AUTONOMOUS ROBOT DANCING DRIVEN BYBEATS AND EMOTIONS OF MUSIC
((in the name of the best creator))
Distributed Artificial IntelligenceCourse
INTRODUCTION
Many robot dances are preprogrammed by choreographers.
choreography:
2
OU
T O
F 1
5
primitive emotions
key frame (static poses)
Our work is made upof two parts:
(1) The first algorithm plans a sequence ofdance movements that is driven by the beats and the emo-tions detected through the preprocessing of selected dancemusic. (2) We also contribute a real-time synchronizing al-gorithm to minimize the error between the execution of themotions and the plan.
3
OU
T O
F 1
5
WE CREATE A LARGE LIBRARY OF MOTION PRIM-ITIVES, BY DIVIDING THE JOINTS OF THE NAO HUMANOID ROBOTINTO 4 CATEGORIES, WHERE EACH CATEGORY OF JOINTS CAN ACTUATEINDEPENDENTLY.
4
OU
T O
F 1
5
PUAL EKMAN PROPOSED 6 PRIMARY EMOTIONS:
1-happy 2-sad 3-surprised 4-angry 5-fear 6-disgust
5
OU
T O
F 1
5
JOINTS:
1. Head (Head): HeadYaw, HeadPitch
2. Left Arm (LArm): LShoulderPitch, LShoulderRoll, LElbowYaw, LElbowRoll
3. Right Arm (RArm): RShoulderPitch, RShoulderRoll, RElbowYaw, RElbowRoll
4. Legs (Legs): LHipYawPitch, LHipRoll, LHipPitch, LKneePitch, LAnklePitch, LAnkleRoll, RHipRoll, RHipPitch, RKneePitch, RAnklePitch, RAnkleRoll
6
OU
T O
F 1
5
emotion extraction:
emotion presentation:
(a , v)
a and v is between -1 , 1 7
OU
T O
F 1
5
SMERS(SVR)
94% agreement
super vector regression
beat tracking
beat tracking+ autocorrelation analysis+ neural network
amplitude ( the best candidate)
8
OU
T O
F 1
5
MAPPING MOTION PRIMITIVE TO ACTIVATION-VALENCE SPACE FROM STATIC POSTURES DATA
We collected 4 static postures of the NAO humanoid robot
for each of Ekman's 6 basic emotions:
Happy, Sad, Angry, Surprised, Fear and Disgust.
we have a totalof 24 emotional static postures.
6*4=24 9
OU
T O
F 1
5
10
OU
T O
F 1
5
EMOTION FOR NEXT MOTION PRIMITIVE
Motion primitives are selected sequentially and stretched
to all a whole number of beat times. To choose the next
motion primitive, we need the emotion at the end of the
previous motion primitive. We simply estimate the emotion
at each beat time by linearly interpolating the (a; v) values
11
OU
T O
F 1
5
THE MARKOV DANCER MODEL
12
OU
T O
F 1
5
exclusion we use an adaptive real-time synchronizing algorithm
primitive motions:
(i) be continuous
(ii) reflect the musical emotion
(iii) be interestingly non-deterministic
13
OU
T O
F 1
5
CONCLUSION
We show that we can automate robot dancing by formingschedules of motion primitives that are driven by the emo-tions and the beats of any music on a NAO humanoid robot.The algorithms are general and can be used on any robot.From emotion labels given for static postures, we can es-timate the activation-valence space locations of the motionprimitives and select the appropriate motion primitives foremotions detected in music.
14
OU
T O
F 1
5
THE END
15
OU
T O
F 1
5