IEEE Robotics & Automation Magazine - December 2022 - 26

human complex skills has become an important method in
physical human-robot interaction [1]. Robots are expected
to have the ability to control both positions and forces to regulate
cooperative behaviors.
Achieving motion
Endowing robots with the
ability to learn human
complex skills has become
an important method in
physical human-robot
interaction.
synchronization is a key
issue in human-robot
cotransporting. Considering
a human-human
cotransporting scenario,
a leader and a follower
cotransport an object
and move it to a target
position, with the leader
aware of the target position
and moving the
object actively, and the
follower sharing the load
of the object and cooperating
with the leader passively, so there exists an inherent
delay between the leader and the follower. As shown
in Figure 1, the human knows the task flow and makes
decisions to perform the task, while the robot measures
Make Decision
Human
Motions
Sensor
Follow Human
Motion Synchronization
Delay
Motion
Prediction
Motion
Recognition
Delay
(a)
Human
(Leader)
1 Pick
Robot
(Follower)
2 Place
Robot
Motion
Planning
Command
Robot
Torque
Control
Robot
Robot
the human motion and plans its own motion to follow the
human's. As illustrated by the black arrows, there are
delays between the human's initiative motion and the
robot's responsive motion, which will decrease the collaboration
efficiency.
If the robot can predict human motion, it can plan its own
motion under the designed controller to decrease the delays,
and motion synchronization between humans and robots
can be achieved [see the red arrows in Figure 1(a)]. Many
human motion prediction methods have been developed in
recent years, such as the hidden Markov model, linear
regression (LR), the autoregressive integrated moving average
model, recurrent NNs (RNNs), and so on. In [2], NNs
are utilized to estimate human motion intent. In [3], human
intent is estimated without force sensors by observing robot
control efforts. In [4], an intentional reaching direction is
used to describe the human motion intention of the wearer's
upper limb, and a novel human-robot interface is designed
to estimate it in real time. Researchers in the field of computer
vision are also working on human motion prediction, and
[5] proposes a simple and scalable RNN architecture for predicting
human motion.
Other methods also can be used to accomplish the
human-robot cotransporting task, such as the vision method
(V method, based only on visual recognition of human
motion), the force method (F method, based on an impedance
control framework), and so on. The visual- force method
(VF method) combines visual and force sensing together
in such tasks, and the robot will adapt to human motion
actively, which can be applied to the scenario of human-robot
cotransporting. However, it is not synchronized between
human and robot motion, and there exist delays of robot
motion that generate a large interaction force in applications.
Yu et al. [6] propose a VF framework in the scenario of
human-robot cocarrying.
Recently, some novel and important control methods
Motion Delay
Starting Area
Storage Location
Motion Synchronization
(b)
Figure 1. A human-robot cotransporting task. (a) Delays in
a human-robot collaborative task. (b) An illustration of the
human-robot cotransporting task.
26 * IEEE ROBOTICS & AUTOMATION MAGAZINE * DECEMBER 2022
for robot control were proposed in [7], [8], [9], and [10].
From a control perspective, controller design in human-
robot cotransporting tasks is challenging because of uncertainties
in the coupled dynamics and kinematics. The
performance of human-robot collaboration with the current
controller design is still far from expected, especially
when compared to the performance of human-human
collaboration. Many types of research combine some
learning method [11] and focus on only force or position
control design to improve efficiency in such human-robot
collaboration tasks. Yang et al. 2018 [12] introduce a physical
haptic feedback mechanism to realize the human adaptive
impedance transfer and apply it to the physical
human-robot interaction. Dong et al. 2021 [13] study the
impedance control of a robot interacting with an unknown
environment, so that the interaction performance between
the robot and the environment is improved. In [14], a
novel hierarchical human-in-the-loop control is proposed
that includes impedance learning and human-in-the-loop
adaptive management.
IEEE Robotics & Automation Magazine - December 2022

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