JP2002219677A - Robot system, and method of controlling behavior of robot system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To exhibit a behavior for acquiring information in an automonous behavior. SOLUTION: This robot system changes emotional movement and the like in a feeling part 130 to exhibit the behavior for acquiring the information as the automonous behavior, based on information obtained by a sensing part 120.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a robot device and a method for controlling the behavior of such a robot device.

[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. The robot device behaves like an animal in accordance with, for example, external information or an internal state (for example, an emotional state). Some of such robot devices take actions such as kicking a ball. Further, some of such robot devices have a learning function. For example,
The learning function includes a language learning function.

[0003]

One ultimate goal of a pet-type robot device is to construct a life-like robot device.
This is, in other words, OPEN-ENDE
D) It can be said that a system is constructed.

[0004] Conventionally, how to show complexity increases
So we are approaching the goal of constructing a life-like robot device. Among them,
Factors such as learning and growth are also considered.

However, the implementation is realized by changing the probability of a stochastic state machine fixedly set for generating an action by reinforcement learning or replacing the entire state machine.

[0006] The generation frequency and pattern of the behavior change due to the interaction with the user and the outside world, and the complexity of the robot device is increased, but the behavior and reaction exceed the behavior and reaction initially assumed by the designer (programmer). is not. Similarly, the objects that can be recognized by the robot device are limited, so that it is not possible to recognize objects other than those set by the programmer. These imply a lack of ability to determine how to act on unknown stimuli.

[0007] What is needed to create an open-ended system that exceeds the settings of the designer is as follows. (1) Ability to recognize unknown stimuli (2) Ability to create new behaviors (3) Ability to select appropriate behaviors for unknown stimuli Furthermore, considering that the robot is a pet type robot, The ability to interact is particularly important. It is also a fact that when learning various unknown things, they often learn through interaction with humans.

[0008] The most important thing in the interaction with humans is communication by language.
Regarding the ability of (1) for recognizing an unknown stimulus, the first step is to acquire an appropriate categorization and the name of the symbol (Symbol) or to acquire the name of the action. . Although this is a research field called so-called language acquisition, it is pointed out that it is particularly important that those symbols are physically associated or grounded.

Regarding this, for example, a report by Kaplan (Kaplan, F. Talking AIBO: Firstexperimentation of
verbal interactions with an autonomous four-legge
d robot.In proceedings of the CELE-Twente worksho
p on interacting agents, October, 2000, hereinafter referred to as Reference 1. ), Reported by Roy et al. (Roy, D. and Pentlan
d A. Learning words from natural audio-visual inpu
t, in proceedings of International Conference on S
poken Language Processing, 1998; ) Or reports by Steels (Steels, L. Percept
ually GroundedMeaning Creation, In proceedings of
the International Conference on Multi-Agent System
s, 1996; ).

[0010] The above-mentioned behavior acquisition (2) is based on imitation, reinforcement learning, and Evolutiona.
One by ry Computing.

In this regard, a report by Damasio (Dam
asio, A. Descartes' Error: Emotion, Reason, and th
e Human Brain, Putman Publishing Group, 1994; ) And reports by Mataric (Mataric,
M., Sensory-motor primitivesas a basis for imitati
on: Linking perception to action and biology to ro
botics, Imitation in Animals and Artifacts, C. Neh
niv and K. Dautenhalm (eds), The MIT Press, 2000,
Hereinafter, this is referred to as Document 5. ).

However, as for the appropriate action to the unknown stimulus of the above (3), only very primitive ones have been reported in the real world. Or there are only a few related things in the virtual world.

The meaning of the above (3) lies in how to obtain the meaning of the object with respect to the robot apparatus. For example, is it food, play equipment, scary, etc. For this purpose, in addition to physically associating or grounding the recognition target, how to affect the internal state of the robot apparatus, that is, the internal state (for example, primary emotion, secondary emotion, etc.) Needs to be grounded.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and has as its object to provide a robot apparatus that is closer to life-like and a method of controlling the behavior of such a robot apparatus. The purpose is.

[0015]

In order to solve the above-mentioned problems, the robot apparatus according to the present invention includes a behavior control means for causing an information acquisition behavior as one of the autonomous behaviors. The robot device having such a configuration expresses the information acquisition behavior as one of the autonomous behaviors.

Further, in order to solve the above-mentioned problem, the behavior control method for a robot device according to the present invention causes an information acquisition behavior as one of autonomous behaviors of the robot device. With such a behavior control method for a robot device, the robot device expresses the information acquisition behavior as one of the autonomous behaviors.

Further, in order to solve the above-mentioned problems, the robot apparatus according to the present invention includes a meaning acquiring means for acquiring a meaning of the object. The robot device having such a configuration acquires the meaning of the target object.

According to another aspect of the present invention, there is provided a method for controlling a behavior of a robot apparatus, the method comprising: controlling an internal state when a robot apparatus acting based on the internal state acts on an object; Is obtained as the meaning of the object.

According to such a behavior control method for a robot device, the robot device behaves based on an internal state,
A change in the internal state when an action is taken on the object is acquired as the meaning of the object.

According to another aspect of the present invention, there is provided a robot apparatus comprising: a voice input unit; a plurality of word sequence feature models classified based on a feature amount of a word sequence when uttered; Utterance input evaluation means for evaluating the utterance input made by the voice input means based on the word sequence feature model; and word sequence identification means for identifying the utterance input word sequence based on the evaluation value of the utterance input evaluation means. Prepare.

The robot device having such a configuration is as follows.
The utterance input made by the voice input means is evaluated by the utterance input evaluation means based on a plurality of word sequence feature models divided based on the feature amount of the word sequence at the time of utterance, and the evaluation value of the utterance input evaluation means is evaluated. Then, the word sequence of the utterance input is specified by the word sequence specifying means. Accordingly, the robot device specifies the input utterance as an optimal word sequence.

Further, in order to solve the above-mentioned problems, the behavior control method of the robot apparatus according to the present invention is characterized in that a speech input step and a speech sequence input in the speech input step are characterized by the characteristics of a word sequence when the speech is made. An utterance input evaluation step of evaluating based on a plurality of word sequence feature models classified based on the amount, and a word sequence identification step of identifying a word sequence of the utterance input based on the evaluation value obtained in the utterance input evaluation step And With such a behavior control method for a robot device, the robot device specifies the input utterance as an optimal word sequence.

Further, in order to solve the above-mentioned problems, the robot apparatus according to the present invention is provided with control means for controlling the behavior of pointing to the subject to be learned. A robot device having such a configuration acts to point to its own learning target.

Further, in order to solve the above-mentioned problems, the behavior control method for a robot device according to the present invention controls the behavior of the robot device so that the robot device that performs autonomous behavior points to its own learning target. According to such a behavior control method for a robot device, the robot device performs a behavior pointing to its own learning target.

According to another aspect of the present invention, there is provided a robot apparatus comprising: a sensor for detecting an object;
A sensory evaluation unit that evaluates an input signal from the sensor, an internal state management unit that receives an evaluation result of the sensory evaluation unit, and manages a pseudo internal state that changes based on the evaluation result; Storage means for storing a relationship with a change in the internal state based on the object.

The robot device having such a configuration is as follows.
When an object is detected, the change in the internal state based on the detected object and the object are stored in the storage unit in association with each other.

