JP2006088251A - User behavior inducing system and method thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a user behavior inducing system and a method thereof, giving output according to a user, and performing inducing behavior suitable for each user. <P>SOLUTION: This user behavior inducing system includes: a data input part for inputting the behavior data of a user and a robot from a sensor; a behavior classifying processing part for classifying the behavior data of a user input by the data input part by each similar behavior of the user behavior; a cause and effect relationship estimating part for obtaining the probability of occurrence of the classified user behavior corresponding to the behavior data of the robot; a designation input part for inputting the user behavior to be induced; a similarity retrieving part for retrieving for the user behavior to be induced and the similar user behavior from the classified user behaviors; a behavior retrieving part for extracting the behavior data of the robot according to the probability of occurrence of behavior data of the robot corresponding to the retrieved user behavior; and a robot behavior processing part for causing the robot to perform the extracted behavior data of the robot. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

The present invention relates to a robot that urges a user's action by using an accumulation result of an action database based on the interaction between the user and the robot through service provision.

  In recent years, entertainment robots that entertain users and robots useful to users have been researched and developed at home. In the former entertainment robot, the purpose is to entertain the user, and one of the challenges is not to get bored. As the latter useful robot, one that acts by listening to a user's command and one that recognizes the user's intention and performs an action suitable for the user have been proposed.

  As an entertainment robot, there is one that causes a robot apparatus to perform reinforcement learning for action acquisition so that the robot does not get tired of the user. In that case, a robot apparatus and a robot control method, a recording medium, and a program characterized by dynamically changing the learning rate of the action value are mentioned (see Patent Document 1).

  As a useful robot, as a controller that controls various devices according to the user's lifestyle, living environment, etc., or according to the user's preferences, the user's operation information and the external situation when the operation is made in advance An integrated controller, a device control method, and a device control program for storing an operation signal suitable for the user from the current user operation information and external situation information have been proposed. An adapted robot is cited (see Patent Document 2).

In addition, there is a method for realizing face-to-face communication between a human and a robot by imitating human behavior by a robot (see Non-Patent Document 1). This means that in face-to-face communication between a human and a robot, voice information and joint angle information are acquired from a human action, and the robot imitates the person by reproducing the acquired information.
JP 2003-340760 A JP 2003-111157 A Fundamental Experiment on Human-Robot Communication Based on Imitation (Proceedings of the 19th Annual Conference of the Robotics Society of Japan (1J15)

  As described above, in order to keep the user from getting bored with the entertainment robot, the output operation pattern for the input signal to the robot is changed by reinforcement learning, and the useful robot records the user's past operation pattern. In addition, from the current situation, the operation that the user will perform next is pre-read, and the robot performs on behalf of the user. could not.

  Furthermore, in the face-to-face communication between a human and a robot, the face-to-face communication between a human and a robot is the purpose. In order for the robot to imitate a human, the acquired information and the output information must be of the same dimension (for example, acquired In order to use human joint angle information, the robot must also have a corresponding joint).

  In order to solve the above-described problems, the present invention provides a user behavior inducing system and a user behavior inducing method that are capable of output tailored to the user and that can execute the induced behavior suitable for each user. The purpose is to do.

  In order to achieve the above-mentioned goal, in the present invention, a data input unit that inputs user and robot behavior data from a sensor, and a user behavior data input by the data input unit is classified for each behavior similar to the user behavior. The action classification processing unit that performs the division, the causal relationship estimation unit that obtains the probability of occurrence of the user behavior classified by the behavior classification processing unit corresponding to the robot behavior data input by the data input unit, and the user behavior to be induced are input A similarity search unit for searching for a user behavior similar to the user behavior to be induced from the user input classified by the behavior classification processing unit, and a robot corresponding to the user behavior searched by the similarity search unit The behavior search unit that extracts the robot behavior data based on the occurrence probability of the robot behavior data, and the robot behavior data extracted by the behavior search unit Characterized by comprising a robot action processing unit for executing data to the robot.

  Thus, when the user wants to perform a desired action, the robot action that has caused the desired action in the past is searched, and the robot executes the action, thereby urging the user to perform a certain action.

