition for transitioning to another node.

In the node NODE ₁₀₀ , even when the recognition result is not input, the behavior model 70 ₁
Of the parameter values of each emotion and each desire held in the emotion model 73 and the instinct model 74 which are periodically referred to by the _n- 70n, “joy” and “surprise” held in the emotion model 73 are stored. )) Or "Sadness (SUDNESS)" when the parameter value is in the range of "50 to 100", it is possible to transition to another node.

[0048] In the state transition table 80, with the node name that can transition from the node NODE ₀ ~ NODE _n in the column of "transition destination node" are listed in the column of "transition probability of other Nodohe", " Input event name ",
Other nodes NOD that can transition when all the conditions described in the rows of “data value” and “data range” are met
The transition probabilities from E _{0 to} NODE _n are respectively described in corresponding portions in the column of “transition probability to other nodes”, and the action to be output when transitioning to that node NODE _{0 to} NODE _n is “other It is described in the row of “output action” in the column of “transition probability to node”. Note that the sum of the probabilities of each row in the column of “transition probability to another node” is 100
[%].

Therefore, in the node NODE ₁₀₀ represented by the state transition table 80 in FIG. 8, for example, “ball is detected (BALL)”, and the “SIZE” of the ball is in the range of “0 to 1000”. Is given, the probability of “30 [%]” and “node N
ODE ₁₂₀ (node 120) "and then" A
The action of “CTION1” is output.

Each of the behavior models 70 ₁ to 70 _n has the node NO described in the state transition table 80 as described above.
DE ₀ to NODE _n are connected to each other, and when a recognition result is given from the input semantics converter module 59 or the like, the probability is calculated using the state transition table of the corresponding nodes NODE ₀ to NODE _n. The next action is determined, and the result of the determination is output to the action switching module 71.

The action switching module 71 shown in FIG.
Each behavior model 70 in the behavior model library 70₁~ 70
_nOf the actions output from
High priority action model 70₁~ 70_nOutput from
Command to select the action and execute it
(Hereinafter, this is called an action command.)
A layer 40 output semantics converter module
To the rule 68. In this embodiment,
Is the behavior model 70 shown on the lower side in FIG. ₁~
70_nThe higher the priority, the higher the priority.

After the action is completed, the action switching module 71 outputs the output semantics converter module 68.
The completion of the action is notified to the learning module 72, the emotion model 73, and the instinct model 74 based on the action completion information given by the user.

On the other hand, the learning module 72 inputs the recognition result of the teaching received from the user, such as “hit” or “stroke”, among the recognition results given from the input semantics converter module 59. I do.

Then, based on the recognition result and the notification from the action switching module 71, the learning module 72 lowers the probability of occurrence of the action when "beaten (scolded)" and "strokes ( praised obtained) "sometimes to increase the expression probability of that action, behavior model 70 _1-7 corresponding in behavioral model library 70
Change the corresponding transition probabilities of 0 _n .

On the other hand, the emotion model 73 indicates “joy (jo
y) "," sadness "," anger ",
For a total of six emotions, “surprise”, “disgust”, and “fear”, a parameter indicating the strength of the emotion is stored for each emotion. Then, the emotion model 73 converts the parameter values of each of these emotions into a specific recognition result such as “hit” and “stroke” given from the input semantics converter module 59 and the elapsed time and action switching module 71. Updates periodically based on the notification from.

Specifically, the emotion model 73 is determined based on the recognition result given from the input semantics converter module 59, the behavior of the robot device 1 at that time, the elapsed time since the last update, and the like. The amount of change of the emotion at that time calculated by the arithmetic expression is expressed by △ E
[T], E [t] of the current parameter value of the emotion, the coefficient representing the sensitivity of the emotion as k _e, (1) the parameter value of the emotion in a next period by equation E [t +
1] is calculated, and the current parameter value E of the emotion is calculated.
The parameter value of the emotion is updated by replacing it with [t]. The emotion model 73 updates the parameter values of all emotions in the same manner.

[0057]

(Equation 1)

The degree to which each recognition result and the notification from the output semantics converter module 68 affect the variation ΔE [t] of the parameter value of each emotion is determined in advance. Is the amount of change in the parameter value of the emotion of “anger” △ E
[T] is greatly affected, and the recognition result such as “stroke” is the variation amount of the parameter value of the emotion of “joy” 喜 び E
[T] is greatly affected.

Here, the notification from the output semantics converter module 68 is so-called action feedback information (action completion information), information on the appearance result of the action, and the emotion model 73 also uses such information. Change emotions. This is, for example, a behavior such as "barking" that lowers the emotional level of anger. The notification from the output semantics converter module 68 is also input to the learning module 72 described above, the learning module 72 changes the corresponding transition probability of the behavioral models 70 ₁ to 70 _n based on the notification.