Further, in order to solve the above-mentioned problems, a behavior control method for a robot apparatus according to the present invention includes a perceptual evaluation step of evaluating an input signal from a sensor for detecting an object;
It has an internal state management step of managing a pseudo internal state that changes based on the evaluation result in the perceptual evaluation step, and a storage step of storing the relationship between an object and a change in the internal state based on the object in storage means. . With such a behavior control method for a robot device, when a target object is detected, the robot device stores the change in the internal state based on the detected target object and the target object in the storage unit in association with each other. I do.

[0028]

Embodiments of the present invention will be described below in detail with reference to the drawings. This embodiment is an autonomous robot device that performs an autonomous action according to the surrounding environment (or external stimulus) or the internal state.

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

(1) Configuration of Robot Apparatus According to the Present Embodiment As shown in FIG. 1, a so-called pet-type robot having a shape imitating an animal such as a "dog" is provided. The unit units 3A, 3B, 3C, and 3D are connected, and the head unit 4 and the 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 each of 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, audio data, and remaining battery data stored in the DRAM 11 in this manner are thereafter transmitted to the robot apparatus 1 by the CPU 10.
It is used when controlling the operation of.

Actually, at the initial stage 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.

[0039] In addition, CPU 10 is configured to determine a subsequent action based on the control program 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 and 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, thereby outputting an audio based on the audio signal to the outside. LE above
D is turned on, off or blinked.

In this way, 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 allowed 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 device 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 includes 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 50-58. 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 n} 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.

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 this 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 column of “input event name”, and further conditions for the transition conditions are listed in the columns of “data name” and “data range”. It is described in the corresponding line.

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.

In the state transition table 80, the names of nodes that can transition from the nodes NODE ₀ to NODE _n are listed in the row of “transition destination node” in the column of “transition probability to another node”. Input event name ",
Other nodes NOD that can transition when all the conditions described in the columns 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 determination result is output to the action switching module 71.

The action switching module 71 shown in FIG. 5 includes the action models 70 ₁ to 70 _n of the action model library 70.
Among the actions which are output from, select an action that is output from a predetermined higher priority behavior model 70 ₁ to 70 _n, the command to the effect that execute the action (hereinafter, referred to as an action command )) Middleware
The output is sent to the output semantics converter module 68 of the layer 40. Incidentally, in this embodiment, the behavior model, labeled on the lower side in FIG. 6 70 ₁ to 70 _n
The higher the priority, the higher the priority.

The action switching module 71 notifies the learning module 72, the emotion model 73, and the instinct model 74 that the action has been completed, based on the action completion information provided from the output semantics converter module 68 after the action is completed. .

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 (scorched)" 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 specific recognition results such as “hit” and “stroke” given from the input semantics converter module 59 and the elapsed time and action switching module 71. Update periodically based on notification from

Specifically, the emotion model 73 is a predetermined calculation 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 equation 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.

[0067]

(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 feedback information of the action (action completion information), information of 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 is output from the action switching modulator 71 (the action to which the emotion is added).
May be performed by

On the other hand, the instinct model 74 is “exercise greed (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 the parameter values of these desires 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, based on the recognition result, the elapsed time, the notification from the output semantics converter module 68, and the like, for “exercise desire”, “affection desire”, and “curiosity”. 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”.

[0073]

(Equation 2)

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, and the coefficient k
_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 “pleased” provided by 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.

As described above, 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) Application of the Present Invention to a Robot Apparatus The technique described here is a technique which is a principle for applying the present invention to a robot apparatus.

(3-1) Outline of System Structure First, acquisition of emotionally related symbols (Emotinally Gounded Sym)
The outline of the system structure that realizes bol acquisition is described.

Here, first, the following problems are raised when constructing a system, and the system to which the present invention is applied solves this problem, and a system for a life-like robot apparatus which could not be achieved conventionally. We are trying to make it happen. (Req-1) How to embed language acquisition behavior in an autonomous behavior system such as the robot device 1. (Req-2) Emotionally Grounded S
ymbol). (Req-3) How to categorize objects in the real world. (Req-4) How is between the robot device 1 and a person
Do you pay attention to the same object? That is, joint attention (Sh
ared Attention).

As described above, a problem was first raised. First,
For (Req-1), an ethological model (Ethologi
The problem was solved by integrating the autonomous behavior generation by the physical model and the method of physically grounded symbol acquisition (Physically Grounded Symbol Acquisition).

Here, an ethological model (Ethologica
l Model), for example, is based on the report of Arkin et al. (Arkin, RC, Fujita, M., Takagi, T., and Hasegaw
a, R.Ethological Model ..., submitted to ICRA-200
1, hereinafter referred to as Reference 6. ) And Bates reports (Bates, J.
The nature of character in interactive worlds and
the oz project.Technical Report CMU-CS-92-200, Ca
rnegie Mellon Unversity, Oct. 1992; )).

Further, physically related symbol acquisition (Physically
The method of Grounded Symbol Acquisition) is, for example, a technique proposed in the above-mentioned Documents 1, 2, and 3.

In particular, as one of the autonomous actions, an information obtaining action is defined as an action that satisfies the hunger of information, and the information obtaining action such as “eating” information is defined as a subsystem similar to the action of eating food. Has been realized.
The information to be acquired is the name and meaning of the object.

Here, the subsystem is the robot device 1
A robot device 1
Has several types of this subsystem depending on the type of action. The subsystem is determined mainly based on perception and internal state.

Further, regarding the emotionally grounded symbols (Req-2) described above, a solution is obtained by associating the change of the internal state that generates the motive of the action with the input and action at that time. I do. Specifically, it is not the internal state itself when there is an input,
By associating the change of the internal state with the input, it is possible to associate the meaning of the object with the individual and the emotional recall when the internal state is satisfied.

As for (Req-3), the object is recognized (Categorize) by perception (Perception), and a statistical model or the like is used as a categorizer (Categorizer) of colors and the like detected as perception. Appropriately categorizes real-world recognition objects.

For example, a report proposed by El-Nasr et al. (El-Nasr, M., Loeger, T., and Yen, J., PETEEI) proposed in a virtual world built in a computer: A
Pet with Evolving Emotionaly Intelligence, in pro
ceedings of InternationalConference on Autonomous
Agents, 2000, hereafter referred to as Reference 8. The difference from the Synthesis Creatur proposed in (2) is that the robotic device must be able to operate in the real world. In the real world, objects such as colors and shapes are continuously distributed in each feature space. Furthermore, just by looking at it, you cannot know what it actually means unless it is programmed in advance. Therefore, regarding the above (Req-3), the categorizer of perception (Categorize)
Solved using a statistical model as r).

Regarding the above-mentioned joint attention problem of (Req-4), this ethological model (Ethological Model)
Take action focusing on a certain object in the action selection of
The problem is solved by executing in a natural form using the part.

Symbol acquisition in the real world (Symbol Acq
An important feature in uisition is what is called Shared Attention or Joint Attention. Bru
ner's report (Bruner, J. Learning how to do things wi
th words, in J. Bruner and A. Garton (Eds.) Human g
rowth and development, Wolfson College Lectures, Cl
aredon Press, 1978; In (1), cognitive psychology and others point out that shared attention plays an important role in the learning process of young children. For example, the ability of an infant to naturally look at the direction in which the finger is pointed or the direction of the line of sight, so that the instructor and the instructor can share the target object.

[0092] Such joint attention is naturally incorporated into autonomous behavior by using the part of taking action focusing on a certain object in the behavior selection of the ethological model (Ethological Model). ing.

Hereinafter, ethological studies (Et.
Behavior control considering behavioral study (Behavior Cont)
rol). For example, the above-mentioned literature 6 is mentioned as a motion control technology of ethology.