  According to the present invention, it is possible to perform output in accordance with a user, and it is possible to execute a triggering action suitable for each user. It is also possible to consider pseudo-communications that do not get bored by inducing user actions such as “enjoying” or “taking rare actions”.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
An embodiment of the present invention will be described with reference to a user behavior induction system using a robot mounted with a camera platform that moves with wheels and has two degrees of freedom (pan and tilt) as a head, for example. FIG. 1 shows a configuration of the present embodiment, and FIG. 2 shows a schematic diagram of the robot of the present embodiment.

  As shown in FIG. 1, the present embodiment is roughly divided into a data accumulation processing unit 110 for a database and a user action induction processing unit 130 using a database. The data accumulation processing unit 110 includes a data input unit 111, a segmentation unit 112, a feature amount extraction unit 113, a feature amount non-dimensionalization unit 114, a behavior classification processing unit 115, a causal relationship estimation unit 116, and a robot behavior storage unit 117. Is done. The user behavior induction processing unit 130 includes an instruction input unit 131, a similarity search unit 132, a behavior search unit 133, and a robot behavior processing unit 134.

  As shown in FIG. 2, the input device that detects the action of the user 200 is a sensing device that is fixedly installed in the room, as a position detection RFID 201 that detects the position of the user 200, and a fixed camera 220 that acquires an image of the user. There is a fixed microphone 230 that acquires the voice of the user 200, and as a sensing device mounted on the robot 210, a camera 211 installed on the head, a microphone 212, and a touch installed on the surface of the robot 210 to detect contact. There is a sensor 213.

First, the operation of the database data storage processing unit 110 will be described. First, in the data input unit 111, outputs (time-series data) from the respective sensing devices such as the fixed sensors 101-1 to 101-N and the sensors 102-1 to 102-M mounted on the robot are input at a certain sampling period. This time series data is stored in the time series DB 121. The output from the sensing device is an indoor coordinate value (X, Y) if it is RFID, image data if it is a camera, sound data if it is a sound, and a pressed pressure value if it is a touch sensor. Moreover, what can be recognized at an abstract level such as a word such as voice data is also stored as the word. These time-series data are stored not only for the user but also for the behavior of the robot. Examples of robot actions to be stored include the robot position (which may be RFID or may be a position calculated by a self-position recognition unit from the internal sensor of the robot), the robot head angle, robot voice data, etc. is there. If there is an abstract word or the like regarding the voice data, the word is also stored. As shown in FIG. 3A, each data becomes time series data x _i (t) (where i is an input parameter) for each sampling period t of the sensor (abstract items such as words are For example, the data at the time of utterance). Here, when a data string is shown on the time axis, time series data for each sampling period t is output from the sensing device as shown in FIG. 5A and stored in the time series DB by the data input unit 111.

Next, the time series data input by the data input unit 111 is converted into time series data as shown in FIG. 3A as user actions (U ₁ , U ₂ ,..., U _n ) having a certain meaning. The portion surrounded by the dotted line is carved by the carving unit 112. As a method of carving, a method of carving at regular intervals, a method of standardizing each data and carving it at a large deviation, or a method of carving audio data when there is audio data are used. For example, by using data from a sensor that detects the user's position and separating the position change amount in a large deviation, it is possible to separate the user action when the user is walking and the user action when the user is stopped. . In addition, the robot action at that time is also divided using the same method. Here, time-series data as shown in FIG. 5 (b) When indicating the data sequence on the time axis is _R 1 _{to R n,} carved with cut portions 112 in U 1 ~U _n.

Next in the feature amount extraction unit 113, user behavior _{(U 1, U 2, ···} , U n) each element time-series information of _{x 1 (t), x 2} (t), ···, x l ( A feature p _j (n) (j is a feature number and n is a user action number) each represented by a single numerical value is extracted. Features include the average value, maximum value, minimum value, standard deviation, etc. of the time series data x _i (t) itself, the change amount x _i ′ (t) calculated from the time series data x _i (t), and the acceleration. The average value, maximum value, minimum value, standard deviation, or the like of the quantity x _i ″ (t), or the time of the user action itself is used. For example, if data from a sensor that detects the user's position is used, this feature p _j (n) is the average moving speed or the maximum moving speed of the user during the user's action, or if data from a microphone is used, The maximum volume or the pitch of sound calculated from audio data. As illustrated in FIG. 3B, the feature value p _j (n) of each user action extracted by the feature value extraction unit 113 is stored in the feature value DB 122.