The feedback of the action result may be made by the output of the action switching modulator 71 (the action to which the emotion is added).

On the other hand, the instinct model 74 is “exercise desire (exer
cise) "," affection "," appetite "
e) ”and“ curiosity ”4
For each of the desires, a parameter indicating the strength of the desire is stored for each of the desires. And instinct model 7
4 periodically updates these desire parameter values based on the recognition result given from the input semantics converter module 59, the elapsed time, the notification from the action switching module 71, and the like.

More specifically, the instinct model 74 determines the “movement desire”, “affection desire”, and “curiosity” based on the recognition result, the elapsed time, the notification from the output semantics converter module 68, and the like. The change amount of the desire at that time calculated by the arithmetic expression is ΔI [k],
Assuming that the parameter value of the current desire is I [k] and the coefficient k _i representing the sensitivity of the desire is a parameter value I [k +
1] is calculated, and the calculation result is replaced with the current parameter value I [k] of the desire to update the parameter value of the desire. Similarly, the instinct model 74 updates the parameter values of each desire except “appetite”.

[0063]

(Equation 2)

Note that the degree to which the recognition result and the notification from the output semantics converter module 68 affect the variation ΔI [k] of the parameter value of each desire is determined in advance. For example, the output semantics converter 68 The notification from the module 68 has a large effect on the variation ΔI [k] of the parameter value of “fatigue”.

In the present embodiment, the parameter values of each emotion and each desire (instinct) are regulated so as to fluctuate in the range of 0 to 100, respectively.
_e, the value of k _i is also set individually for each emotion and each desire.

On the other hand, as shown in FIG. 4, the output semantics converter module 68 of the middleware layer 40 provides “forward” and “pleasure” given from the action switching module 71 of the application layer 41 as described above. , "Sound" or "tracking (follow the ball)" is given to the corresponding signal processing modules 61 to 67 of the output system 69.

The signal processing modules 61 to 6
7, given the behavior command based on the action command, and servo command value to be applied to the actuator 25 1 _{to 25} n _(Fig. 2) corresponding to perform that action, the output from the speaker 24 (FIG. 2) Generating sound data of the sound to be played and / or driving data to be given to the LED of the “eye”,
These data are transferred to the virtual robot 33 of the robotic server object 32 and the signal processing circuit 14.
The corresponding actuators 25 ₁ to 25 ₁ through (FIG. 2)
_25n or the speaker 24 or the LED.

In this way, the robot apparatus 1 can perform autonomous actions according to its own (internal) and surrounding (external) conditions and instructions and actions from the user based on the control program. It has been made possible.

(3) Learning of Information of Robot Apparatus The above-described robot apparatus identifies a user to be interacted with, and learns an operation in which the identified user gives a high evaluation.
The user can be identified based on the learning result from the information input about the user, and the learned highly evaluated operation can be expressed according to the identified conversation partner.

More specifically, the robot device learns the face (image information) and voice (sound information) of the owner (user) as information input about the user. With respect to the user's face, captured image information input by the CCD camera 20 shown in FIG. 2 and signal-processed by the signal processing circuit 14 or the like is used. For the user's voice, voice information (or acoustic information) input by the microphone 23 shown in FIG. 2 and obtained by signal processing by the signal processing circuit 14 or the like is used. In addition, examples of the operation in which the user has a high evaluation include an operation praised by the user or an operation praised by the user.

The robot controls the operation by such learning. The robot device performs a user identification learning step of learning information input about the user, an operation learning step of learning a highly evaluated operation, and From the user based on the learning result in the user identification learning step, the operation having a high evaluation value learned in the operation learning step is
And an operation control step of displaying an operation according to the identified user.

The robot apparatus constructs and executes such processing steps as modules or objects. That is, the robot device includes a learning module or a learning object and an operation control module or an operation control object that are constructed by a program that realizes such processing steps. Further, for example, such a program is provided as being recorded on a disk-shaped recording medium or the like, or is distributed on the so-called Internet, and the robot apparatus takes in the program via such a medium, and Such processing is realized.

As described above, the robot apparatus learns user identification and motion, and examples of such a learning method include an RNN (Recurrent Neural Network) method. For example, Long Ji Li reported in his paper "Reinforcement Learning W.
ith Hidden States. "

FIG. 9 shows an example of learning by the RNN 100 and use of the learned RNN 100. (A) in FIG. 9 illustrates a use mode of the RNN 100 during learning, and (B) in FIG. 9 illustrates a use mode of the RNN 100 after the learning.