Next, the whole structure and how to integrate the information acquisition behavior into the software of the autonomous robot device 1 will be described, and a solution method for joint attention will be described.

(3-2) Association between external stimulus and internal state (Fusion of External Stimuli and Internal Variab)
les) The important points in the ethological model are that behavior is selected by both internal states and external stimuli, and that motivation generation from internal states and motivation from external stimuli are important. Perceptual stimulus (Releas
e signal) are evaluated independently, and the action evaluation value (Be
havior Value) is associated (fusion) when it is generated. As a result, it is possible to generate a homeostasis behavior that stops the internal state within a certain range. Here, the homeostasis behavior is, for example, a behavior expressed so as to keep the internal state constant.

FIG. 9 shows a configuration for independently evaluating an internal state and an external stimulus corresponding to a certain action. FIG. 10 shows a configuration for realizing homeostasis behavior. Specifically, the behavior is systematically configured, the external environment is obtained to obtain an external stimulus, and the internal state is kept constant. 4 shows a configuration for selecting an action to be kept.

FIG. 9 will be described by taking the case of ingestive behavior as an example. Motivation generator (Mo
The tivation creator) 101 evaluates and outputs a motivation value (motivation value) of a feeding behavior from an internal state such as the degree of hunger. On the other hand, the release mechanism (release
mechanism) 102 is an external stimulus related to eating behavior,
For example, if there is food, the perception signal of its eating behavior (releas
e signal) and output. Then, the motivation value and the release signal or the perception signal (release signal) are evaluated independently.

Behavior evaluation unit (Behavior evaluator) 103
Then, a motivation value (motivation value) and a release signal (release signal) are evaluated, and an evaluation value of the action itself is output as an action evaluation value (behavior value). There are a plurality of such actions, each of which independently calculates a behavior evaluation value (behavior value) and outputs it to an action selection unit (Action selection) described later. In the action selection unit (Action selection), the action with the highest evaluation is selected and executed.

Here, the motivation generation section (motivati
On creator) 101 is defined to be an action that can return to the original range when the internal state deviates from an appropriate range, so that if the object of the action exists in the outside world, A general behavioral definition of acquiring is realized, thereby realizing homeostasis behavior.

(3-3) Construction of Emotionally Grounded Symbol Acquisition Architec
ture) Emotionally Grounded Symbol
Acquisition) realizes information acquisition behavior for unknown objects as part of autonomous behavior. Emotionally Grounded Symbol Acquisit
ion) is realized as a part of the autonomous action, for example, as shown in FIG. The points in this system construction are as follows. (I) A categorizer for each channel that can determine whether the input is unknown or known. (Ii) Associative memory for storing the result of categorization of each channel at the timing when the internal state changes. (Iii) Ethological model of internal state and external stimulus (Eth
biological model).

The above is the point in the system construction. If the external stimulus is a known stimulus, an autonomous behavior based on a homeostasis behavior based on a normal ethological model is performed.

Also, an emotionally grounded symbol acquisition (Emotionally Grounded Symbol Acquisit
ion) also has a feature of memorizing what the object is important to in the internal state. In this regard, in this point, a normal physically-related symbol acquisition (Physically Grounded Symbol A) is performed.
cquisition).

As described above, the acquisition of emotion-related symbols (Emot
In ionally Grounded Symbol Acquisition, emotionally related (Emotionally Grounded) information is associated with the target object and the information is used as information. By associating the emotion with the target object in this way, any action (Action) can be performed for a new target object. Release mechanism (Releas)
e Mechanism).

Further, by having the change of the internal state as the associative memory in relation to the object, when the learned input is presented, the secondary emotion (secondary emot) is obtained from the associative memory.
ion) internal state (internal variables)
Can be output to generate secondary emotions. For example, joy or fear, emotion
Produce as

Thus, in response to seeing an object or the like, it becomes possible to create an expression as an emotional expression action, influence the action selection, or to modulate the action.

(3-4) Information acquisition behavior (Information Ea
ting Behavior) In order to realize the information acquisition behavior as a part of the autonomous behavior, there is a variable relating to the information acquisition desire as a factor of the internal state (hereinafter referred to as an information acquisition variable) as a model indicating the internal state. Defined subsystem (hereinafter referred to as an information acquisition behavior subsystem).

For example, the information acquisition behavior subsystem is defined as an internal model in which the information acquisition variable increases when the associative memory learns an unknown input and decreases with time. The information acquisition behavior subsystem generates motivation for the information acquisition behavior when the information acquisition variable becomes deficient.

Further, in this case, the release mechanism generates a release signal if the input (information) is unknown. As a result, an action of acquiring information can be generated as an association (fusion) between an internal state and an external stimulus, as in the case of eating food.

As a specific action expressed as an information acquisition action, as a typical example, when the desire to acquire information increases, an action to search for an unknown object occurs, and further, "What is this?" And taking question behaviors. In general, such an action is formed as a dialogue with a user.

By constructing such a system, it is possible to realize information acquisition by dialogue based on curiosity, and to embed such information acquisition behavior naturally in autonomous behavior. become. That is, the information acquisition behavior is realized as a new element of the interaction realized as an autonomous behavior in the robot device 1.

(3-5) Shared Attention Some systems include Shared Attention or Joint Attent.
Ion is embedded as a natural form. Information acquisition based on joint attention in the system structure (Informatio
n Eating) action is performed as follows.

As described above, the information acquisition behavior is determined by the association between the internal state and the external stimulus (fusion).
on selection) 116.

[0113] Release mechanism
m) The object that caused the release signal 102 is the target for acquiring information. Further, if this action is selected only from the feeling of hunger in the internal state, a search is performed, and as a result, an object becomes a target of the information acquisition action. The target for acquiring information in this way is a target for shared attention.

In the case of the robot center, that is, in the case of the information acquisition behavior caused by the hunger of the internal state, the robot apparatus 1 approaches the target, points the finger, and asks the question “What is this?” Shared Attention is achieved by having the object pay attention to the object.

On the other hand, if the user holds the initiative,
In other words, release mechanism
When the target is specified based on the release signal output by the controller 102, the robot apparatus 1 firstly alerts the user by moving a sound or an object. In response to this, it is assumed that the user asks the question "What is this?"
When an action acquisition action is selected by the finger or the question, the robot apparatus 1 specifies the object pointed by the finger as the target object. This allows shared attention to the same object (Shared) even when the user is in the initiative.
Attention) is achieved.

As described above, in the present invention, shared attention (shared attention) is part of the general idea that the system pays attention to what the internal state wants or the external stimulus is strong. ).

(3-5) Internal state change and emotion (INTERNAL
VARIABLES AND EMOTIONS) As shown in FIG.
Is roughly divided into a perceptual internal state unit 131, an internal state unit 132, and an emotion unit 133.

The first internal state section 132 manages the dynamics of the internal state itself. Nutrients, water, fatigue,
Curiosity and the like exist as pseudo variables (see FIG. 2).
3). However, these internal states may be other internal states found in living things or animals in addition to those described above. The internal state unit 132 monitors the state necessary for holding the individual, and detects that it deviates from an appropriate value. Further, the internal state unit 132 is a motivation generation unit (Motivation Creator) for an action required to keep the internal state constant, that is, to maintain homeostasis.
This is a part for transmitting a signal prompting an action necessary for maintaining the internal state.

