JP7425681B2 - Social ability generation device, social ability generation method, and communication robot - Google Patents

Description

The present invention relates to a social ability generation device, a social ability generation method, and a communication robot.

Today, smart speakers and communication robots are being developed. Such systems focus on functions such as turning lights on or off, accessing calendars, reading email, and setting appointments based on commands. In such a system, the input of instructions is limited to, for example, selection using a touch panel or predetermined voice commands, making it difficult to build relationships with people.

For this reason, a system that allows for relationships with people is desired. For example, Patent Document 1 proposes a system in which a companion device interacts with a person and a person is involved in interaction and operation with the device. In the technology described in Patent Document 1, a companion device detects the user's utterances and actions, expresses them through movement, graphics, sound, light, and fragrance, and provides a friendly presence.

Special Publication No. 2019-521449

However, with the technology described in Patent Document 1, it is difficult to form an emotional connection between a robot and a human.

The present invention has been made in view of the above problems, and provides a social ability generation device, a social ability generation method, and a communication robot that can form an emotional connection between a robot and a person. The purpose is to provide

(1) In order to achieve the above object, a social ability generation device according to one aspect of the present invention acquires person information about a person, extracts characteristic information about the person from the acquired person information, and performs communication. Multimodal emotional interaction between people using cognitive means that recognizes the interactions that occur between devices and people, and the extracted characteristic information about the person. and a behavior generating means that generates a behavior based on the learned emotional interaction information of the person.

(2) Furthermore, in the social ability generation device according to one aspect of the present invention, the learning means performs learning using an implicit reward and an explicit reward, and the implicit reward is The reward is multimodally learned using feature information about the person, and the explicit reward is a reward based on the result of evaluating the behavior of the communication device toward the person generated by the behavior generation means. You can do it like this.

(3) Furthermore, the social ability generation device according to one aspect of the present invention includes a sound collection section that picks up an acoustic signal, and a shooting section that shoots an image including the user, and the recognition means includes: Speech recognition processing is performed on the collected acoustic signal to extract feature information related to the sound, image processing is performed on the captured image to extract feature information related to human behavior included in the image, and the The feature information about the person includes feature information about the voice and feature information about the human behavior, and the feature information about the voice includes the voice signal, voice volume information, voice intonation information, and the meaning of the utterance. The characteristic information regarding human behavior is at least one of the following: facial expression information, gesture information performed by the person, head posture information, face orientation information, line of sight information, and communication between people. It may be at least one of the distances between.

(4) Furthermore, in the social ability generation device according to one aspect of the present invention, the learning means performs learning using social norms, social constituent elements, psychological knowledge, and humanities knowledge. It's okay.

(5) In order to achieve the above object, a social ability generation method according to one aspect of the present invention includes a recognition unit that acquires person information about a person, extracts characteristic information about the person from the acquired person information, The learning means recognizes the interactions that occur between a person and a communication device that communicates, recognizes the interactions that occur between people, and uses the extracted characteristic information about the person to learn about the person's emotional state. The interaction is multimodally learned, and the behavior generation means generates the behavior based on the learned emotional interaction information of the person.

(6) In order to achieve the above object, a communication robot according to one aspect of the present invention acquires human information about a person, extracts characteristic information about the person from the acquired human information, and connects a communication device and a person to communicate. Multimodally learns emotional interactions between people using a cognitive means that recognizes the interactions that occur between people and the extracted characteristic information about the person. The device includes a learning device and a behavior generating device that generates a behavior based on the learned emotional interaction information of the person.

(7) Furthermore, the communication robot according to one aspect of the present invention includes a display unit, and the behavior generating means improves the relationship with the person by causing the person to behave in a manner that maximizes positive emotions. An image that maintains the state may be generated, and the generated image may be displayed on the display unit.

According to (1) to (7), an emotional connection can be formed between robots and humans.
According to (2), learning can be performed without using much teaching data.
According to (3), a lot of information can be obtained based on people's reactions.
According to (4), they are socially intelligent and able to navigate social scenarios.
According to (7), it is possible to make people behave in a manner that maximizes positive emotions, and it is possible to maintain good relationships with people.

FIG. 3 is a diagram illustrating an example of communication of a communication robot according to an embodiment. FIG. 1 is a block diagram showing a configuration example of a communication robot according to an embodiment. 1 is a diagram showing an example of the external shape of a communication robot according to an embodiment. FIG. 3 is a diagram showing the flow of recognition, learning, and social ability performed by the social ability generation device of the embodiment. FIG. 6 is a diagram showing an example of data recognized by the recognition unit according to the embodiment. It is a diagram showing an example of a system that performs deep reinforcement learning using raw data in a comparative example. FIG. 3 is a diagram illustrating an example of an agent creation method used by the behavior generation unit according to the embodiment. 2 is a flowchart illustrating an example of a procedure for social ability generation processing according to an embodiment. FIG. 3 is a diagram illustrating an example of communication between a communication robot and a user according to an embodiment. FIG. 3 is a diagram illustrating an example of an image displayed on a display unit during communication between a communication robot and a user according to an embodiment. FIG. 3 is a diagram illustrating an example of an image displayed on a display unit during communication between a communication robot and a user according to an embodiment. FIG. 3 is a diagram illustrating an example of communication by a communication robot when communicating with a user's friend according to the embodiment. FIG. 2 is a diagram showing an example in which the communication robot of the embodiment is applied to a car navigation system in a vehicle. FIG. 2 is a diagram illustrating an example of connections with various devices in the home when applied to a car navigation according to an embodiment.