As shown in FIG. 9A, at the time of learning, learning information to be learned is input to the RNN 100. In response, the RNN 100 outputs corresponding data corresponding to the learning information. This correspondence data is grasped as so-called teaching data or target data.

The learning by the RNN 100 causes the corresponding data to correspond to the learning information. That is, after learning, as shown in FIG. 9B, by inputting certain information to the RNN 100 after learning, certain data is output, but the input information is learned in advance. If the information is learning information, corresponding data (teaching data or target data) corresponding to the learning information is output as an identical or approximate value. Therefore, by learning the RNN 100, for example, by inputting certain information to the RNN 100, it is possible to know the characteristics of the input information according to the output data (corresponding data).

FIG. 10 shows a configuration example of the RNN 100. As shown in this FIG. 10, RNN100 includes an input layer 100 _1, the intermediate layer 100 ₂ and the output layer 100 _3. The input layer 100 ₁ , the intermediate layer 100 _2, and the output layer 100 ₃ have a predetermined number of neurons, and each neuron in each layer is connected. Each neuron stores a predetermined weight coefficient.

[0078] In RNN100 configured in this manner, the learning information is input to the neurons of the input layer 100 _1. Learning information input to the input layer 100 _1, through the intermediate layer 100 ₂ is outputted from the output layer 100 _3. Data outputted from the output layer 100 ₃ is data corresponding to the learning information.

At the time of learning, each neuron multiplies an input by a weighting coefficient and outputs the result to another neuron connected to the layer on the output side. By the learning, the neuron outputs a predetermined value according to the information input to the RNN 100. Also, part of the output of a given neuron in the output layer 100 _3, context (cont
As _{ext) C t + 1,} is fed back to the input layer 100 ₁ neuron.

Thus, at the time of learning, the RNN 10
0 inputs learning information at the current time (step) t, predicts and outputs corresponding data as information at the next time t + 1, and outputs predetermined corresponding data in response to the input of learning information. Will be able to do it.

In the present embodiment, such learning is realized by the robot device, and the learning information is voice information (voice information) and face information (image information) of the user. Further, the data format of the input learning information and the output corresponding data is, specifically, a vector format.
For example, by treating as vector data, learning information and corresponding data can be specified as a feature amount in a feature space.

As described above, by learning using the RNN, after learning, information is input to the RNN, so that the RNN is used.
Output certain data based on learning.
That is, in response to input of certain information, the RNN outputs, as its output, data as certain vector data indicating its characteristics.

Further, in the embodiment, the RNN after learning is
And the information is obtained by the reverse operation. That is, using the RNN after learning, this time, certain information is specified by inputting data indicating such characteristics. Specifically, as shown in (A) of FIG.
As shown in the middle (B), a new RNN is generated, and learning information is specified from the corresponding data using the new RNN 110.

Here, the RNN 110 newly generated from the RNN 100 after learning is a so-called inverse RNN.
(Inverse-RNN, RNN ⁻¹ ) 110, which enables information to be specified using a so-called inverse-forward.

As described above, information can be learned by the RNN, and various information can be obtained by using the learned RNN. By applying such a learning method, the robot device identifies the face and voice of the user (owner) and realizes the performance when the user wants to be praised. The robot apparatus can identify and execute the performance obtained by learning when it is desired to be praised. Next, more specific processing in the robot apparatus for realizing such an operation will be described.

(3-1) Modeling and User Modeling by Learning
(3-1-1) Modeling of User by Learning FIG. 12 shows RNN 100 during learning, that is, learning information.
As information, sensor information Pt, voice information (information of the voice of the user)
Report) S_tAnd image information (user face information) F _tInput
And the corresponding output B_{t + 1}RNN100 to do
Is shown.

Here, the sensor input information _Pt is, for example,
This is conventional sensor input information, specifically, information on the external environment and information indicating the internal state of the robot apparatus 1. Such sensor input information _Pt and the above-described audio information S
_t and image information _{F t} is input to RNN100 as vector data as input sequence of time series. Also, data B _{t + 1} output for such an input
Are vector data {B (1),
B (2),..., B (n)}. Note that the description of the vector data as {B ₁ , B ₂ ,..., B _n } is based on the above {B (1), B (2),.
(N) It is the same as the description of｝.

[0088] RNN1 by learning entered such a sensor input information _{P t,} audio information _{S t} and the image information _{F t}
Data B as a feature sequence specified by 00
The characteristics of the user are determined based on _{t + 1} , and a modeled user (hereinafter, referred to as a user model) M _user is obtained. For example, the user model M _user is specified on the feature space by the vector data B _{t + 1} . That is, the sensor input information P _t forming information describing a specific _user, by learning audio information S _t and the image information F _t, the user model M _{use r} as an index indicating characteristics of the user are obtained.