The second internal state unit 131 for perception analyzes input from an internal sensor or an external sensor and inputs the analysis result to the internal state management unit. here,
In the case of an original animal, the sensor signal analysis corresponds to information on a meal, information on fatigue, and the like, which are detected from the ratio of sugar in blood. In the robot apparatus 1, the remaining battery level analysis or the like corresponds to this. However, in the robot apparatus 1, appropriate behavior (Acti
On), an input signal for keeping the internal state constant is generated.

The third emotion part 133 is a part for generating pleasantness and discomfort from changes in the internal state, and generating emotions corresponding to joy, anger and the like. This emotion part 1
33 is also called a secondary emotion, and generates a pleasant or unpleasant signal depending on how the internal emotion (this is called a primary emotion) is satisfied. Further, the emotion unit 133 generates emotions such as so-called joy, sadness, and anger from the pleasant / unpleasant signals and the arousal level, the conviction level, and the like. The secondary emotion is used for an operation for expressing an emotion, for example, generation of a facial expression, generation of an LED light pattern corresponding thereto, and the like.

This change in the internal state is used for the learning timing of the learning memory (associative memory) 140 as shown in FIG. In other words, learning is performed when the internal state changes significantly. In addition, the internal state and the emotional state are the actions (Be
havior) generation unit (Motivation C)
reator) and used as the cause of each behavioral motive.

(3-6) Perception of unknown stimulus (PERCEPTION F
OR UNKNOWN STIMULI) "Recognition" in the development of the robot device 1 in the real world
Has been a major challenge. In particular, in real-time recognition in a real environment, a major problem arises as to whether an input that changes due to various factors should be identified with already learned information or whether it is determined as a new stimulus.

In recent years, statistical pattern recognition (St.
atistical (or probabilistic) Pattern Classificatio
n) there. This is a recognition method that treats input samples distributed in the feature space as minimizing a risk function as a statistical problem and obtains parameters for the statistical problem. Hidden-Markov-Model, which is currently the mainstream for speech recognition described later
(Hereinafter referred to as HMM) is also a recognition method in this category, and is also a typical recognition method in image recognition.

In the present system, it is determined whether the input is an unknown object or a known object by using this statistical pattern recognition technique.

In the statistical pattern recognition, a probability or a likelihood is given as to whether or not an object is a prototype, and the probability or the likelihood is used to determine whether the object is an unknown stimulus or a known stimulus. It is carried out.
Furthermore, even if the distance in the feature space of a certain sensor channel is short and it is difficult to determine only with this channel, when using another channel, a significant difference is observed, and the adjustment of the discrimination parameter in the original space can be performed. You can do it.

(3-7) Learning Memory for Emotions (ASSOCIATIVE MEMORY WITH EMOTIONS) The learning memory (Associative Memory) is triggered by the change of the internal state by the output of each perceptual channel. This is for learning. Here, the learning means, specifically, the change of the internal state triggered as such and the change of the internal state,
That is, associative storage of the object that has affected the internal state is performed.

Here, the change in the internal state can be determined, for example, by referring to “(3-5) Internal state change and emotion (INTERNAL VARIABLES)” described above.
AND EMOTIONS) ”, the amount that can be actually sensed in sensor signal analysis (current consumed by the joints, etc.) and the amount that can be simulated (eating for simulated food). Detection). Here, the current consumed by the joint is determined by, for example, the number of times of operation and the like, and constitutes, for example, a factor of “fatigue”.

As an association, the perception channel (Percepti
on channel) prototype (prototyp)
Based on the number of e) and the probability or likelihood belonging to the prototype, the connection of simultaneously occurring events is learned. The phenomena here include so-called physically grounded symbols such as names of the objects sent through the action generation, and these are also acquired as learning.

Further, at the same time, the change of the internal state serving as a trigger and the action performed on the object are also stored in the associative memory. As a result, what kind of action is performed on the object and what kind of change in the internal state occurs is stored. Such learning is an emotion-related symbol (Emot
ionally Grounded Symbol).

Here, the emotion (Emotion) is directly referred to as a primary emotion because it is a change in the internal state. However, a secondary emotion can be caused by a change in the primary emotion. Because it is possible, it is a symbol (Symbol) that is related (grounded) to fear and the like.

(3-8) Subsystem and action (SUBSYS
TEMS the AND Behaviors) behavior, subsystem _(subsystem) 115 1 ~115 _n being a plurality of behavioral groups capable classified as shown in FIG. 10
It is controlled based on. Subsystems 115 ₁ to 1
_15n has a hierarchical structure and a tree structure, and the highest layer is an actual subsystem.

For example, in the study of ethology reported by Arkin et al. In the above-mentioned reference 6, a subsystem considered to be necessary and sufficient as canny behavior is listed. The feature of the technology reported in Reference 6 is that, as shown in FIG. 11, one of the subsystems, feeding behavior (Investigat)
ive) is defined as the action of eating information. For example, the ingestive subsystem is defined as eating food (electricity). This ideally creates an action that keeps the battery remaining in an internal state and keeps it within a certain range, and when the battery runs low, creates an action of searching for a charging place, charging desire, or automatic charging It is possible to generate motivation.

In the present system, such a concept is introduced in the information acquisition step, and an item corresponding to “learning amount of new information” is provided as an item of the internal state. Defines state dynamics. Then, in the dynamics of such an internal state, an action corresponding to the “learning amount” is generated as in the case of the battery. That is, for example, the robot apparatus 1 acts so as to keep the “learning amount” within a certain range, and searches for an unknown object, obtains new information if the “learning amount” decreases, If is present as an external stimulus, approach it and point your finger at "what is
(What is this?) ", or to generate an action of learning the name spoken by a person using associative memory. Here, the learning amount is, for example, the learning amount of the learning object. The learning amount is a change amount that is determined according to the feature, and is reduced over time.

Furthermore, if the name of the object was learned,
You can also define actions that capture what it means for the internal state. This can be realized by trying an action on the target object and associatively learning the action and the internal state change when the internal state changes.

(4) Application to an actual robot device (IMPL
(4-1) Structure of a four-legged walking robot device (Enhanced)
Four-legged Robot Platform) A four-legged robot device 1 on which the above-described system is mounted will be described. FIG. 13 illustrates an example of a network system including the robot device 1 as a configuration.

In this network system, the robot apparatus 1 includes a wireless LAN card (wireless LAN card).
d) By using 161, TCP / IP (Transmi
ssion Control Protocol / Internet Protocol).

The robot apparatus 1 includes, for example, a CPU having a characteristic of about 100 MIPS of MIPS R4XXX and a 16 MB main memory. And this robot device 1
Output primitive behaviors (basic posture t
ransition, to search an object, to track an objec
t, to close to an object, to kick an object.to ea
tan object, etc.) and a Speech object (object) to which a phonetic symbol string is input. In addition, the robot apparatus 1 has an L corresponding to the eye.
Commands for creating some expressions using the ED are also provided.

In such a robot apparatus 1, the system as described above is constructed, and the robot apparatus 1
For example, information acquisition behavior is expressed as a part of autonomous behavior.

Further, the same processing as the processing in the robot apparatus 1 can be executed on the workstation 163 by the network system to which the robot apparatus 1 is connected. For example, this makes it possible to confirm the operation of the robot device 1 on the workstation 163. The processing performed on the workstation 163 is performed as follows.

The robot apparatus 1 captures an image signal as input, and transmits the image to the access point 162 via the wireless LAN (wireless LAN) by the wireless LAN card 161. Then, the image is transmitted from the access point 162 to the Ethernet (ethern).
et) (registered trademark) to the workstation 163.