Embodiments of the present invention will be described below with reference to the drawings. Note that in the drawings used in the following explanation, the scale of each member is changed as appropriate in order to make each member a recognizable size.

<Summary>
FIG. 1 is a diagram showing an example of communication by the communication robot 1 according to the present embodiment. As shown in FIG. 1, a communication robot 1 communicates with an individual or multiple people 2. Communication mainly consists of dialogue g11 and gestures g12 (actions). The motion is expressed by an image displayed on the display unit in addition to the actual motion. Furthermore, when an e-mail is sent to the user via an Internet line or the like, the communication robot 1 receives the e-mail and notifies the user that the e-mail has arrived and the content (g14). Furthermore, for example, when a reply to an e-mail is required, the communication robot 1 communicates with the user to see if advice is needed and makes a proposal g14. The communication robot 1 sends a reply (g15). Furthermore, the communication robot 1 presents a weather forecast g19 for a location according to the scheduled date and time and location, for example, in accordance with the user's schedule.

In this embodiment, the social ability of the robot is generated so that an emotional connection can be formed between the robot and the human, and the robot communicates with the human based on, for example, the human's reactions and actions. In this embodiment, humans and robots communicate with each other by empathizing with each other on an emotional level. In this embodiment, so to speak, communication between humans and pets is realized by learning social norms and the like. In this embodiment, the above is achieved by learning users' social backgrounds, interactions between people, etc. in communication.

<Example of configuration of communication robot 1>
Next, a configuration example of the communication robot 1 will be explained.
FIG. 2 is a block diagram showing a configuration example of the communication robot 1 according to this embodiment. As shown in FIG. 2, the communication robot 1 includes a receiving section 101, a photographing section 102, a sound collecting section 103, a sensor 104, a social ability generation device 100, a storage section 106, a first database 107, a second database 109, and a display section. 111, a speaker 112, an actuator 113, and a transmitter 114.

The social ability generation device 100 includes a recognition section 105 (recognition means), a learning section 108 (learning means), and a motion generation section 110 (motion generation means).
The motion generation section 110 includes an image generation section 1101, an audio generation section 1102, a drive section 1103, and a transmission information generation section 1104.

<Function and operation of communication robot 1>
Next, the functions and operations of each functional section of the communication robot 1 will be explained with reference to FIG. 1.

The receiving unit 101 acquires information (eg, e-mail, blog information, news, weather forecast, etc.) from, for example, the Internet via a network, and outputs the acquired information to the recognition unit 105 and the motion generation unit 110. Alternatively, for example, when the first database 107 is on the cloud, the receiving unit 101 acquires information from the first database 107 on the cloud and outputs the acquired information to the recognition unit 105.

The imaging unit 102 is, for example, a CMOS (Complementary Metal Oxide Semiconductor) imaging device, a CCD (Charge Coupled Device) imaging device, or the like. The photographing unit 102 outputs a photographed image (person information that is information about a person; a still image, a continuous still image, a moving image) to the recognition unit 105 and the motion generation unit 110. Note that the communication robot 1 may include a plurality of imaging units 102. In this case, the imaging unit 102 may be attached to the front and rear of the housing of the communication robot 1, for example.

The sound collection unit 103 is, for example, a microphone array composed of a plurality of microphones. The sound collection unit 103 outputs the acoustic signals (person information) collected by the plurality of microphones to the recognition unit 105 and the motion generation unit 110. Note that the sound collection unit 103 may output each of the acoustic signals picked up by the microphone to the recognition unit 105 after being sampled with the same sampling signal and converting the analog signal into a digital signal.

The sensor 104 is, for example, a temperature sensor that detects the temperature of the environment, an illuminance sensor that detects the illuminance of the environment, a gyro sensor that detects the inclination of the housing of the communication robot 1, and an acceleration sensor that detects the movement of the housing of the communication robot 1. , an atmospheric pressure sensor that detects atmospheric pressure, etc. The sensor 104 outputs the detected value to the recognition unit 105 and the motion generation unit 110.

The storage unit 106 stores, for example, items to be recognized by the recognition unit 105, various values (threshold values, constants) used during recognition, algorithms for recognition, and the like.

The first database 107 stores, for example, a language model database, an acoustic model database, a dialogue corpus database, and acoustic features used in speech recognition, a comparison image database and image features used in image recognition, and the like. do. Note that each data and feature amount will be described later. Note that the first database 107 may be placed on a cloud or may be connected via a network.

The second database 109 stores data related to relationships between people, such as social components, social norms, social customs, psychology, humanities, etc., which are used during learning. Note that the second database 109 may be placed on the cloud or may be connected via a network.

The social ability generation device 100 recognizes the interaction that occurs between the communication robot 1 and a person or the interaction that occurs between multiple people, and based on the recognized content and the data stored in the second database 109, the social ability generation device 100 generates human emotions. learning interactive interactions. Then, the social ability generation device 100 generates social abilities of the communication robot 1 from the learned content. Note that social ability is, for example, the ability to perform interactions between people, such as dialogue, action, understanding, and empathy between people.