(3-1-2) Identification of User Using Learning Model As described above, the user can be modeled as data using the RNN 100. In the identification of a user using the RNN 100, the user is identified using a user model M _user specified by learning. FIG. 13 shows recognition of the user performed using the RNN 100 after learning the user and the user model M _user . Specifically, the user is identified as follows.

The information similar to the above-described learning information input format at the time of learning is input to the RNN 100 after learning as the information of the user to be identified. That is, RNN
The input to 100 is sensor input information P _t , voice information _St
And an image information _{F t.} Then, the RNN 100 outputs data Bt _{+ 1} in response to such an input. Here, the data format of the data _{Bt + 1 is the same as} the data format at the time of learning, and the vector data {B (1), B (2),
.., B (n)}.

The data B thus obtained is
_{t + 1} and user model M composed of vector data
Compare and compare with _user to get difference. Based on the difference, the input information identifies the user. That is, when the difference Err is equal to or smaller than the predetermined threshold α, it is determined that the user has learned in advance,
On the other hand, when the difference Err is larger than the predetermined threshold α, it is determined that the user is different from the user who has learned in advance, and the user is identified based on the input information.

As described above, the information about the user used for user identification is learned by the RNN, and the user model M corresponding to the user obtained as the learning result is obtained.
Based on _{us er,} it is possible to identify the user.

(3-2) Learning of Actions to be Praised and Expression of Learned Actions (3-2-1) Learning of Behaviors to Be Praised Learning of actions to be expressed when the user is praised will be described. FIG. 14 shows an RNN 100 at the time of learning, that is, a performance sequence A _t as learning information.
RNN1 correspondingly outputs its output B _{t + 1}
00 is shown.

Here, the performance sequence A
_t is associated with a predetermined action expressed by the robot device. Actions are associated with the performance sequence A _t is the action when the compliments the user. For example, the action to be associated with a performance sequence A _t, operation or attitude such as the so-called "Sitting" and the like. The performance sequence A
_t is vector data {A (1), A (2),.
A (n)}.

[0095] at the time of learning, performance sequence _{A t}
Is input to the RNN 100, and the data B
_{t + 1} is output. Here, data _{Bt + 1} is sequence data, and vector data {B (1), B
(2),..., B (n)}.

Then, the performance sequence A
_t is input to the RNN 100 and learned assuming that the data indicates the behavior when praised, the output data B _{t + 1} is the feature data (or the characteristic data indicating the characteristic of the behavior when praised) Characteristic sequence).

(3-2-2) Expression of Behavior Learned Using Inverse RNN As described above, the behavior when praised can be learned by the RNN. Then, the robot device can learn the behavior learned in this manner and express it using the RNN. Specifically, an action is expressed using the RNN after learning as follows.

First, as described with reference to FIG. 11, the RNN 10 that has learned the action to be expressed when a praise is desired.
From 0, Inverse-Forward
It is possible to specify the action sequence using
So-called Inverse-RNN (Inverse-RNN, RNN)
^-1 ) 110 is generated.

[0099] Figure 15 is a sequence _{B t + 1} is the input indicates the inverse RNN110 outputting performance sequence _{A t} correspond. Here, the performance sequence _{A t} is the vector data {A outputted in response to an input sequence _{B t + 1 (1),} A
, A (n)}. Note that the description of vector data as {A ₁ , A ₂ ,..., A _n } is based on the above-described {A (1), A (2),.
(N) It is the same as the description of｝.

[0100] Accordingly, by inputting the sequence _{B t + 1} to the inverse RNN110, performance sequence _{A t} which is data indicating an operation to be compliments is to be outputted. Then, based on such performance sequence A _t, robotic device to actually expose the operation praised the user such association has been that "Sitting", and the like.

As described above, the robot apparatus learns the RNN at the timing when the user wants to be praised.
Inverse RNN 110 obtained from 100, RNN 10
By inputting the sequence _{Bt + 1} obtained at the time of learning using 0, an operation praised by the user can be expressed.

(3-3) Identification of Performance Sequence when Prescribed User is Praised Here, an outline of a processing procedure for identifying a prescribed user and expressing an operation to be praised will be described. I have. FIG.
Shows the procedure.

First, as step S1, learning by an RNN for identifying a predetermined user is performed in advance. Learning for identifying a predetermined user is realized by the procedure described with reference to FIG. That is, by learning here, the user model M corresponding to the predetermined user
You can get a _user .