Further, similarly to the case of transferring an image from the robot apparatus 1 to the workstation 163,
Sensor detection information from the joint angle detection and the touch sensor, the acceleration sensor, and the like in the robot device 1 is transferred to the workstation 163. Further, for example, when processing is performed in the workstation 163 as described above, sound can be input using the microphone of the workstation 163 instead of using the microphone of the robot apparatus 1.

On the workstation 163, the above-mentioned Perception, Evaluati
on for internal variable, Behavior subsystem, acti
Execute on selection etc. These features are, for example, Li
An OPEN-R (a system provided by Sony Corporation) mounted on nux allows an OPEN-R object (OP
This is achieved by designing EN-R objects and connecting them freely on a network. For example, currently Matlab
It works with a mixture of programs and OPEN-R objects on Linux.

(4-2) Function of actual machine and experimental results (Implem
ented Functions and Experimental Results) By applying the present invention, the robot apparatus 1 is finally configured to express an information acquisition action or an emotion confirmation action by joint attention as a part of the autonomous action. . More specifically, as shown in FIG. 14, the robot apparatus 1 roughly divides the steps into an autonomous action (step S1), an input of an object (step S2), and an action selection (step S3). Then, an information acquisition action and an information confirmation action by joint attention are displayed (step S4). And the robot apparatus 1 processes such steps as part of the autonomous behavior.

(4-2-1) Perception Par
t) As shown in FIG.
Is provided within. Specifically, as shown in FIG. 12, the perception unit 121 is a color perception unit 1 for perceiving an image.
22 and a pattern perception unit 123, a contact perception unit (tactile unit) 124 for perceiving contact, and a voice perception unit 125 for perceiving sound.

More specifically, the color perception section 122 is a section for performing automatic color segmentation, which will be described later, based on information on the object, and the type perception section 123 is a section for analyzing the type of the object based on the image information. And the voice perception unit 12
Reference numeral 5 denotes an utterance recognition unit for utterance input from a microphone. The following description is about the processing performed in each of these perception units.

The contact perception unit 124 detects contact with an object by a signal from a so-called meat ball sensor provided on the sole of the robot device 1, for example.

(4-2-1-1) Automatic Color Segmentation Color segmentation using colors is first performed when a perceptual stimulus is input. In the color segmentation, it is possible to separate a plurality of objects having an arbitrary single color. In the color segmentation, a method based on a clustering algorithm based on unsupervised learning is used.

FIG. 15 shows an artificially painted object (A in the figure) and the result of the color segmentation (B in the figure). FIG. 16 shows a natural image including a human hand and a natural image including a human face (A in the figure).
And the result of the color segmentation (B in the figure).

Here, the input image has a relatively narrow viewing angle (5
(3 x 41 degrees) already low when input to the system from the camera
It has been reduced to 88 x 60 pixels through a pass filter. Taking this into account, segmentation is performed only by independent processing for each pixel. By doing so, good results as shown in B in FIG. 15 and B in FIG. 16 can be obtained almost in real time.

In general, color segmentation is often performed in RGB or normalized RGB space. However, since the camera signal is in Y, Cr, Cb format, (Nr, Nb) = (atan (Cr / Y), atan (Cb / Y)) is defined as a color space. This is a very efficient process in consideration of the amount of calculation that occurs when mapping to the RGB space and errors during quantization.

Note that such color segmentation is used as an initial process for shape analysis.

Hereinafter, examples of the processing steps (i) to (vi) of the clustering algorithm in the above-described color segmentation will be described.

In step (i), an appropriate number of prototypes (prototypes) are uniformly arranged.

In step (ii), the closest prototype to all pixels (pr
ototype).

[0155]

(Equation 3)

Here, σhue and σsat are distributions corresponding to hue and saturation, respectively, as shown in FIG. 17 and are obtained in advance from the distribution of appropriate sample images. Generally, σhue <σsat is there. That is, hu
It can be considered as a distance in which the error in the e direction is weighted.

In step (iii), if the number of pixels belonging to the prototype is small, the prototype is changed.

In step (iv), the prototype is moved to an average position with the same class label.

In step (v), if two or more prototypes are less than a certain distance, they are combined into one.

In step (vi), the process is terminated when the number of updates of the prototype position has decreased or when the number of updates has reached an appropriate number. Otherwise, the process returns to the above step (ii) and starts the process again.

FIG. 18 shows a state of clustering for an input image. In the example shown in FIG.
This figure shows a case in which a skin color area stored in advance is analyzed to detect a direction in which a finger is pointing and to expose an object on an extension thereof. For example, this information is used in Shared Attention described below.

(4-2-1-2) Type Analysis
is) Type analysis (Shape Analysis) is performed using a Fourier descriptor (FD), which is a universal feature of size and rotation. For example, in this type analysis, the categorization is Fourier
L2 norm in Descriptor space (64 dimensions) is used.
The input object is represented in the FD space, and it is determined whether a new prototype is to be used by using a distance from the closest prototype. FIG. 19 shows the result of shape analysis of an object cut out by color segmentation.

(4-2-1-3) Speech Recognition (Speech Rec
ognition) Continuous speech recognition using an HMM is used as speech recognition (Speech Recognition). As this technique, there is a technique proposed in the above-mentioned Document 5.

As shown in FIG. 20, this system includes a voice input unit 171, an HMM register 172 having a plurality of HMMs, an HMM 173 for inputting unknown words, and a comparison unit 174.

The HMM in the HMM register 172 is an HMM that has learned Japanese phonemes, and necessary words are registered in advance. Further, the HMM in the HMM register 172 includes those for which words acquired later have been learned. Here, for example, the registered or acquired words include nouns and verbs.
The input phoneme sequence is stored in such an HMM register 17.
2 is evaluated as a certainty factor in the HMM.

The unknown word input HMM 173 is an HMM for acquiring an unknown word. This unknown word input HMM 173
, As shown in FIG. 21, all phoneme models are states, and are connected to all phoneme states. For example, the HMM 173 for inputting an unknown word, as shown in FIG.
When the utterance input of “uruu” is made, “booru”
Recognize as

The input phoneme sequence includes the HMM of the word registered or acquired and the HMM 17 for inputting the unknown word.
3 and the confidence level (verifi
The comparison unit 174 evaluates the distance from the HMM that has the largest match using the cation value. If the certainty factor (verification value) is equal to or greater than a certain value, a new label is assigned as a new phoneme sequence, and the new label is registered as an HMM in the HMM register 172.

As an example, the HMM register 172 stores the HM
A case will be described in which, as M, only those in which two words “tomare (stop)” and “kere (kick)” are registered. FIG. 22 shows the result when such a system is used.

In FIG. 22, the right side shows the value of the certainty factor (verification) of the input signal for the registered word. The lower the value of the certainty factor (verification), the higher the certainty factor.

For example, for the utterance “tomare”,
The system estimates that the input is a phonological sequence “tomoare”, and the value of the verification is 0.136.

On the other hand, in FIG. 22, the third "bo
oru (ball) ”, the best matching model is“ tomare ”, and its confidence (verificatio
Since n) is so large as 4.835, a new symbol called unknown-1 is assigned and registered. This allows
For the next utterance input of “booru (ball)” shown in FIG. 22, which is the fourth utterance input, the HMM corresponding to unknown-1 is the closest, and its credibility (verification) is Take a small value of 0.41 and correctly un
"booru (ball)" will be acquired by known-1.

Also, in this system, since the HMM can recognize continuous utterances, the “boor kere” obtained in advance from “boru kere” as in the seventh utterance from the top in FIG.
Following the label unknown-1 for u ", it is possible to recognize the symbol kere.