The recognition unit 105 recognizes the interaction that occurs between the communication robot 1 and a person, or the interaction that occurs between multiple people. The recognition unit 105 acquires an image photographed by the photographing unit 102, an acoustic signal collected by the sound collection unit 103, and a detection value detected by the sensor 104. Note that the recognition unit 105 may acquire the information received by the receiving unit 101. The recognition unit 105 recognizes the interaction between the communication robot 1 and a person, or the interaction between multiple people, based on the acquired information and the data stored in the first database 107. Note that the recognition method will be described later. The recognition unit 105 outputs the recognized recognition results (features related to sounds, feature information related to human behavior) to the learning unit 108 . Note that the recognition unit 105 performs known image processing (for example, binarization processing, edge detection processing, clustering processing, image feature quantity extraction processing, etc.) on the image photographed by the photographing unit 102. The recognition unit 105 performs well-known speech recognition processing (sound source identification processing, sound source localization processing, noise suppression processing, speech interval detection processing, sound source extraction processing, acoustic feature amount calculation processing, etc.) on the acquired acoustic signal. Based on the recognized result, the recognition unit 105 extracts the audio signal (or audio signal) of the target person, animal, or object from the acquired audio signal, and recognizes the extracted audio signal (or audio signal). The result is output to the motion generation section 110. The recognition unit 105 extracts an image of the target person or object from the acquired image based on the recognition result, and outputs the extracted image to the motion generation unit 110 as a recognition result.

The learning unit 108 uses the recognition results output by the recognition unit 105 and the data stored in the second database 109 to learn human emotional interactions. The learning unit 108 stores the model generated through learning. Note that the learning method will be described later.

The motion generation section 110 acquires information received from the reception section 101 , an image photographed from the photographing section 102 , an acoustic signal collected from the sound collection section 103 , and a recognition result from the recognition section 105 . The behavior generation unit 110 generates behavior (utterances, gestures, images) for the user based on the learned results and the acquired information.

The image generation unit 1101 generates an output image (still image, continuous still image, or video) to be displayed on the display unit 111 based on the learned result and the acquired information, and displays the generated output image. It is displayed on the display unit 111. As a result, the motion generation unit 110 causes the display unit 111 to display an animation such as a facial expression, presents an image to be presented to the user, and communicates with the user. The displayed images include images that correspond to the movements of a person's eyes, images that correspond to the movements of a person's mouth, and information such as the user's destination (maps, weather maps, weather forecasts, information on shops and recreational areas). etc.), an image of a person who has made a video call to the user via the Internet line, etc.

The audio generation unit 1102 generates an output audio signal to be output to the speaker 112 based on the learned result and the acquired information, and causes the speaker 112 to output the generated output audio signal. Thereby, the motion generation unit 110 outputs an audio signal from the speaker 112 to communicate with the user. The output audio signal is an audio signal from a voice assigned to the communication robot 1, an audio signal from a person who has made a video call to the user via the Internet line, or the like.

The drive unit 1103 generates a drive signal for driving the actuator 113 based on the learned result and the acquired information, and drives the actuator 113 with the generated drive signal. Thereby, the motion generation unit 110 controls the motion of the communication robot 1 to express emotions, etc., and communicate with the user.

Based on the learned results and the acquired information, the transmission information generation unit 1104 generates transmission information (audio signal, image ) is generated, and the generated transmission information is transmitted from the transmitter 114.

The display unit 111 is a liquid crystal image display device, an organic EL (Electro Luminescence) image display device, or the like. The display unit 111 displays an output image output by the image generation unit 1101 of the social ability generation device 100.

The speaker 112 outputs an output audio signal output by the audio generation unit 1102 of the social ability generation device 100.

The actuator 113 drives the operating unit according to a drive signal output by the drive unit 1103 of the social ability generation device 100.

The transmission unit 114 transmits the transmission information output by the transmission information generation unit 1104 of the social ability generation device 100 to a destination via the network.

<Example of external shape of communication robot 1>
Next, an example of the external shape of the communication robot 1 will be explained.
FIG. 3 is a diagram showing an example of the external shape of the communication robot 1 according to the present embodiment. In the example of the front view g101 and side view g102 of FIG. 3, the communication robot 1 includes three display units 111 (111a, 111b, 111c). Further, in the example of FIG. 3, the photographing section 102a is attached to the top of the display section 111a, and the photographing section 102b is attached to the top of the display section 111b. The display units 111a and 111b correspond to human eyes and present image information. The speaker 112 is attached to the housing 120 near a display section 111c that displays an image corresponding to a human mouth. The sound collection section 103 is attached to the housing 120.

The communication robot 1 also includes a boom 121. The boom 121 is movably attached to the housing 120 via a movable part 131. A horizontal bar 122 is rotatably attached to the boom 121 via a movable part 132.
Furthermore, the display section 111a is rotatably attached to the horizontal bar 122 via a movable section 133, and the display section 111b is rotatably attached via a movable section 134.
Note that the external shape of the communication robot 1 shown in FIG. 3 is an example, and is not limited to this.

<Data stored in the first database>
Next, an example of data stored in the first database will be explained.
The language model database stores language models. The language model is a probability model that gives a probability that any character string is a Japanese sentence or the like. Further, the language model is, for example, an N-gram model, a hidden Markov model, a maximum entropy model, or the like.