Next, the feature quantity non-dimensionalization unit 114 calculates the standard deviation S _j (n) for each user action of each feature as shown in Equation 1, and from this standard deviation S _j (n) to Equation 2 As shown, dimensionless and feature parameter P _j (n) is calculated.

Then, the feature quantity non-dimensionalization unit 114 performs dimensionless processing for each user action as illustrated in FIG. 3C and stores the feature parameter P _j (n) in the dimensionless feature quantity DB 123. Also, expressed in the form of a position vector u _n a multidimensional space in which the user behavior itself and its characteristic parameters and the axis as indicated by the number 3.

The action classification processing unit 115 determines the similarity of each behavior from short and long distance position vector u _n indicating the user behavior, perform classification for each similar behavior. For example, as shown in FIG. 4A, the position vectors u ₁ and u ₃ indicating the user behavior are clustered and determined as the user behavior U _a . Similarly, the position vectors u ₂ and u ₄ are determined as user actions U _b and U _c , respectively. Further, the classification result of FIG. 4A is stored in the user behavior classification DB 124.

  The causal relationship estimation unit 116 estimates the classified behavior and the robot behavior that caused the classified behavior. As the estimation method, for example, as shown in FIG. 5C, the robot behavior that was performed immediately before the user behavior occurred, or the robot behavior that was frequently performed before a certain user behavior occurred from the database. It is estimated that the user behavior is caused by. And as shown in FIG.4 (b), the generation | occurrence | production probability of the classified user action with respect to each robot action is calculated | required, and it memorize | stores in causal relationship corresponding | compatible DB125.

The robot behavior storage unit 117 stores in the robot behavior DB 126 the robot behavior that is estimated by the causal relationship estimation unit 116 and causes a certain user behavior.
Subsequently, an operation of the user behavior induction processing unit 130 will be described. First, an instruction input unit 131 inputs a user action to be induced. The input person may be the user who performs the action or another user. Next, the similarity search unit 132 searches the user action classification DB 124 from the input information and extracts the closest user action.

  The behavior search unit 133 searches the causal relationship correspondence DB for the robot behavior that has derived (caused) the extracted user behavior, and the robot behavior processing unit 134 extracts the robot behavior from the robot behavior DB 126 and executes it. To do. For example, the behavior retrieval unit 133 retrieves the robot behavior with the highest probability from the causal relationship correspondence DB 125, and the robot behavior processing unit 134 extracts the robot behavior with the highest probability from the robot behavior DB 126 and outputs it to the robot body. By doing so, the robot body performs an action that prompts the user action to be induced.

  Specifically, for example, consider a case where a child is woken up as shown in FIG. When the parent wants the robot to raise the child A, the parent instructs the robot to “wake up the child A” (S1). The instruction input unit 131 searches the user behavior classification DB 124 for the user behavior closest to the event “child A has occurred” (S2). Then, the robot action that caused the user action is searched, and the robot is made to execute it. The event that happened to child A is “Robot Wake Up! "10% when speaking", "" Morning! "5% when speaking" and "90% chance if robot hits wall". In this case, the robot executes “the robot collides with the wall” having the highest probability of occurrence (S3). Whether or not the child A as a result has occurred is detected and determined by the data storage processing unit 110 (S4). When the child A wakes up, each database is updated (S5). If the child A does not wake up, the robot performs the robot action with the next highest probability (S6), and the data storage processing unit 110 detects and determines whether the child A has woken up.

By repeating the above processing, it is possible to execute a user's behavior effectively and a triggering method tailored to each user.
Here, the data accumulation processing unit 110, the user action induction processing unit 130, and the databases (DB) 121 to 126 may be configured in the robot main body 210 or may be configured outside the robot main body 210. Further, even if some of these are configured in the robot body 210, the present invention can be applied.