Then, as step S2, a process for identifying a user is performed based on the user model M _user . That is, this is the process described with reference to FIG. 13, in which the information of the human being to be identified is input to the RNN after learning, and the output from the RNN obtained thereby is compared with the user model M _user. To obtain the difference Err.
If the difference Err is equal to or smaller than a predetermined threshold, it is determined that the user is a predetermined user learned by the RNN. If the difference Err exceeds the predetermined threshold, the user is not a learned user. Judge. Alternatively, in the example feature space, if the output from the RNN is a value that satisfies predetermined requirements for the user model M _user grasped as vector data, it may be specified as a predetermined user.

On the other hand, in step S3, when the praise is specified, learning by the RNN is performed. The learning of the action to be praised is realized by the procedure described with reference to FIG. Then, as described with reference to FIG. 15, an inverse RNN is generated from the RNN after learning,
In step S4, a sequence _{Bt + 1} for expressing an operation is input to the inverse RNN. In particular,
The sequence _{Bt + 1} corresponding to the user identified in step S2 is input. For example, in the robot apparatus, a user is associated with a sequence _{Bt + 1} and stored as a table, and by referring to such a table, a sequence _{Bt + 1} corresponding to the identified user is specified. As shown in FIG. 17, associating such a user with the sequence _{Bt + 1} , as shown in FIG. 17, a user feature sequence indicating the feature of the user as an output of the RNN at the time of learning for identifying the user, and a praised operation Is associated with an operation feature sequence indicating the feature of the operation as an output of the RNN at the time of learning.

In step S5, the sequence B _{t + 1} specified as being associated with the identified user is
By input to the inverse RNN, identifies the predetermined performance sequence A _t. And the robot device is
Such a predetermined operation corresponding to the performance sequence A _t of "Sitting" and the like behavior (Action) to expose the.

The above-described processing is executed when the robot apparatus wants to be praised. For example, the timing at which the processing for displaying such an operation to be praised is started. Is realized by having a model (for example, a module or an object) capable of obtaining such timing. That is, for example, as described above, the robot device has a model that defines the timing of being praised as a model for increasing motivation, and for expressing an operation to be praised when this model reaches a predetermined level. Execute the process.

As described above, the robot device recognizes the face and voice of the user (owner), and learns the operation (performance) to appear when the owner wants to praise, based on the RNN method. Then, when the owner wants to be praised by the owner, the performance is executed using the inverse RNN. Thereby, for example, the user is delighted to see the operation that the robot device is trying to pay attention to to be praised. Will also be higher. As described above, the robot device has the emotion model whose character changes in the dialogue of the user or the like, and can affect the character of the operation result to be praised, thereby further increasing the entertainment. In this way, the robot device has an improved learning ability, and further has an improved entertainment.

(3-4) Specific Example of Learning by RNN For example, input data on RNN (here,
_Pt , _Ft , _St , _Ct , θ) and output data (predicted value)
The relationship with B _{t + 1} can be expressed as in equation (3). That it can be shown by a function _{R NN} as shown in the RNN (3) expression. Where B _{t + 1} , P
_t and θ are shown as Expressions (4) to (6), respectively. Θ is a bias value of a unit (neuron) constituting the RNN.

[0110]

(Equation 3)

[0111]

(Equation 4)

[0112]

(Equation 5)

[0113]

(Equation 6)

As described above, the output data B _{t + 1} is a function R with the input data P _t , F _t , S _t , C _t , and θ as variables.
It can be seen that it can be obtained by _NN . here,
I make the following hypothesis. For example, a hypothesis is made using equation (7). Equation (7) is a so-called membership function or sigmoid function.

[0115]

(Equation 7)

Here, as shown in FIG. 18, consider a unit j which is grasped as a neuron constituting the RNN. This unit has n _{j, 1} (t), ...,
n _{j , Nj} (t) are input, and the unit _{responds to} this input by using o _{j, 1} (t),.
_{j, Lj} (t) is output. Such a relationship is (8)
It is shown as an equation. In equation (8), W _1,1 ,.
, W _{Nj and Lj} are weighting factors. And (8)
The equation can also be expressed as equation (9).

[0117]

(Equation 8)

[0118]

(Equation 9)

Here, it is assumed that the left side of equation (9) is the left side as in equation (10). That is, it is assumed that the output value o _{j, i} from the unit is given as a function of the data t, P _t , F _t , S _t , and C _t that are the actual inputs to the RNN. Then, Expression (11) is obtained from the relationship between Expression (10) and Expression (7). Equation (11) defines an RNN that outputs predetermined data in response to an input.

[0120]

(Equation 10)

[0121]

[Equation 11]

Next, RN grasped by the above equation
The learning using N will be described using mathematical expressions. The input data to the RNN can be represented as vector data (or sequence data) as shown in equation (12).