The acoustic model database stores sound source models. The sound source model is a model used to identify the sound source of a collected acoustic signal.

The acoustic feature is a feature calculated after converting a collected acoustic signal into a frequency domain signal by performing Fast Fourier Transform. The acoustic features are, for example, a static Mel-Scale Log Spectrum (MSLS), a delta MSLS, and one delta power, which are calculated at predetermined time intervals (for example, 10 ms). Note that MSLS is obtained by performing inverse discrete cosine transform on MFCC (Mel Frequency Cepstrum Coefficient) using a spectral feature as a feature for acoustic recognition.

The dialogue corpus database stores dialogue corpora. The dialogue corpus is a corpus used when the communication robot 1 and the user have a dialogue, and is, for example, a scenario depending on the content of the dialogue.

The comparison image database stores, for example, images used in pattern matching. Images used during pattern matching include, for example, images of the user, images of the user's family, images of the user's pets, images of the user's friends and acquaintances, and the like.

The image feature amount is, for example, a feature amount extracted from an image of a person or an object by well-known image processing.
Note that the example described above is just an example, and the first database 107 may store other data.

<Flow of cognition, learning, and social abilities>
Next, the flow of recognition and learning performed by the social ability generation device 100 of this embodiment will be explained. FIG. 4 is a diagram showing the flow of recognition, learning, and social ability performed by the social ability generation device 100 of this embodiment.

Recognition result 201 is an example of a result recognized by recognition unit 105. The recognition result 201 is, for example, interpersonal relationships, interpersonal mutual relationships, etc.

Multimodal learning and understanding 211 is an example of learning content performed by the learning unit 108. The learning method 212 is machine learning or the like. Furthermore, the learning objects 213 are social constituent elements, social models, psychology, humanities, and the like.

Social abilities 221 are social skills, such as empathy, individualization, adaptability, and emotional aphorism.

<Data to be recognized>
Next, an example of data recognized by the recognition unit 105 will be explained.
FIG. 5 is a diagram showing an example of data recognized by the recognition unit 105 according to this embodiment. In this embodiment, personal data 301 and interpersonal relationship data 351 are recognized as shown in FIG.

Personal data is behavior that occurs within one person, and includes data acquired by the photographing unit 102 and sound collection unit 103, and data obtained by performing voice recognition processing, image recognition processing, etc. on the acquired data. . Personal data includes, for example, voice data, semantic data resulting from voice processing, voice volume, voice intonation, spoken words, facial expression data, gesture data, head posture data, face direction data, and gaze data. , co-occurrence expression data, physiological information (body temperature, heart rate, pulse rate, etc.). Note that the designer of the communication robot 1 may select, for example, what kind of data to use. In this case, for example, the designer of the communication robot 1 may set important characteristics of personal data in communication for actual communication or demonstration between two people. Furthermore, the recognition unit 105 recognizes the user's emotions as personal data based on information extracted from each of the acquired utterances and images. In this case, the recognition unit 105 recognizes based on, for example, the volume and intonation of the voice, duration of speech, facial expression, etc. The communication robot 1 of this embodiment controls the user's emotions to maintain good emotions and to maintain a good relationship with the user.

Here, an example of a method for recognizing a user's social background will be explained.
The recognition unit 105 estimates the user's nationality, place of birth, etc. based on the acquired utterances and images and data stored in the first database 107. The recognition unit 105 extracts the user's daily life schedule, such as wake-up time, going out time, returning home time, and bedtime, based on the acquired utterances and images and the data stored in the first database 107. The recognition unit 105 determines the user's gender, age, occupation, hobbies, career, preferences, family composition, and religious beliefs based on the acquired utterances, images, life schedule, and data stored in the first database 107. , the degree of attachment to the communication robot 1, etc. is estimated. Note that since the social background may change, the communication robot 1 updates information regarding the user's social background based on the conversation, images, and data stored in the first database 107. In order to enable emotional sharing, the social background and the degree of attachment to the communication robot 1 are not limited to the level that can be entered such as age, gender, career history, etc., but also the emotional ups and downs depending on the time of day. It is recognized based on the volume and intonation of the voice regarding the topic. In this way, the recognition unit 105 learns things that the user is not aware of based on daily conversations and facial expressions during conversations.

Interpersonal relationship data is data regarding relationships between users and other people. By using interpersonal relationship data in this way, social data can be used. The interpersonal relationship data includes, for example, the distance between people, whether or not the people who are interacting are looking at each other, the intonation of their voices, the volume of their voices, and the like. As will be explained later, the distance between people varies depending on interpersonal relationships. For example, interpersonal relationships between husband and wife and friends are L1, and interpersonal relationships between businessmen are L2, which is larger than L1.

Note that, for example, the designer of the communication robot 1 may set important features of interpersonal data in communication for actual communication or demonstration between two people. Note that such personal data, interpersonal relationship data, and information regarding the user's social background are stored in the first database 107 or the storage unit 106.

Furthermore, when there are multiple users, for example, when there are users and their families, the recognition unit 105 collects and learns personal data for each user, and estimates the social background of each person. Note that such social background may be acquired, for example, via the network and the receiving unit 101, and in that case, the user may input his/her social background or select an item using, for example, a smartphone. Good too.