The block diagram which shows the user action induction system of embodiment Environment schematic diagram showing a user behavior induction system of an embodiment The figure which shows the data structure of time series DB, feature-value DB, dimensionless feature-value DB. The figure which shows the data structure of user action classification DB and causal relationship corresponding | compatible DB. Time chart showing behavior of robot and user of embodiment The flowchart which shows the user action induction system of embodiment

Explanation of symbols

101-1 to 101-N: Fixed sensor 102-1 to 102-M: Robot-mounted sensor 110: Data storage processing unit 111: Storage unit 112: Separation unit 113: Feature amount extraction unit 114: Feature amount non-dimensionalization unit 115 : Behavior classification processing unit 116: Causal relation estimation unit 117: Robot behavior storage unit 121: Time series DB
122: Feature value DB
123: dimensionless feature DB
124: User action classification DB
125: Causal relationship correspondence DB
126: Robot action DB
130: User behavior induction processing unit 131: Instruction input unit 132: Similarity search unit 133: Behavior search unit 134: Robot behavior processing unit 200: User 201: RFID for position detection
210: Robot 211: Camera 212: Microphone 213: Touch sensor 220: Fixed camera 230: Fixed microphone

Claims (8)

A data input unit for inputting user and robot behavior data from sensors;
An action classification processing unit that performs classification for each action similar to the user action from the user action data input by the data input unit;
A causal relationship estimation unit for determining the occurrence probability of the user behavior classified by the behavior classification processing unit corresponding to the behavior data of the robot input by the data input unit;
An instruction input unit for inputting a user action to be induced;
From the user behavior classified by the behavior classification processing unit, a similarity search unit that searches for user behavior similar to the user behavior to be induced,
An action search unit for extracting robot action data based on the occurrence probability of the robot action data corresponding to the user action searched by the similarity search unit;
A user behavior induction system, comprising: a robot behavior processing unit that causes a robot to execute robot behavior data extracted by the behavior retrieval unit. Furthermore, a separation unit that separates the input behavior data for each user behavior;
A feature amount extraction unit that extracts the action data feature amount carved by the carving unit;
A feature quantity non-dimensionalization unit that non-dimensionalizes the feature quantity for each user action and extracts a feature parameter;
The user behavior induction system according to claim 1, wherein the behavior classification processing unit classifies behavior data for each similar behavior from a position vector in a multidimensional space with the feature parameter as an axis. Furthermore, it comprises a user behavior classification database for storing user behavior data classified by behavior similar to user behavior in the behavior classification processing unit,
The user behavior induction system according to claim 1, wherein the similarity search unit searches a user behavior similar to a user behavior to be induced from user behavior data stored in the user behavior classification database. And a causal relation correspondence database for storing the occurrence probability of the user behavior classified by the behavior classification processing unit for the behavior data of the robot input by the data input unit obtained by the causal relationship estimation unit,
The user behavior induction system according to claim 1, wherein the behavior search unit extracts robot behavior data based on a probability of occurrence of robot behavior data corresponding to a user behavior in the causal relationship correspondence database.   2. The user behavior induction system according to claim 1, further comprising a robot behavior storage unit that stores robot behavior data corresponding to the generated user behavior classified by the behavior classification processing unit in a robot behavior database. .   6. The user according to claim 5, wherein the robot behavior processing unit extracts the behavior data of the robot extracted by the behavior search unit from the behavior data stored in the robot behavior database, and causes the robot to execute the data. Action induction system.   6. The database according to claim 3, wherein the robot behavior processing unit causes the robot to execute behavior data, and then inputs user behavior data from the sensor through the data input unit, and updates the database. User behavior induction system. A data input step for inputting behavior data of the user and the robot from the sensor;
An action classification step for classifying each action similar to the user action from the user action data input in the data input;
A causal relationship inference step for obtaining the occurrence probability of the user behavior classified in the behavior classification step corresponding to the behavior data of the robot inputted in the data input step;
An instruction input step for inputting a user action to be induced;
A similar search step of searching for a user behavior similar to the user behavior to be induced from the user behavior classified in the behavior classification step;
An action search step of extracting robot action data based on the occurrence probability of the action data of the robot corresponding to the user action searched in the similarity search step;
And a robot action step for causing the robot to execute the robot action data extracted in the action search step.

Images