[0123]

(Equation 12)

For example, the elements A ₁ , A
_{2, ··· A t, ···,} A n , respectively, as shown in (13), a value specified by the speech information _{S t} and the image information _{F t.} Note that the original audio information _St and image information _Ft are data as feature amounts.

[0125]

(Equation 13)

In this case, the vector data B _{t + 1} is output in response to the input to the RNN. The source of the vector data B _{t + 1} is also expressed by the audio information _{St + 1} and the image data B _{t + 1} as shown in the equation (14). The value specifies information _{Ft + 1} .

[0127]

[Equation 14]

FIG. 19 shows the input of data {A ₁ , A ₂ ,..., A _n } to the RNN during learning and the corresponding output data {B ₁ , B ₂ ,. B _n }. For example, in the case of learning the user's voice and face, FIG as shown in 20, when the vector data {A 1 to the original image information F _t and audio information S _t which is input as a series _data, A _2, .., A _n } are input to the RNN.

[0129] Here, the image information F _t constituting the original, and detecting the feature quantity from one single continuous captured images obtained by capturing, the characteristic amount is an image information F _t. Also, the audio information S _t, from a continuous speech signal obtained by the microphone or the like, for example, by detecting a feature from each by sampling by a predetermined interval, the feature amount is the audio information S _t. Thus, for example, vector data ｛A composed of a time series of data based on the feature amount of the voice at each time point such as “sitting” and the feature amount at each time point of the user's face changing at that time. _{1, a} 2, can be obtained., the _{a n}.} Thus, based on _{A 1, A} 2 of the vector data _{A t,} · ·
·, A _n are each a data indicating a feature of each time point of the image information F _t and audio information S _t as time-series data at the time of learning. Such a vector data A _t
As input, if learned by RNN, function _{R NN}
With the relationship between the input data _{A t} and the output data _{B t + 1} can be expressed as (15).

[0130]

(Equation 15)

[0131] Here, the W and theta are the parameters of the function R _NN, W is a weighting factor RNN of neurons (units) holds, theta, such bias value of the neuron (units) It is. Data _{A t}
Is learned, such weighting factor W and bias value θ are determined as weighting factor W ^* and bias value θ ^* . Here, the weighting coefficient W ^* and the bias value θ ^* obtained by learning are represented by vector data {A ₁ , A ₂ ,.
, A _n }, and the weight coefficient W ^* and the bias value θ ^* can be given by the equations (16) and (17).

[0132]

(Equation 16)

[0133]

[Equation 17]

That is, for example, when the user issues “sitting”, the function _RNN can be obtained as equation (18) using the weight coefficient W ^* and the bias value θ ^* as variables.

[0135]

(Equation 18)

Then, the RNN trained in this way is
To generate the inverse RNN. This inverse RN
By using N (function R _NN ⁻¹ ), as shown in FIG. 21, data {B ₁ ′, B ₂ ′,.
By inputting B _n ′, data {A ′ ₁ , A ₂ ′,..., A _n ′} input during learning is output.

[0137] For example, the inverse RNN (function _{R NN}
⁻¹ ), the input value B _t ′, and the output value A _t ′ corresponding to the input value B _t ′ can be expressed as in equation (19).

[0138]

[Equation 19]

[0139] Here, 'and the value _{B t} inputted' values _{A t} to be outputted from, for example, (20) as in equation and equation (21), the user's voice data _{(S t} or _{S t} ') Or face data ( _Ft or _Ft ').

[0140]

(Equation 20)

[0141]

(Equation 21)

As described above, the learning process by the RNN uses the mathematical formula.
And explained. By such a learning process, the user
"Sitting" from the data @ A₁, A₂, ...,
A_nTime series of learning as an input of 、 and its output
Data No. B₁, B₂, ..., B _n学習, after learning
Input to the inverse RNN obtained from the RNN after learning.
By inputting the data corresponding to "sitting"
｛A₁, A₂, ..., A _nSo that you can get
become.

The data {A}₁, A₂,
... A_n対 応 付 け is associated with the execution of the “sitting” operation
By doing, the data to inverse RNN after learning
｛B ₁, B₂, ..., B_nRobot input by inputting｝
The device now displays the actual `` sitting '' behavior
You.

Further, the data {B ₁ , B
_2, ..., data _{A 1 obtained from _{B n}, A} 2, · ·
By setting the timing at which the action of “sitting” is expressed based on A _n } as the timing at which the user wants to be praised, the robot apparatus can perform the action of “sit down” when the user wants to be praised. Will be expressed.

By having the learning model constructed by the above mathematical formulas, the robot device can identify a predetermined user by learning, and can express a praised operation.