Here, an example of a method for recognizing interpersonal relationship data will be explained.
The recognition unit 105 estimates the distance (interval) between the people communicating with each other based on the acquired utterances and images and the data stored in the first database 107. The recognition unit 105 detects whether the lines of sight of the people communicating with each other are intersecting, based on the acquired utterances and images, and data stored in the first database 107. Based on the acquired utterance and the data stored in the first database 107, the recognition unit 105 determines the content of the utterance, the volume of the voice, the intonation of the voice, the received e-mail, the sent e-mail, and the sending and receiving of the sent and received e-mail. Estimates friendships, work colleagues, relatives, and parent-child relationships based on the previous partner.

In addition, in the initial state used, the recognition unit 105 randomly selects one combination of initial values of social background and personal data stored in the first database 107, and performs communication. may be started. Then, if the behavior generated by the randomly selected combination makes it difficult to continue communication with the user, the recognition unit 105 may reselect another combination.

<Learning procedure>
In this embodiment, the learning unit 108 performs learning using the personal data 301 and interpersonal relationship data 351 recognized by the recognition unit 105 and data stored in the second database 109.

Here, we will explain social composition and social norms. In spaces where people participate in social interactions, interpersonal relationships differ depending on people's careers, for example. For example, a relationship with a distance of 0 to 50 cm from a person is an intimate relationship, and a relationship with a distance of 50 to 1 meter from a person is a personal relationship. A relationship where the distance between one person and another person is 1 to 4 meters is a social relationship, and a relationship where the distance between one person and another person is 4 meters or more is a public relationship. Such social norms are used as rewards (implicit rewards) during learning, based on whether or not gestures and utterances conform to social norms.

Furthermore, the interpersonal relationship may be set according to the environment and user in which the system is used by setting the feature amount of the reward at the time of learning. Specifically, you could set multiple intimacy settings, such as setting a rule to not talk to people who don't like robots much, and setting a rule to actively talk to people who like robots. good. In a real environment, the recognition unit 105 recognizes what type the user is based on the results of processing the user's utterances and images, and the learning unit 108 selects a rule. good.

Furthermore, the human trainer may evaluate the behavior of the communication robot 1 and provide rewards (implicit rewards) according to the social structure and norms that the human trainer knows.

<Data stored in the second database>
Next, an example of data stored in the second database will be explained.
Social components include, for example, age, gender, occupation, and relationships between multiple people (parents and children, husband and wife, lovers, friends, acquaintances, colleagues, neighbors, teachers and students, etc.).

Social norms are rules and manners between individuals or multiple people, and are associated with utterances, gestures, etc. depending on age, gender, occupation, and relationships between multiple people.

Data related to psychology is, for example, data on findings obtained through past experiments and verifications (for example, the attachment relationship between mothers and infants, complexes such as the Oedipus complex, conditioned reflexes, fetishism, etc.).

Data related to the humanities include, for example, data on religious rules, customs, national characteristics, regional characteristics, and characteristic acts, actions, and utterances of a country or region. For example, data shows that Japanese people express their agreement by nodding without saying words. Data related to the humanities includes, for example, what is considered important and what is given priority depending on the country or region.

FIG. 6 is a diagram illustrating an example of a system that performs deep reinforcement learning using raw data in a comparative example.
In the comparative example, when using raw data 902 of a captured image 901 and a captured audio signal 901 for learning, it is necessary to perform deep reinforcement learning 903. The system of this comparative example is difficult to implement. The reason is that it is necessary to collect enough teaching data for deep reinforcement learning, but it is difficult to collect it. The reason why it is difficult to collect is that the number of times the required feature appears in the raw data is limited.

Therefore, in this embodiment, instead of directly using raw data (audio signals, images) for learning, features are detected from the raw data and those features are used for learning, which enables reinforcement learning instead of deep reinforcement learning. .

FIG. 7 is a diagram illustrating an example of an agent creation method used by the behavior generation unit 110 according to the present embodiment.
The area indicated by reference numeral 300 is a diagram showing the flow from input to creation of an agent to output (agent).
The image photographed by the photographing unit 102 and the information 310 collected by the sound collecting unit 103 are information about a person (a user, a person related to the user, another person) and information about the environment around the person. Raw data 302 acquired by the imaging unit 102 and the sound collection unit 103 is input to the recognition unit 105.

The recognition unit 105 collects a plurality of pieces of information from the input raw data 302 (voice volume, voice intonation, utterance content, uttered words, user's line of sight, user's head posture, user's face orientation). , user's ecological information, distance between people, whether or not people are looking at each other, etc.). The recognition unit 105 uses a plurality of pieces of extracted and recognized information to perform multimodal understanding using, for example, a neural network.
The recognition unit 105 identifies an individual based on at least one of an audio signal and an image, and assigns identification information (ID) to the identified individual. The recognition unit 105 recognizes the motion of each identified person based on at least one of the audio signal and the image. The recognition unit 105 performs well-known image processing and tracking processing on the image, for example, to recognize the line of sight of the identified person. The recognition unit 105 performs, for example, speech recognition processing (sound source identification, sound source localization, sound source separation, speech interval detection, noise suppression, etc.) on the speech signal to recognize speech. The recognition unit 105 performs well-known image processing on the image, for example, to recognize the head posture of the identified person. For example, when two people are photographed in a photographed image, the recognition unit 105 recognizes the interpersonal relationship based on the content of the utterance, the distance between the two people in the photographed image, and the like. The recognition unit 105 recognizes (estimates) the social distance between the communication robot 1 and the user, for example, according to the results of processing each of the photographed image and the collected audio signal.

