Apparatus and method for learning emotion of robot

1. An apparatus for a robot capable of learning emotion, the apparatus comprising:

an internal state calculation unit configured to detect external environment information and calculate an internal state vector value, corresponding to one point of an internal state coordinate system, based on the detected external environment information;

an emotion decision processing unit configured to generate a set of emotion values within each emotion value group, indicated by coordinates of an internal state input value of the robot on the internal state coordinate system, as an emotion of the robot;

a behavior generation unit configured to decide a behavior of the robot based on the set of emotion values decided by the emotion decision processing unit and make an emotional expression based on the decided behavior by performing a facial expression processing or a voice output processing of the robot, in which the robot comprises an emotional expression organ including at least one of a mouth, an eye, a head, an arm, a leg, and a body; and

an emotion learning unit configured to generate feedback information regarding the set of emotion values, wherein the set of emotion values corresponds to one point of the internal state coordinate system, based on the set of emotion values decided by the emotion decision processing unit and information regarding a type of emotion received from a user input unit and configured to change an emotion probability distribution of the internal state coordinate system using the feedback information,

wherein the emotion probability distribution comprises a first plurality of Gaussian distributions, each having weight, a central value, and a variance, or a second Gaussian distribution comprising the first plurality of Gaussian distributions, and

wherein the emotion learning unit is configured to retrieve a Gaussian distribution corresponding to the feedback information using a matching function and is configured to change the Gaussian distribution of the emotion probability distribution based on the retrieved Gaussian distribution.

2. The apparatus of claim 1 , wherein the matching function match j,m (k+1) satisfies the following equation:

match

j

,

m

⁡

(

K

+

1

)

=

{

1

,

if

⁢

⁢



x

⁡

(

k

+

1

)

-

μ

j

,

m

⁡

(

k

)



<

δσ

j

,

m

⁡

(

k

)

0

,

otherwise

,

wherein match j,m (k+1) means a matching function of an m th Gaussian distribution of a j th emotion distribution in a time k+1, δ indicates a matching range, σ j,m (k) indicates a standard deviation of the m th Gaussian distribution of the j th emotion distribution in a time k, x(k+1) indicates the feedback information, and μ j,m (k) indicates the central value of the m th Gaussian distribution of the j th emotion distribution in the time k.

3. The apparatus of claim 2 , wherein if the matching function is 1, a Gaussian distribution is changed so that weight ω j,m (k+1), a central value μ j,m (k+1), and a variance σ 2 j,m (k+1) are satisfied based on the feedback information as follows:

ω j,m ( k+ 1)=(1−λ)ω j,m ( k )+λ;

μ j,m ( k+ 1)=(1−ρ j,m ( k+ 1))μ j,m ( k )+ρ j,m ( k+ 1) x ( k+ 1);

σ 2 j,m ( k +1)−(1−ρ j,m ( k+ 1))ρ 2 j,m ( k ); and

+ρ j,m ( k+ 1)[ x ( k+ 1)−μ j,m ( k+ 1)] T [( k+ 1)−μ j,m ( k+ 1)]

wherein ω j,m (k+1) is the weight of the m th Gaussian distribution of the j th emotion distribution in the time k+1, λ is a predetermined basic weight, ω j,m (k) is the weight of the m th Gaussian distribution of the j th emotion distribution in the time k, μ m (k+1) is the central value of the m th Gaussian distribution of the j th emotion distribution in the time k+1, x(k+1) is the feedback information, μ j,m (k) is the central value of the m th Gaussian distribution of the j th emotion distribution in the time k, σ 2 j,m (k+1) is the variance of the m th Gaussian distribution of the j th emotion distribution in the time k+1, σ 2 j,m (k) is the variance of the m th Gaussian distribution of the j th emotion distribution in the time k, and [ ] T is a transposed matrix, and

ρ j,m ( k+ 1)=λη( x ( k+ 1)|μ j,m ( k ),Σ j,m ( k )), and

wherein Σ j,m (k) is a bandwidth of the m th Gaussian distribution of the j th emotion distribution in the time k.

4. The apparatus of claim 2 , wherein if the matching function is 1, an m th Gaussian distribution of an unselected j th emotion distribution is changed so that weight (ω j,m (k+1)) is satisfied based on the feedback information as follows:

ω j,m ( k+ 1)=(1−λ)ω j,m ( k ),

wherein ω j,m (k+1) is the weight of the m th Gaussian distribution of the j th emotion distribution in the time k+1, λ is the predetermined basic weight, and ω j,m (k) is the weight of the m th Gaussian distribution of the j th emotion distribution in the time k.

5. The apparatus of claim 3 , wherein the emotion learning unit normalizes weights of all Gaussian distributions corresponding to each emotion after the Gaussian distribution is changed.

6. The apparatus of claim 4 , wherein the emotion learning unit normalizes weights of all Gaussian distributions corresponding to each emotion after the Gaussian distribution is changed.

7. The apparatus of claim 2 , wherein if the matching function is 0 (if there is no matched Gaussian distribution), the emotion learning unit generates a Gaussian distribution that satisfies weight ω j,M j (k+1) (k+1), a central value μ j,M j (k+1) (k+1), and a variance σ j,M j (k+1) 2 (k+1) based on the feedback information as follows:

ω j,M j (k+1) ( k+ 1)=λ

μ j,M j (k+1) ( k+ 1)= x ( k+ 1); and

σ j,M j (k+1) 2 ( k+ 1)=σ 2 ,

wherein ω j,M j (k+1) (k+1) is the weight of a Gaussian distribution having the j th emotion distribution in the time k+1, λ is the predetermined basic weight, μ j,M j (k+1) (k+1) is the central value of the Gaussian distribution having the j th emotion distribution in the time k+1, x(k+1) is the feedback information, σ j,M j (k+1) 2 (k+1) is the variance of the Gaussian distribution having the j th emotion distribution in the time k+1, σ 2 is a predetermined basic variance, and M j (k+1) is a number of Gaussian distributions included in the j th emotion distribution in the time k+1.

8. The apparatus of claim 6 , wherein the emotion learning unit normalizes weights of all Gaussian distributions corresponding to each emotion after the Gaussian distribution is changed.

9. A method for a robot capable of learning emotion, the method comprising the steps of:

detecting, by an internal state calculation unit, external environment information, calculating an internal state vector value corresponding to one point of an internal state coordinate system based on the detected external environment information, and generating an internal state based on the calculated internal state vector value;

deciding, by an emotion decision processing unit equipped with an emotion model in which an emotion probability distribution for j types of emotions is formed on an n-dimensional internal state coordinate system, an emotion value of the robot by matching an input value of the internal state calculation unit with the internal state coordinate system;

deciding, by a behavior generation unit, a behavior of the robot based on the emotion value decided by the emotion decision processing unit and making an emotional expression based on the decided behavior; and

generating, by an emotion learning unit, feedback information on the emotion value corresponding to one point of the internal state coordinate system based on the emotion value decided by the emotion decision processing unit and information on a type of emotion received from a user input unit and changing an emotion probability distribution of the internal state coordinate system based on the feedback information,

wherein the emotion probability distribution comprises a first Gaussian distribution having weight, a central value, and a variance, or a second Gaussian distribution comprising a plurality of the first Gaussian distribution, and

wherein the emotion learning unit is configured to retrieve a Gaussian distribution corresponding to the feedback information using a matching function and is configured to change the Gaussian distribution of the emotion probability distribution based on the retrieved Gaussian distribution.