With such an utterance recognition system,
For example, it is keel and rare. If the noun “ball” is acquired, the robot device 1 can kick the ball by the command “ball kick”.

(4-2-1-4) Emotion Par (Emotion Par)
t) FIG. 23 shows the relationship between internal states (Internal Variables) and actions (subsystems) related thereto.

In this example, referring to a physiological model of eating behavior and the like, in order to maintain a general internal state, a virtual internal nutrition storage buffer and an excretion buffer are assumed, and the storage amounts are stored in the internal state. Is defined as They are, for example, the amount of Energy-2 (fake food) and the amount of Fake excrement.

For example, as shown in FIG. 24, by associating a virtual stomach (intravital storage buffer) with a virtual bladder or intestine (excretion buffer), a decrease in the virtual stomach storage volume can be achieved. , To increase the amount of storage such as a virtual bladder.

As shown in FIG. 23, there is a dynamic that increases or decreases by a certain factor. The basic operation of the motivation creator (Motivation Creator) is to keep the internal state variables within a certain allowable range by motivating the corresponding action group (subsystem).
ion).

[0178] Fake food and water are considered to be mainly implemented for the purpose of entertainment of the robot apparatus 1, but they also correspond to electric energy and fatigue in the original sense. Some internal state variables exist. These also compose the dynamics by the increase and decrease factors shown in FIG. 23, and the corresponding subsystems
The motivation creator is designed to motivate the behavior to keep this constant. However, although an automatic charging action in which the robot apparatus 1 is attached to a so-called charging device as an autonomous action is also conceivable, if the robot apparatus 1 does not have such an automatic charging action,
The robot apparatus 1 performs an action requesting the charging, and causes another person (human) to charge.

Further, similar internal state variables are prepared for the information obtained in the associative memory. The amount of internally acquired information having a virtual meaning is calculated and transmitted by the associative memory. In this case, if there is no forgetting, the amount of internal information of the associative memory only increases, but forgetting may not be implemented. The motive of the information acquisition behavior subsystem is constructed by constructing simple dynamics of the integration factor of each information amount within an appropriate time range and the increase factor and the time decrease factor.

(4-2-1-5) Learning Memory Unit (Asso
FIG. 25 shows a specific configuration of a learning memory (Associative Memory) 140 used by the robot apparatus 1 for acquiring information. As shown in FIG. 25, the learning memory 140 includes a short-term memory 181, a long-term memory 182, and an attention target memory 183. The learning memory 140 is specifically provided as shown in FIG.

Learning Memory (Associative Memory) 14
0 functions as a storage unit having a certain name with a certain color and a certain shape by such a configuration, and furthermore, a storage unit indicating what meaning itself has with respect to the internal state of the robot apparatus 1. Function as

Short-term memory (Short Term Memory,
In (STM) 181, information on the ID numbered object in the image is stored. At this time, the information of the object is information of a color prototype number (CP-i) and a shape prototype number (SP-j). In addition, short-term memory (Short Te
rm Memory) 181 receives a word sequence for one utterance input from voice processing.

The data from the image is obtained by inputting the color prototype number (CP-i) and the shape prototype number (SP-j) and inputting the object name (HMM-k) and the effect on the internal state (Del).
ta-I) is obtained, and these are put together and sent to a behavior generator (Behavior Generator) 150 as shown in FIG. If the object name (HMM-k) and the effect on internal state (Delta-I) cannot be obtained, leave it blank (ni
l) Sent with information. The utterance data is sent to a behavior generator (Behavior Generator) 150 as it is.

On the other hand, an action selection unit (Action Selection) 1
At 16, the action and its target object (Obj-ID)
Is selected, but this information is stored in a learning memory (Asso
ciative Memory) 140. The information corresponding to this target object (Obj-ID) is stored in the short-term memory (ShortTer
m Memory) 181 to Attention Ob
ject memory (AOM) 183. At this time, the uttered word sequence stored in the short term memory (Short Term Memory) 181 is sent to the attention target memory (AttentionObject Memory) 183 as it is.

Attention Object Mem (Attention Object Mem)
ory) 183 from the original learning memory (Associative Mem
ory) as long term memory (Long Term Memor)
y) The learning timing for 182 is triggered by an internal state change as a trigger. Thus, when an internal state changes when a certain action is performed on a certain object (Object), the internal state change is stored in association with the target object.

(4-2-1-6) Behavior generator (Behavior
Generation Part) Here, the information acquisition behavior subsystem among the subsystems that regulate the behavior of the robot device 1 will be described. As shown in FIG. 26, the information acquisition behavior subsystem 1
51 _n is configured as having a hierarchical structure.

There is one software object at the behavior subsystem layer.

This software object (software
object) Motivation Creato
r) 101 is configured to output a motivation value when the above-mentioned primary internal storage amount is out of an appropriate range.

[0189]

(Equation 4)

On the other hand, the release mechanism (Release Mech
anism) 102 is a learning memory (Associative Memor)
y) This is done by checking the object sent from 140. Release mechanism (Release
Mechanism 102 considers what is currently unknown and the factors of perception (release) of human pointing. Here, for example, the perceptual factors are the name of the object (Object) (Obj: Name), the name of the Color (Color: Name), and the shape.
(Act: Delta-I) of the object (Object) and its influence on the internal state change of the object (Object).

Release mechanism (Release Mechanis
m) 102 is a release signal (Release sign
al). And the release mechanism (Releas
e Mechanism) 102 outputs a release signal (release
The value of signal) is determined corresponding to the object as an accumulation of an undefined number for one object. For example, only the name of the object (Obj: Name) and the effect on the internal state change (Act: Delta-I) can be targeted.

The release mechanism (Release Me
chanism) 102 evaluates a release signal for an existing object, selects an object having the largest value, and specifies an ID and a release for identifying the selected object (Obj). Output a signal (release signal).

For example, when an apple is specified as an object, the robot apparatus 1 analyzes the type and color using the above-described type analysis and color segmentation, and obtains the name of Color as a perceptual factor. (Color: Name), Sha
Evaluates the name of the pe (Shape: Name). When an apple is registered in advance, a high evaluation value is obtained as a result, and it is recognized that the target object is an apple. Then, an ID for identifying the selected target apple and a release signal at that time (releasesignal)
Is output. If an apple is not registered in advance, an undefined number is accumulated, and this is made to correspond to an apple as an unknown object.

On the other hand, the release mechanism (Release Mech
anism) 102 is configured to generate a larger release signal for human pointing. And release mechanism (Re
The lease mechanism 102 generates a release signal when an object is detected by pointing, regardless of whether the object is unknown or known. The idea is that the pointing is clearly a request for information acquisition or information confirmation from a human, and the user wants to induce the information acquisition action without largely depending on the internal state, or to perform the confirmation action on a known thing. It is due to.

Then, the information acquisition behavior subsystem 151
_nNow, this release signal and the mochibe
Multiplied by the motivation value
Obtained as a behavior evaluation value. Also,
Then, each other subsystem (subsys
tem), the information acquisition behavior subsystem 151 _n
Release signal and mochibe input to
Behavior evaluation using the motivation value
Get the behavior value.

Then, in the action selecting section 116, the action evaluation value (behavior
value) and compare the largest behavioral evaluation value (behavior va
lue) is selected as the subsystem to be executed. In the description here, the information acquisition behavior subsystem 1 is obtained by comparing such evaluation values.
51n is the case _where the behavior evaluation value (behavior value) is maximized.