[0146] In the above-described embodiment, a case has been described where the conversation partner is a user assumed to be a human. However, the present invention is not limited to this. For example, the conversation partner may be a robot device. In this case, the robot device is identified by learning, and an operation corresponding to the identified robot device is expressed.

Further, in the above-described embodiment, information input about a conversation partner has been described as audio information or image information. However, the present invention is not limited to this, and other information that can identify the conversation partner can be used as information for the robot apparatus to identify the conversation partner.

[0148]

According to the motion control method for a robot apparatus according to the present invention, a learning step of causing a robot apparatus to learn an operation to evaluate a conversation partner with a high evaluation by identifying the conversation partner, and to input the conversation partner. An operation control step of identifying a conversation partner based on the learning result in the learning step from the obtained information, and expressing a learned high evaluation operation by the robot apparatus according to the identified conversation partner. The robot device is made to learn an operation in which the other party is identified and the identified dialogue partner gives a high evaluation, and the dialogue partner is identified based on the learning result from the information input with respect to the dialogue partner. The movement can be expressed by the robot device according to the identified conversation partner.

According to the operation control method of the robot device, the robot device learns the conversation partner and the operation that the conversation partner evaluates highly, and based on the learning result,
Highly evaluated actions can be expressed according to the conversation partner.

Further, the program according to the present invention includes a learning step of causing a robot apparatus to learn an operation in which a dialogue partner is identified and the identified dialogue partner gives a high evaluation, and a learning step based on information input with respect to the dialogue partner. A conversation control unit that identifies the conversation partner based on the learning result in the process, and causes the robot apparatus to express the learned highly evaluated operation by the robot device according to the identified conversation partner. Then, the robot device learns the operation that the identified conversation partner evaluates high, and identifies the conversation partner based on the learning result from the information input about the conversation partner, and identifies the learned operation with high evaluation. It can be displayed by the robot device according to the conversation partner.

A robot apparatus whose operation is controlled by such a program learns a conversation partner and an operation that the conversation partner evaluates highly, and, based on the learning result, performs a movement with a high evaluation according to the conversation partner. Can be revealed.

Further, the recording medium according to the present invention includes a learning step of causing the robot apparatus to learn an operation in which a dialogue partner is evaluated highly by identifying the dialogue partner, and a learning step in which information is input with respect to the dialogue partner. A program is recorded for executing an operation control step of identifying a conversation partner based on a learning result in the learning step and expressing a learned highly evaluated operation by a robot device according to the identified conversation partner. According to this, the robot device is made to learn the operation in which the conversation partner is identified, and the identified conversation partner gives a high evaluation, and the conversation partner is identified based on the learning result from the information input with respect to the conversation partner, and the learning is performed. A highly evaluated operation can be expressed by the robot apparatus according to the identified conversation partner.

A robot apparatus whose operation is controlled by a program recorded on such a recording medium learns a conversation partner and an operation that the conversation partner gives a high evaluation, and based on the learning result, gives the conversation partner. Accordingly, a highly evaluated operation can be expressed.

Further, the robot apparatus according to the present invention identifies the conversation partner, and learns the operation that the identified conversation partner gives a high evaluation, and the learning result of the learning unit from the information input with respect to the conversation partner. And an operation control means for displaying a learned high evaluation operation in accordance with the identified dialogue partner, thereby identifying the dialogue partner and increasing the identified dialogue partner. The operation to be evaluated is learned, the conversation partner is identified based on the learning result from the information input with respect to the conversation partner, and the learned high evaluation operation can be expressed according to the identified conversation partner. Therefore, the robot apparatus can learn the conversation partner and the operation that the conversation partner evaluates high, and can express the highly evaluated operation according to the conversation partner based on the learning result.

[Brief description of the drawings]

FIG. 1 is a perspective view illustrating an external configuration of a robot device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a circuit configuration of the robot device described above.

FIG. 3 is a block diagram illustrating a software configuration of the robot device described above.

FIG. 4 is a block diagram illustrating a configuration of a middleware layer in a software configuration of the robot device described above.

FIG. 5 is a block diagram showing a configuration of an application layer in the software configuration of the robot device described above.

FIG. 6 is a block diagram showing a configuration of the behavior model library of the application layer.

FIG. 7 is a diagram used to explain a finite probability automaton that is information for determining an action of the robot apparatus.

FIG. 8 is a diagram showing a state transition table prepared for each node of the finite probability automaton.

FIG. 9 is a diagram used for explaining information learning by the RNN.

FIG. 10 is a diagram showing a specific example of the structure of an RNN.

FIG. 11 shows an RNN to an inverse RNN (RNN ⁻¹ ).
FIG. 7 is a diagram used for explaining learning performed by generating a.

FIG. 12 is a diagram used to explain learning of user information by the RNN.