The learning unit 108 performs reinforcement learning 304 instead of deep learning. Reinforcement learning involves learning to select the most relevant features (including social constructs and social norms). In this case, multiple pieces of information used in multimodal understanding are used as features for input. The input to the learning unit 108 is, for example, raw data itself, name ID (identification information), facial influence, recognized gestures, keywords from voice, etc. The output of the learning unit 108 is the behavior of the communication robot. The behavior to be output may be anything that is desired to be defined according to the purpose, such as a voice response, a robot routine, or an angle for the robot to rotate. Note that in multimodal understanding, a neural network or the like may be used for detection. In this case, different modalities of the body may be used to detect human activity. Furthermore, which feature to use may be selected in advance by, for example, the designer of the communication robot 1. Furthermore, in this embodiment, by using implicit rewards and explicit rewards during learning, social models and social constructs can be incorporated. The result of reinforcement learning is the output, which is the agent 305. In this manner, in this embodiment, the agent used by the behavior generation unit 110 is created.

The area indicated by reference numeral 350 is a diagram showing how to use the reward.
Implicit rewards 362 are used to learn implicit responses. In this case, the raw data 302 includes user reactions, and the raw data 302 is subjected to the multimodal understanding 303 described above. The learning unit 108 generates an implicit reaction system 372 using the implicit reward 362 and social models stored in the second database 109. Note that the implicit reward may be obtained by reinforcement learning or may be given by a human. Further, the implicit reaction system may be a model acquired through learning.

For explicit response learning, for example, a human trainer evaluates the behavior of the communication robot 1 and gives a reward 361 according to the social structure and social norms that the human trainer knows. Note that the agent adopts the action that maximizes the reward for the input. As a result, the agent adopts behaviors (utterances, gestures) that maximize positive emotions toward the user.

The learning unit 108 uses this explicit reward 361 to generate an explicit reaction system 371. Note that the explicit reaction system may be a model acquired through learning. Note that the explicit reward may be given by the user after evaluating the behavior of the communication robot 1. Based on the user's utterances and actions (gestures, facial expressions, etc.), the communication robot 1 may, for example, The reward may be estimated based on whether or not the user was able to take the desired action.
During operation, the learning unit 108 outputs the agent 305 using these learning models.

Note that in this embodiment, an explicit reward that is a user's reaction is given priority over an implicit reward. The reason for this is that user reactions are more reliable in communication.

<Processing procedure example>
Next, an example of the processing procedure will be explained. FIG. 8 is a flowchart illustrating an example of a procedure for social ability generation processing according to the present embodiment.

(Step S11) The recognition unit 105 acquires the image photographed by the photographing unit 102 and the acoustic signal collected by the sound collection unit 103.

(Step S12) The recognition unit 105 recognizes, detects, or extracts feature information about voices from the acoustic signal, and recognizes, detects, or extracts feature information about people from the image. Note that the feature information regarding the voice is at least one of a voice signal, information on the volume of the voice, information on the intonation of the voice, and the meaning of the utterance. The characteristic information regarding a person is at least one of the following: facial expression information, gesture information made by the person, head posture information, face direction information, and gaze information.

(Step S13) The recognition unit 105 recognizes the interaction that occurs between the communication robot 1 and a person, or the interaction that occurs between multiple people, based on the acquired information and the data stored in the first database 107. .

(Step S14) The learning unit 108 learns human emotional interaction using the recognition result output by the recognition unit 105 and the data stored in the second database 109.

(Step S15) The behavior generation unit 110 generates behavior (utterances, gestures, images) for the user based on the learned results and the acquired information.

<Communication between communication robot 1 and people>
Next, an example of communication between the communication robot 1 and a person will be explained.
The timing of the speech of the communication robot 1 is set as an initial value, for example, when the robot returns home or when the robot wakes up. Then, by repeating communication, the communication robot 1 may learn the timing to start speaking.
Alternatively, the communication robot 1 may start speaking in response to the user's utterances or actions. In this case, the communication robot 1 does not start a conversation in response to a command adopted by a smart speaker, etc., but rather, for example, the communication robot 1 does not start a conversation in response to a command adopted by a smart speaker or the like. The communication robot 1 starts a conversation by recognizing that a face is looking at the communication robot 1, etc.) or that a light in the room is turned on. Furthermore, the communication robot 1 finishes the conversation by recognizing, for example, the content of the user's utterances, the user's facial expressions, and the like. The communication robot 1 may switch the start timing and end timing of speech depending on the user's social background (including age and gender), attachment to the communication robot 1, and the like.

Examples of communication between the communication robot 1 and the user will be explained using FIGS. 9 to 12.
FIG. 9 is a diagram showing an example of communication between the communication robot 1 and the user according to the present embodiment. The example in FIG. 9 shows the communication robot 1 talking to the user and having a conversation with the user when the user returns home. During communication, the communication robot 1 acts (utterances, gestures, image presentation) so as to maintain a good relationship with the person. Note that the communication robot 1 controls gestures and gestures by driving the boom horizontal bar and the like using the drive unit 1103 and the actuator 113.