The selected subsystem (subsyste
m) needs to be selected for a while, but this can be achieved, for example, by mutual suppression or fatigue factor.

When the information acquisition behavior subsystem 151 _n is selected, as shown in FIG.
Proceed to a hierarchy called D. In the mode (Mode) MD, information selection processing is performed similarly. Specifically, in the mode (Mode), the selection of the upper layer is performed by pointing to an object by pointing,
An object selected by itself, that is, an unknown object is distinguished. In this Mode MD, when a distinction is made, as shown in FIG. 26, a specific behavior is evaluated in a layer called Module MJ which is a lower layer. Based on this evaluation, a specific action is selected in the action selecting unit 116.

As a result, the robot apparatus 1
If the target object is known, the confirmation action is taken. If the target object is unknown, the acquisition action is taken. For example, the information acquisition behavior affects the name of the object (Obj: Name) and the effect on the internal state change (Act: Delta-In
When two types of information are obtained in t), the object having the maximum value evaluation in the subsystem may be examined and either one may be selected.

For example, as a process for executing the confirmation action, an instruction is sent to the state machine corresponding to the confirmation action, and the confirmation action of the name is executed. And
The robot apparatus 1 approaches the object pointed by the finger with the visual tracking, and then points the object with the finger, that is, the front leg, and says, "Is this XX?" Such actions are expressed.
Such an action is realized by controlling with a state machine in which an action sequence that defines such an action is described.

As the processing for obtaining the object name (Obj: Name), the output is sent to the state machine for obtaining the corresponding object name (Obj: Name). In the acquisition of the name of the object (Obj: Name),
Visual tracking for the object
ing), approach, and point your finger at it, expressing an action such as "What is this name?" At this time, appropriate behavior control is performed using the distance to the target. Such an action is realized by controlling with a state machine in which an action sequence that defines such an action is described.

After "What is this name?", If there is a valid output from the utterance recognition unit that has been input, it is possible to incorporate a state machine that repeatedly checks the phoneme sequence. .

On the other hand, the effect on the internal state change (Act: Delta-
When an acquisition action by Int), that is, an acquisition action on an unknown object based on a change in internal state is selected, some actions are randomly selected and executed on the object. Then, the effect on the internal state change (Delta-Int) that occurs at that time is evaluated by the associative memory. This allows
Since this object is associated with the internal state (grounding), the influence of the new object on the internal state change is obtained as meaning acquisition.

For example, if an internal state changes to "pleasant" at the time of looking at an apple, the change in the internal state is made to correspond to the apple as an object. Thereafter, the robot apparatus 1 interprets the apple as having a pleasant meaning, which means that the robot apparatus 1 has acquired the meaning of the apple. As described above, by applying the present invention, the robot apparatus 1 causes the information acquisition behavior to be expressed as a part of the autonomous behavior, executes the joint attention as the optimal behavior, and furthermore, In an information acquisition behavior, the meaning of an unknown object obtained as a change in the internal state can be acquired. Thereby, the robot device 1 becomes closer to life-like.

The application of the present invention to the robot device 1 as described above can be realized by, for example, software.

[0206]

The robot apparatus according to the present invention can exhibit the information acquisition behavior as one action of the autonomous action by providing the action control means for causing the information acquisition action as one action of the autonomous action.

Further, in the behavior control method for a robot apparatus according to the present invention, the robot apparatus performs the information acquisition behavior as one of the autonomous actions of the robot apparatus. Can be revealed.

Further, the robot apparatus according to the present invention includes meaning obtaining means for obtaining the meaning of the object,
The meaning of the object can be acquired.

In the behavior control method for a robot device according to the present invention, a change in the internal state as a meaning of the object when the robot device acting on the basis of the internal state acts on the object. By acquiring, the robot apparatus can act based on the internal state, and can acquire a change in the internal state when acting on the object as the meaning of the object.

[0210] The robot apparatus according to the present invention also includes a speech input unit, a plurality of word sequence feature models classified based on the feature amount of the word sequence at the time of utterance, and a speech input performed by the speech input unit. By providing utterance input evaluation means for evaluating based on the word sequence feature model, and word sequence identification means for specifying the word sequence of the utterance input based on the evaluation value of the utterance input evaluation means, the speech input means The utterance input is evaluated by the utterance input evaluation means based on a plurality of word sequence feature models classified based on the feature amount of the word sequence at the time of the utterance, and the utterance input is performed based on the evaluation value of the utterance input evaluation means. Can be specified by the word sequence specifying means. Thereby, the robot device can specify the input utterance as an optimal word sequence.

Further, the behavior control method of the robot apparatus according to the present invention is characterized in that a voice input step and a plurality of utterance inputs made in the voice input step are classified based on a feature amount of a word sequence at the time of the utterance. The robot apparatus has an utterance input evaluation step of evaluating based on the word sequence feature model, and a word sequence identification step of identifying a word sequence of the utterance input based on the evaluation value obtained in the utterance input evaluation step. The input utterance can be specified as an optimal word sequence.

Further, the robot apparatus according to the present invention is capable of performing an action pointing at its own learning target object by providing a control means for controlling an action pointing at its own learning target object. As a result, joint attention between the robot device and the user is ensured.

Further, the behavior control method for a robot device according to the present invention controls the behavior of the robot device so that the autonomously acting robot device points to the object to be learned by itself. Be able to point to things. As a result, joint attention between the robot device and the user is ensured.

Further, in the robot apparatus according to the present invention, a sensor for detecting an object, a perceptual evaluation section for evaluating an input signal from the sensor, and an evaluation result of the perception evaluation section are input, and the evaluation result is input to the evaluation result. When an object is detected by providing an internal state management unit that manages a pseudo internal state that changes based on the object and a storage unit that stores a relationship between the object and a change in the internal state based on the object. In addition, the change in the internal state based on the detected object and the object can be stored in the storage unit in association with each other.

Further, the behavior control method of the robot apparatus according to the present invention includes a perceptual evaluation step of evaluating an input signal from a sensor for detecting an object, and a pseudo internal state which changes based on the evaluation result in the perceptual evaluation step. The robot apparatus has an internal state management step of managing the object and a storage step of storing the relationship between the target object and a change in the internal state based on the target object in a storage unit. ,
The change in the internal state based on the detected object can be stored in the storage unit in association with the object.

[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 block diagram showing a component for selecting an action.

FIG. 10 is a block diagram showing a component for selecting an action by perception.

FIG. 11 is a diagram illustrating a specific example of a subsystem.

FIG. 12 is a block diagram showing a more specific component of a configuration for selecting an action.

FIG. 13 is a flowchart showing a series of procedures until the robot device expresses an information acquisition action or an information confirmation action by joint attention.

FIG. 14 is a diagram illustrating a configuration of a network system including a robot device.

FIG. 15 is a diagram used for describing color segmentation of an input image composed of an arbitrary single color.

FIG. 16 is a diagram used for describing color segmentation of an input image including a human.

FIG. 17 is a diagram used for describing clustering of color segmentation.

FIG. 18 is a diagram illustrating a state of clustering of an input image.

FIG. 19 is a diagram showing a result of an outline analysis cut out by color segmentation.

FIG. 20 is a block diagram showing components for realizing speech recognition.

FIG. 21 is a diagram illustrating a configuration example of an unknown word input HMM.

FIG. 22 is a diagram illustrating a result of speech recognition.

FIG. 23 is a diagram showing information on an internal state.

FIG. 24 is a diagram showing a relationship between a virtual stomach and a virtual bladder and the like.

FIG. 25 is a block diagram illustrating a configuration of a learning memory.

