IEEE Robotics & Automation Magazine - December 2022 - 105

combined with image-based visual servoing (IBVS) to
dispose of the uncertainties of a seven-degree-of-freedom
(7-DoF) redundant robot manipulator. The proposed
algorithm is substantiated through comparison
experiments based on the Sawyer robot and constructed
visual servoing physical platform.
Introduction
In recent decades, human-like behavior has increased
research interest in various areas, such as industrial robots,
service robots, and wearable robotic systems. Industrial application
is converting from mechanization to anthropomorphism,
and in the case of human-robot collaborative work, it
can imply a threat to the physical safety of workers. Human-
robot interaction (HRI) research is a feasible strategy to fix
similar problems [1], [2]. Robots with human-like behavior
will promote HRI task performance, particularly when industrial
robotic manipulators cooperate with workers to accomplish
assembly tasks. The human-like motion of an industrial
manipulator helps elevate the safety of human-robot cooperation
while avoiding discomfort or tension from nearby
humans. The upper limb of the human can be modeled as
kinematic chains, and the kinematic systems often have
human-redundant degrees of freedom [3]. Typical anthropomorphic
serial redundant robot manipulators, such as
MOTOMAN-SIA (Yaskawa, Fukuoka, Japan) and LBRiiwa
(KUKA, Augsburg, Germany), can facilitate HRI task performance
because of their human-like behavior and similar
mechanical structure to the human arm, which will make robot
manipulators more approachable and socially accepted [4].
Human-like motions have been implemented on the
robot's end effector to improve the task performance of HRI
[5]. A collaborative human-robot manufacturing cell for
assembly is equipped with the end effector of a redundant
robot, which can help alleviate the burden on the operator [6].
In [7], remote center of motion constraints are employed in
combination with an improved recurrent neural network
(NN) for trajectory control of redundant robotic manipulators.
However, focusing on only the human-like behavior of
the end effector is not enough as the arm pose will enhance
the humanoid's control performance.
Employing robot kinematic redundancy can actualize the
human-like motion control of redundant robotic manipulators.
An elbow elevation angle motion of a human limb is
defined to characterize human arm motions [8]. In [9], to
resolve redundancy of the wearable robotic system, which is
similar to human-arm redundancy, they define swivel angle
W , as shown in Figure 1. The human-like behavior of the
redundant robot manipulator can be introduced by the kinematic
model that predicts the relationship between the
manipulator and the human arm motion. In [10], incremental
learning is employed to optimize redundancy of an
anthropomorphic robot manipulator, and human-like behavior
is realized by utilizing swivel-motion reconstruction. Deep
learning can be considered to achieve humanoid control of
the robot manipulator. In [11], deep learning is utilized to
develop a human activity recognition system that could perceive
human dynamic motions.
To make industrial robots behave more like humans,
industrial robots are equipped with visual perception functions,
which allow robots to complete more complex and flexible
work. Visual servoing is widely used in industrial fields to
complete tasks such as grasping and assembly. IBVS with an
eye-in-hand construction is utilized in this article to reduce
reliance on accurate camera models and avoid the computational
burden of pose estimation. IBVS improves the efficiency
and local stability of vision-guided tasks. As IBVS relies on
only image feature points as a control input, the loss of image
features will directly lead to the failure of vision-guidance
tasks. Therefore, constraining image feature points within the
camera's FoV is vital to the success of visual servoing tasks.
Numerous research works have been suggested for the
problem of image feature loss in visual servoing tasks [12]-
[19]. Some research works have attempted to use path planning
to avoid image feature trajectories that exceed the
camera's FoV. In [12], image moment trajectories are designed
to solve the image-based control task of the quadrotor by
defining them in the virtual image plane. A trajectory-planning
algorithm is developed to overcome the shortcomings of
visual servoing by parameterizing the velocity screw of the
camera and providing an accurate initial depth from the proposed
depth-estimation technique [14]. Hybrid visual servoing
schemes are also employed. In [17], a control framework
is designed that combines the advantages of position based
and IBVS to control a robotic arm-equipped aerial vehicle
without losing features during operations. Mekonnen et al.
[19] present a new wireless hybrid control algorithm for the
visual servoing of mobile robots. The hybrid algorithm utilizes
the position-based visual servoing method for global routing
and the IBVS method for fine navigation. The mentioned
strategies can guarantee FoV constraints in the process of
visual servoing tasks, but there are many online nonlinear
optimization problems, and their real-time feasibility in
robotic systems is suspectable.
From the perspective of constraining the image feature
output error, the image feature output error does not violate
the safety boundary, so it indirectly constrains the image feature
points to remain within the preset camera's FoV. The BLF
has been widely used in recent years to constrain the
Shoulder (S)
()
Wrist (W)
ist (W)
Ψ
Elbow (E)
Elb
(E)
Figure 1. Definition of the swivel angle of the human arm.
DECEMBER 2022 * IEEE ROBOTICS & AUTOMATION MAGAZINE *
105
IEEE Robotics & Automation Magazine - December 2022

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2022