10 and 11 are diagrams showing examples of images displayed on the display unit during communication between the communication robot 1 and the user according to the present embodiment.
The example in FIG. 10 is an example in which heart marks representing attachment are displayed on the display sections 111a and 111b, and an image corresponding to an image of a smiling mouth is displayed on the display section 111c.
The example shown in FIG. 11 is an example in which an illustration illustrating the location of the scheduled date and the weather forecast for the day are presented, and the weather forecast is uttered and read out by the speaker 112. Although the example in FIG. 11 shows an example in which an illustration illustrating a location is presented, the communication robot 1 acquires an image (photo or video) of the planned location via the receiving unit 101 and displays the acquired image. may be presented.
In this manner, the display unit 111 displays an image that maintains a good relationship with the person by causing the person to behave in a manner that maximizes positive emotions. As a result, according to the present embodiment, it is possible to make a person behave in a manner that maximizes positive emotions, and it is possible to maintain a good relationship with the person.

FIG. 12 is a diagram showing an example of communication by the communication robot 1 when communicating with a user's friend according to the present embodiment. In the example of FIG. 12, the communication robot 1 sends a message to the user's friends through dialogue with the user. The terminal 200 owned by the user's friend receives this message and displays it on the display (g301). Then, the terminal 200 transmits a reply g302 to the user to the communication robot 1 according to the operation result of the user's friend. Based on the information received from the terminal 200, the communication robot 1 presents replies from the user's friends through actions (speech, gestures, image presentation).

In the example described above, the communication robot 1 communicates with the user using voice, motion (gesture), and images, but the present invention is not limited to this. In order to communicate with the user, it is preferable that the communication robot 1 uses two or more output means, and any two or more of voice, movement (gesture), and image may be used. Alternatively, the output means may be, for example, text and motion, text and voice, or the like. Further, in order to prevent the user from getting bored with the communication robot 1, it is preferable that there be a plurality of output means.

Furthermore, the input from the user to the communication robot 1 is not limited to the above-mentioned voices and images. It is sufficient if the user's behavior can be acquired, and other information may also be acquired. Other information is, for example, contact information such as when the user touches or hits the communication robot 1.

As described above, in this embodiment, the communication robot 1 is made to recognize the interaction that occurs between the communication robot 1 and a person, or the interaction that occurs between multiple people. In this embodiment, human emotional interactions are learned from the recognized content using machine learning, psychology, social customs, humanities, etc., and the robot's social abilities are generated from the learned content. did. Furthermore, in this embodiment, explicit rewards are used in addition to implicit rewards in learning.

As a result, according to this embodiment, social skills of the robot can be generated based on emotional interactions with people. According to this embodiment, it is possible to provide a home social robot and agent that foster relationships with people. According to this embodiment, empathetic communication and interaction between machines and humans can be generated. According to this embodiment, the concept of a "machine" friend, such as a pet friend, or a robot friend can be provided. According to this embodiment, it is possible to provide a machine that is socially intelligent and capable of navigating social scenarios. Thereby, according to this embodiment, an emotional connection can be formed between the robot and the person.
Furthermore, according to this embodiment, features are extracted from each of the collected audio signals and the captured images, and reinforcement learning is performed using the extracted features, so deep machine learning using raw data is performed. Learning can be performed without using a large amount of teaching data.

<Modified example>
In the embodiment, a communication robot 1 will be described as an example of a device that performs communication, but the present embodiment is also applicable to other devices, such as an in-vehicle navigation device, a smartphone, a tablet terminal, and the like. For example, when applied to a smartphone, a still image of the communication robot 1 as shown in FIG. 3 may be displayed on the display section of the smartphone, and communication may be performed mainly by voice. Alternatively, the gestures of the communication robot 1 may be displayed in animation on the display section of the smartphone.

FIG. 13 is a diagram showing an example in which the communication robot of this embodiment is applied to a car navigation system 300 in a vehicle. Note that the car navigation system 300 may be a smartphone, a tablet terminal, or the like.
The car navigation system 300 displays an image of a communication robot on the display unit. In this case, the communication robot operates as an agent. Then, the agent realizes the functions of the communication robot 1 (excluding the drive unit, actuator, etc.) using the imaging unit, sound collection unit, display unit, speaker, etc. included in the car navigation system 300.

When applied to the car navigation system 300, the communication robot displayed on the display unit may be a still image or an animation. In this case, the agent performs at least a voice interaction as a response. Even in this case, when communicating, the agent acts (utterances, gestures, image presentation) to maintain a good relationship with the person.

FIG. 14 is a diagram showing an example of connections with various devices in the home when applied to a car navigation according to the present embodiment. Note that also in FIG. 14, the car navigation system 300 may be a smartphone, a tablet terminal, or the like. It is assumed that the car navigation system 300 includes a communication section (a receiving section and a transmitting section), and each device at home is connected via a network. The agent applied to the car navigation system 300 responds to communication with the user by, for example, opening/closing the parking lot shutter 401, instructing the rice cooker to start operating 402, instructing the air conditioner to start operating or setting the temperature, etc. 403. , an instruction 404 to start turning on lights in a room, etc., an instruction 405 to start operating an automatic lawn mower, and the like. In addition, the agent does not simply give operation instructions, but also tells the user what time they plan to return home, the user's preferred temperature setting, and the user's preferred room brightness settings, based on the user's utterances. The user may also learn how to communicate and, based on the results of the learning, provide the optimal timing and setting instructions so that these tasks are completed by the time the user returns home.

