IEEE Robotics & Automation Magazine - December 2013 - 132

Apart from the neuroscience literature, the research communities of automatic control, optimization, and biomechanics are very active in the field as well. Many optimality criteria have been introduced to explain human motions. The
pioneer work [10] presented the first result in modeling natural trajectories for the human arm as resulting from jerk
minimization. Further research led to more detailed objective functions. For instance, in [11], the redundancy of the
human arm during a reaching trajectory is resolved minimizing the magnitude of the total work done by joint
torques. In [11], a relationship between the swivel angle
(used to parameterize the redundancy) and the hand pose is
investigated and used to synthesize a humanlike behavior of
a humanoid robot. The same relationship is exploited in [12].
Recently, a biomechanics-based investigation was performed by Khatib, et al. [5]. In [13], the muscular redundancy
is taken into account explicitly. It was argued that in a static
configuration, among all the possible muscular activations,
the optimal posture can be predicted using a weighted leastsquare approach. Hidden Markov Model (HMM) or
Gaussian Mixture Regression (GMR) have been recently proposed to learn natural movements [14].
When it comes to industrial robotics, attempts to duplicate
the natural motion of the human arm with the motion of the
robotic arm must comply with the following constraints:
● End-effector motion should be accurately controlled.
● The algorithm adopted for redundancy resolution must be
compatible with the architecture of a real industrial robot
controller.
This article contributes deriving a simple relationship
between the hand pose (position and orientation) and the
swivel angle in the motion of the human arm. This relationship effectively explains the motion of the arm and lends
itself to easy application in the kinematic control of a
robotic arm. To identify such a relation, an extensive experimental study on volunteers performing simple assemblylike tasks has been conducted. The existence of the relationship is first proven with statistical methods using a

Gleno-Humeral
Joint Center (O)

Elbow (E)
Wrist
Center (W) Backward
q6
q5

Left
Rotate

Right

Forward

q7
(Self-Motion)
Figure 1. A hand's admissible motion on the spherical surface.

132

IEEE ROBOTICS & AUTOMATION MAGAZINE

DECEMBER 2013

clustering approach and multivariate correlation statistics
and then identified using a least-squares algorithm. Once
the relationship between hand poses and swivel angle is
identified, it is implemented in an industrial robot controller to resolve the kinematic redundancy of the ABB 14-DoF
dual-arm concept robot Frida. Remarkably, no modification of the controller architecture is necessary to implement
the method. Experiments performed on Frida and
reported in this article demonstrate the method. This article is an extension of [15], discussing the identification
experiments in detail and, above all, adding the implementation of the humanlike motion on the Frida robot with
the related demonstration.
The remainder of this article is organized as follows. In
the "Motion-Capture Experiments" section, the protocol of
the experiments on volunteers and the procedure to elaborate the data are described. The kinematic model of the
human upper limb is also introduced. In the section
"Redundancy Resolution in the Human Arm," a black-box
model explaining how humans exploit arm redundancy is
presented and validated against experimental data. Finally,
the section "Humanlike Trajectories for a Dual-Arm Robotic
Industrial Manipulator" discusses the applicability to the
Frida prototype and presents experimental results.
Motion-Capture Experiments
This section describes the experimental campaign organized
to get all the data needed to study human arm motion. The
concept exploited in the experiments is the motion capture,
whereby some volunteers have been instructed on the operations they were supposed to perform, and the motion of their
arm was tracked through visual recording of markers
attached to the arm itself. The goal of the experiments was to
investigate whether a relationship between the hand pose and
the swivel angle exists and how it can be characterized. The
experimental setup as well as the protocol of the experiments
is described in this section.
Description of the Experiments
During the experiments, the hand was constrained to move
with the palm on a rigid smooth sphere of a 7-cm radius
placed on a work bench. The height of the work bench has
been raised to 109 cm, as prescribed by the standards on
anthropometric requirements for the design of workstations
at machinery [16]. The motion of the human arm, which in
this way resembles basic assembly operations, is partially constrained, but the motion of the hand along the spherical surface can be performed freely as depicted in Figure 1. Thanks
to motion-capture techniques, the motion of the volunteer's
arm was tracked through visual recording of markers
attached to the arm itself.
A commercial optoelectronic motion-capture system,
Vicon (Oxford, United Kingdom), consisting of six T10 cameras (1-MP @200 Hz) with near infrared strobes was used.
Eight healthy volunteers were selected between the 50th and
95th percentile. Anthropometric measurement of each

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2013