FIG. 13 is a diagram used for describing user identification using an RNN after learning.

FIG. 14 is a diagram used for explaining learning of a motion to be praised.

FIG. 15 is a diagram used to explain that an operation praised using the inverse RNN is expressed.

FIG. 16 is a flowchart illustrating a processing procedure for identifying a user and expressing a praised operation.

FIG. 17 is a diagram used to explain an example of association between a user feature sequence obtained by learning user identification and an operation feature sequence obtained by learning operation.

FIG. 18 shows a unit (neuron) of RNN,
It is a figure which shows an input and its output.

FIG. 19 is a diagram used for describing RNN expressed by a mathematical expression.

FIG. 20 is a diagram used to explain vector data as user information input to the RNN during learning.

FIG. 21 is a diagram used for describing an inverse RNN in a mathematical expression.

[Explanation of symbols]

 1 robot device, 10 CPU

 ──────────────────────────────────────────────────の Continued on the front page F term (reference) 2C150 CA02 DA05 DA24 DA25 DA26 DA27 DA28 DF03 DF04 DF06 DF33 ED42 ED52 EF07 EF16 EF23 EF29 EF33 EF36 3F059 AA00 BA00 BB06 DB04 DC00 DC01 FC15 3F060 AA00 BA10 CA14

Claims (16)

[Claims] A learning step of causing a robot apparatus to learn an operation in which the conversation partner is highly evaluated by identifying the conversation partner, based on the learning result in the learning step from information input with respect to the conversation partner. An operation control step of causing the robot apparatus to express a learned high evaluation operation by the robot apparatus according to the identified dialogue partner. 2. The method according to claim 1, wherein the learning step includes a dialogue partner learning step of causing the robot apparatus to learn information input with respect to the dialogue partner, and an operation learning step of learning the robot apparatus with a highly evaluated operation. In the operation control step, the conversation partner is identified based on the learning result in the conversation partner learning step from the information input with respect to the conversation partner, and the operation having a high evaluation value learned in the operation learning step is identified. 2. The operation control method for a robot device according to claim 1, wherein the robot device is displayed in accordance with the condition. 3. In the operation learning step, the robot device learns the operation by an RNN (recurrent neural network). In the operation control step, an inverse RNN (RNN ⁻¹ ) having an inverse functional relationship of the learned RNN. 3. The operation control method for a robot device according to claim 2, wherein the operation is expressed by the robot device based on the following. 4. The operation control method for a robot apparatus according to claim 1, wherein the input information on the conversation partner is face information and voice information of the conversation partner. 5. The operation control method for a robot device according to claim 1, wherein the highly evaluated operation is an operation in which the robot device is praised by a conversation partner. 6. The method according to claim 1, wherein said robot device has an appearance simulating an animal. 7. The method according to claim 1, wherein the conversation partner is a human. 8. A learning step in which the robot apparatus learns an operation in which the conversation partner is identified by giving a high evaluation to the conversation partner, based on the learning result in the learning step based on information input about the conversation partner. And a motion control step of causing the robot apparatus to express a learned operation with a high evaluation according to the identified dialogue partner. 9. A learning step for causing a robot device to learn an operation in which a conversation partner is identified and the identified conversation partner evaluates high, based on a learning result in the learning step from information input about the conversation partner. And a motion control step of causing the robot apparatus to display a learned high-reputation operation by the robot apparatus according to the identified conversation partner. 10. A learning means for identifying a conversation partner and learning an operation in which the identified conversation partner gives a high evaluation, and identifying the conversation partner based on the learning result of the learning means from information inputted about the conversation partner. And a motion control means for displaying the learned highly evaluated operation in accordance with the identified conversation partner. 11. The learning means includes a dialogue partner learning means for learning information inputted with respect to a dialogue partner, and an operation learning means for learning an operation having a high evaluation. Identifying the conversation partner based on the learning result of the conversation partner learning unit from the input information, and expressing a highly evaluated operation learned by the operation learning unit according to the identified conversation partner. The robot device according to claim 10, wherein 12. The operation learning means learns an operation by an RNN (Recurrent Neural Network), and the operation control means operates based on an inverse RNN (RNN ^-1 ) which is an inverse functional relationship of the learned RNN. The robot device according to claim 10, wherein the operation is expressed. 13. The robot apparatus according to claim 10, wherein the input information relating to the conversation partner is face information and voice information of the conversation partner. 14. The robot apparatus according to claim 10, wherein the operation having a high evaluation is an operation in which the robot apparatus is praised by a conversation partner. 15. The robot device according to claim 11, wherein the robot device has an appearance shape imitating an animal. 16. The robot apparatus according to claim 11, wherein the conversation partner is a human.