Note that a program for realizing all or part of the functions of the social ability generation device 100 according to the present invention may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be read into a computer system. , all or part of the processing performed by the social ability generation device 100 may be performed. Note that the "computer system" herein includes hardware such as an OS and peripheral devices. Furthermore, the term "computer system" includes a WWW system equipped with a home page providing environment (or display environment). Furthermore, the term "computer-readable recording medium" refers to portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, and storage devices such as hard disks built into computer systems. Furthermore, "computer-readable recording medium" refers to volatile memory (RAM) inside a computer system that serves as a server or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. This also includes programs that are retained for a certain period of time.

Further, the program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in a transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. Moreover, the above-mentioned program may be for realizing a part of the above-mentioned functions. Furthermore, it may be a so-called difference file (difference program) that can realize the above-described functions in combination with a program already recorded in the computer system.

Although the mode for implementing the present invention has been described above using embodiments, the present invention is not limited to these embodiments in any way, and various modifications and substitutions can be made without departing from the gist of the present invention. can be added.

1... Communication robot, 101... Receiving unit, 102... Photographing unit, 103... Sound collecting unit, 104... Sensor, 100... Social ability generation device, 106... Storage unit, 107... First database, 109... Second database, 111... Display section, 112... Speaker, 113... Actuator, 114... Transmission section, 105... Recognition section, 108... Learning section, 110... Motion generation section, 1101... Image generation section, 1102... Sound generation section, 1103... Drive section , 1104...Transmission information generation unit

Claims (6)

Acquire human information about a person, extract characteristic information about the person from the acquired human information, recognize the interaction that occurs between the communication device and the person, and recognize the interaction that occurs between the people. cognitive means,
A learning means for multimodally learning emotional interactions between people using the extracted feature information about the person;
a behavior generating means for generating a behavior based on the learned emotional interaction information of the person;
Equipped with
The learning means is
Learning using implicit and explicit rewards,
generating an implicit response system using the implicit reward and social model;
evaluating the behavior of the communication device and generating an explicit response system using the explicit reward according to social composition and social norms;
When used, the implicit reaction system and the explicit reaction system are used to generate an output that incorporates at least one of a social model and a social construct;
The implicit reward is a reward learned multimodally using feature information about the person,
The social ability generating device , wherein the explicit reward is a reward based on a result of evaluating the behavior of the communication device toward the person generated by the behavior generating means . a sound collection unit that collects the acoustic signal;
A photography department that photographs images including users;
The recognition means performs voice recognition processing on the collected acoustic signal to extract feature information related to the sound, and performs image processing on the captured image to extract feature information related to human behavior included in the image. extract,
The characteristic information regarding the person includes characteristic information regarding the voice and characteristic information regarding the human behavior,
The feature information regarding the voice is at least one of a voice signal, voice volume information, voice intonation information, and meaning of the utterance,
The feature information regarding human behavior includes at least one of the following: facial expression information, gesture information performed by a person, head posture information, face direction information, line of sight information, and distance between people. There is one
The social ability generation device according to claim 1 . The learning means is
Learning using social norms, social components, psychological knowledge, and humanities knowledge,
The social ability generation device according to claim 1 or claim 2 . The recognition means acquires person information about the person, extracts characteristic information about the person from the acquired person information, recognizes the interaction that occurs between the person and the communication device that performs communication, and recognizes the interaction that occurs between the person and the person. Recognizing the efforts,
a learning means multimodally learns the emotional interaction of the person using the extracted characteristic information about the person;
The learning means is
Learning using implicit and explicit rewards,
generating an implicit response system using the implicit reward and social model;
evaluating the behavior of the communication device and generating an explicit response system using the explicit reward according to social composition and social norms;
When used, the implicit reaction system and the explicit reaction system are used to generate an output that incorporates at least one of a social model and a social construct;
The implicit reward is a reward learned multimodally using feature information about the person,
The explicit reward is a reward based on the result of evaluating the behavior of the communication device towards the person generated by the behavior generation means,
the action generation means generates the action based on the learned emotional interaction information of the person;
Social competence generation method. Acquire human information about a person, extract characteristic information about the person from the acquired human information, recognize the interaction that occurs between the communication device and the person, and recognize the interaction that occurs between the people. cognitive means,
A learning means for multimodally learning emotional interactions between people using the extracted feature information about the person;
a behavior generating means for generating a behavior based on the learned emotional interaction information of the person;
Equipped with
The learning means is
Learning using implicit and explicit rewards,
generating an implicit response system using the implicit reward and social model;
evaluating the behavior of the communication device and generating an explicit response system using the explicit reward according to social composition and social norms;
When used, the implicit reaction system and the explicit reaction system are used to generate an output that incorporates at least one of a social model and a social construct;
The implicit reward is a reward learned multimodally using feature information about the person,
The communication robot , wherein the explicit reward is a reward based on a result of evaluating the behavior of the communication device toward the person generated by the behavior generation means . Equipped with a display section,
The action generation means generates an image that maintains a good relationship with the person by causing the person to behave in a manner that maximizes positive emotions, and causes the display unit to display the generated image.
The communication robot according to claim 5 .

Images