IEEE Robotics & Automation Magazine - September 2017 - 32

because the Kinetography Laban system operates in the
physical space while the main differences between two
movements might be in the motor-control space. The second experience accounts for a dance imitation that is also
reported in [6]. The dance score is translated in terms of a
robot program, i.e., the so-called SoT [5]. Even if the dance
movements are simple and not challenging for a humanoid
robot, significant differences appear between the original
human movements and the robot movements. Such differences are understood by comparing the Kinetography
Laban scores that describe the human motions and the
humanoid robot motions. The differences between the
scores provide a better understanding of what makes a
movement natural.
We first review research related to the segmentation of
complex movements, dancing robots, and computational
scoring. We then introduce the basics of the Kinetography
Laban system. This is followed by the first experience of picking up a ball. The motion
performed by the HRP-2
robot to pick up the ball is
The Kinetography Laban
notated with the Kinetography Laban system. The
system provides a way
objective is to point out
the flexibility and the limto segment and analyze
itations of this notation
system in expressing ancomplex movements of
thropomorphic movements
with different levels of
humanoid robots.
details. We then share the
experience reported in [6]
of translating the Kinetrography Laban score of a particular
dance, known as the Tutting Dance, into a hierarchical
sequence of tasks to be executed by the humanoid robot
Romeo. The Kinetography Laban system is used to compare
Romeo's movements with the dancer's. It appears that Romeo's
movements differ from the human ones. We will see how
these differences might refer to recent neuroscience and biological studies.
Segmenting Complex Movements
Several reported experiments promote the idea that motor
actions and movements in both vertebrates and invertebrates
are composed of elementary building blocks [7]; i.e., complex movement is segmented into simple movements, and
the combination of these elementary steps results in a complex action or movement, just the same as how the letters of
the alphabet make up more complex ideas in the form of
words. The motion segmentation of complex human movements is widely studied, and there is much research to be
found in the literature. The main objective is to determine
this alphabet of human movements. For instance, in [8], the
authors automatically constructed a directed graph called a
motion graph that encapsulated connections among the database from human motion capture data. Motion was generated simply by building walks on the graph.
32

IEEE ROBOTICS & AUTOMATION MAGAZINE

SEPTEMBER 2017

In [9], the role of a parameter that characterizes the twothirds power law was investigated. This parameter was nearly constant during extended parts of the movement and only
shifted abruptly at certain points of the trajectory. This was
interpreted as an indicator for segmented control. In [10],
the authors showed that imagined trajectories follow the
two-thirds power law. These findings support the conclusion
that the coupling between velocity and curvature originates
in centrally represented motion planning. However, for particular cyclic or repetitive actions, such as elliptical and figure-eight patterns of different sizes and orientations
performed by using the whole arm, there is no evident segmentation in the motor-control space but rather continuous
oscillatory patterns [11].
In the field of learning by demonstration, a general
approach for learning robotic motor skills from human demonstration was introduced [12]. By using a nonlinear differential equation to be learned to represent an observed
movement, the researchers built a library of movements by
labeling each recorded movement according to task and context (e.g., grasping, placing, and realizing). In [13], a hierarchical framework capable of learning complex sequential
tasks from human demonstrations was proposed. Through a
task-segmentation and action-primitive discovery algorithm,
both the high-level task decomposition and low-level motion
parameterizations were achieved for each action. Finally, in
[14], the authors proposed the use of nonnegative matrix
factorization to address the problem of segmenting combinations of initially unknown human motion primitives associated with ambiguous sets of linguistic labels during training.
This technique allowed the system to find the combinatorial
structure of parallel combinations of unknown primitives.
Dance Notations
Dance notation is to dance what musical notation is to music
and what the written word is to drama. It is basically a symbolic description of human movements and forms by using
graphic symbols and figures, numerical systems, path mapping, as well as letters and words. A recorded dance notation
that describes a dance through symbols is known as a dance
score. The most frequently used dance notation systems are
the Kinetography Laban system, originally created and published by Rudolf Laban in the late 1920s, the Benesh movement notation system, invented by Joan and Rudolf Benesh
in the late 1940s, and the Eshkol-Wachman movement notation system, created in Israel by dance theorist Noa Eshkol
and Avraham Wachman in the late 1950s. All of them allow
the notation of every kind of human movement [15],
although they differ in the way they represent the human
body and its movements.
The Benesh movement notation system is very similar to
the modern staff music notation. It is recorded on a five-line
stave from left to right with vertical bar lines to mark the transition of time. For this reason, Benesh notation is often synchronized with a musical staff. It draws the position of a
dancer as seen from behind, from the top of the head down to

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - September 2017