FIG. 26 is a diagram showing a process from the information based on the external stimulus and the internal state until the information acquiring action or the information confirming action by joint attention is expressed.

[Explanation of symbols]

Reference Signs List 1 robot apparatus, 10 CPU, 101 motivation generation unit, 102 release mechanism, 103 operation evaluation unit, 111 perception unit, 112 internal state unit for perception,
113 internal state unit, 114 emotion unit, 115 subsystem, 116 action selection unit, 140 learning memory

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Rika Horinaka 6-35-3 Kita-Shinagawa, Shinagawa-ku, Tokyo Inside Sony Corporation (72) Inventor Jun Jun-Yokono 6-35, Kita-Shinagawa, Shinagawa-ku, Tokyo Inside Sony Corporation (72) Inventor Gabriel Costa 6-35, Kita-Shinagawa, Shinagawa-ku, Tokyo Sonny Corporation (72) Hideki Shimomura 6-35, Kita-Shinagawa, Shinagawa-ku, Tokyo Soni -In-house (72) Inventor Kiki Minamino 6-35, Kita-Shinagawa, Shinagawa-ku, Tokyo F-term in Sony Corporation (reference) 2C150 CA01 CA02 DA05 DA24 DA26 DA27 DA28 DF03 DF04 DF06 DF33 ED42 ED07 ED52 EF07 EF16 EF23 EF28 EF29 EF33 EF36 3C007 AS36 CS08 KS10 KS31 KS36 KS39 KT01 LW12 MT14 WA04 WA14 WB00 WB15 WB19 5D015 KK01

Claims (31)

[Claims] 1. A robot apparatus that performs an autonomous action, comprising: an action control unit that performs an information acquisition action as one action of the autonomous action. 2. The information acquisition action according to claim 1, wherein the information acquisition action is a language acquisition action for acquiring a language as information.
The robot device as described. 3. The robot according to claim 1, wherein the robot performs an autonomous action based on at least an internal state, and the action control unit causes the information acquisition action based on a homeostasis of the internal state. apparatus. 4. An information acquisition model in which the information acquisition desire as a change factor of the internal state is provided as a parameter, wherein the action control means is configured to perform the operation when the information acquisition desire parameter reaches a predetermined threshold value. 4. The robot apparatus according to claim 3, wherein the information acquisition action is expressed. 5. The robot apparatus according to claim 4, wherein the parameter of the information acquisition desire changes with time. 6. The robot apparatus according to claim 4, wherein the parameter of the desire to acquire information indicates a feeling of hunger for learning. 7. The robot device according to claim 4, wherein the parameter of the information acquisition desire is an information acquisition amount. 8. The robot apparatus according to claim 4, wherein the parameter of the desire to acquire information is a learning amount of new information. 9. The learning amount is a learning amount decreasing with time, and the behavior control means causes the information acquisition behavior to be expressed when the learning amount falls below a predetermined threshold. The robot device according to claim 8, wherein 10. A behavior control method for a robot device, wherein an information acquisition behavior is performed as one of the autonomous behaviors of the robot device. 11. A robot apparatus comprising a meaning obtaining means for obtaining a meaning of an object. 12. An action based on an internal state, wherein the meaning obtaining means obtains a change in the internal state as a meaning of the object when the action on the object is performed. The robot device according to claim 11, wherein 13. An internal state monitoring means for detecting the internal state, wherein the meaning obtaining means changes the internal state detected by the internal state monitoring means when the action on the object is performed. 13. The robot apparatus according to claim 12, wherein the robot acquires the meaning of the object. 14. The robot apparatus according to claim 11, further comprising associative storage means for associatively storing the meaning of the acquired object in association with the object. 15. The behavior of the robot device, wherein when the robot device acting based on the internal state acts on the object, the change in the internal state is acquired as the meaning of the object. Control method. 16. A speech input means, a plurality of word sequence feature models segmented based on a feature amount of a word sequence at the time of utterance, and an utterance input made by the speech input means are used as the word sequence feature model. A robot apparatus comprising: utterance input evaluation means for performing evaluation based on the utterance input evaluation means; and word sequence identification means for identifying a word sequence of the utterance input based on the evaluation value of the utterance input evaluation means. 17. The utterance input evaluating means includes: a feature amount detecting unit that detects a feature amount related to a word sequence of the utterance input performed by the voice input unit; and a feature amount of the utterance input detected by the feature amount detecting unit. 17. The robot apparatus according to claim 16, further comprising: an evaluation unit that performs evaluation based on the voice feature model. 18. A model registering means for registering the utterance input as a new speech sequence feature model using the feature amount of the word sequence when the feature amount of the utterance input detected by the feature amount detection unit is low. 18. The robot device according to claim 17, wherein: 19. The robot apparatus according to claim 16, wherein the word sequence feature model has a word sequence feature amount obtained by phonemic learning. 20. A speech input step, and an utterance input for evaluating the utterance input made in the speech input step based on a plurality of word sequence feature models classified based on a feature amount of the word sequence at the time of the utterance. A behavior control method for a robot apparatus, comprising: an evaluation step; and a word sequence specifying step of specifying a word sequence of the utterance input based on the evaluation value obtained in the utterance input evaluation step. 21. A robot apparatus which performs an autonomous action, further comprising control means for performing action control indicating a self-learning object. 22. The robot apparatus according to claim 21, further comprising control means for giving priority to an object specified based on an external stimulus and causing the object to act as a learning object. 23. The robot apparatus according to claim 21, wherein the action indicating curiosity indicates the object to be learned. 24. The control device according to claim 21, wherein the control means controls the behavior of pointing to the learning object by the legs. The robot device as described. 25. A behavior control method for a robot device, wherein the behavior of the robot device is controlled so that the robot device that behaves autonomously points to its own learning target. 26. A robot device which autonomously behaves, comprising: a sensor for detecting an object; a perceptual evaluation unit for evaluating an input signal from the sensor; and an evaluation result of the perceptual evaluation unit. An internal state management unit that manages a pseudo internal state that changes based on the evaluation result; and a storage unit that stores a relationship between the object and a change in the internal state based on the object. The robot device stores the change in the internal state based on the detected object and the object when the operation is performed in the storage unit. 27. The apparatus according to claim 27, further comprising an emotion unit configured to generate a pseudo emotion based on the change in the internal state, wherein the object and emotion-related information based on the object are stored in the storage unit. Item 27. The robot device according to item 26. 28. The apparatus further comprising an action generation unit, wherein the internal state management unit manages the internal state so as to maintain the constancy of the internal state. 27. The robot apparatus according to claim 26, wherein a first signal is transmitted, and the action generation unit generates an action for maintaining the homeostasis based on the first signal. 29. The robot apparatus according to claim 28, wherein the change in the internal state and the action for maintaining the homeostasis are stored in the storage unit in association with each other. 30. The internal state management unit has an information acquisition desire variable, and transmits a second signal to the action generation unit based on a value of the information acquisition desire variable. 29. The robot apparatus according to claim 28, wherein the information acquisition behavior is generated based on the second signal. 31. A behavior control method for a robot apparatus that performs an autonomous behavior, comprising: a perceptual evaluation step of evaluating an input signal from a sensor for detecting an object; and a pseudo-state changing based on an evaluation result in the perceptual evaluation step. An internal state management step of managing an internal state, and a storage step of storing the relationship between the object and a change in the internal state based on the object in a storage unit. A behavior control method for a robot apparatus, wherein a change in the internal state based on a detected object and the object are stored in the storage unit in association with each other.