Cobalt-catalyzed regioselective syntheses of indeno[2,1-: C] pyridines from nitriles and diynes bearing propargyl fragments

Article abstract of DOI:10.1039/c8ob02419g

A highly efficient CoI₂/o-phenanthroline catalyzed cycloaddition reaction of diynes bearing TBS protected propargylic alcohol fragments with nitriles has been developed. This methodology offers regioselective access, with good functional group tolerance, to various indeno[2,1-c]pyridine derivatives in moderate to excellent yields. It was found that o-phenanthroline as a bidentate nitrogen ligand showed high efficacy in this cycloaddition reaction.

Full text of DOI:10.1039/c8ob02419g

2018 3rd International Conference on Advanced Robotics and
Mechatronics (ICARM)
Human-robot Interaction Oriented Human-in-the-loop Real-time
Motion Imitation on a Humanoid Tri-Co Robot
Wenbin Xu1, Xudong Li1, Wenda Xu1, Liang Gong1, Yixiang Huang1, Zelin Zhao1, Lujie Zhao1, Binhao Chen1,
Haozhe Yang1, Li Cao2, Chengliang Liu1
Abstract—Inspired by the concept of Tri-Co Robots, a real- are to be substituted or to be strengthened, robots that are
time motion imitation based human-robot interaction method is
proposed to telemanipulate a life-size humanoid robot. First, a
motion capture system is established to retarget the accurate
motion data to a skeletal model. Second, the real-time data
transfer for a human-in-the-loop control system is realized
equipped with human morphology are required. With respect
to HRI, humanoid robots have numerous advantages. First,
since humanoid robots possess human-like structures and
scales that have evolved for millions of years, they are com-
through the publish-subscribe messaging mechanism in ROS pletely capable of integrating themselves into daily life, human
(Robot Operating System). Third, a fast mapping algorithm
is developed to convert BVH (BioVision Hierarchy) data into
corresponding joint angles which are then encapsulated in a
designed communication protocol and sent to a low-level slave
controller through a serial port. Finally, visualization terminals
society and uncertain environments seamlessly to emulate
human behaviors and acquire necessary knowledege and skills.
Second, humanoid robots can provide a direct and natural
platform for HRI. Because they are entirely able to mirror
render it more convenient to make comparisons between two human motion, manipulators can obtain overall comprehensive
different but simultaneous motion systems. Experimentation
on complicated gesture imitation shows the feasibility of the
proposed methodology, and the synchronous latency of less than
0.5 seconds validated its real-time performance. The proposed
human-in-the-loop imitation control method enhances robot in-
cognition of surroundings, analyze complicated information
and make strategies for humanoids from a first-person view
so that robots can be gradually endowed with the ability to
repeat similar processes or even to make autonomous decisions
teraction capability in a more natural way and improves robot through such a natural teaching. Third, humanoid robots can
adaptability to uncertain and dynamic environments.
Index Terms—Motion Imitation, Life-size Humanoid Robot,
BioVision Hierarchy, Motion Capture, Human-in-the-loop Con-
trol, Human-robot Interaction
resort to rich fruits of ontology and epistemology of human
civilization to accumulate necessary knowledge, which renders
the cost of HRI lowest and achieves the most satisfactory
performance.
To facilitate the interaction between human and humanoid
robots under social and industrial circumstances, novel be-
haviors that are commonly used in human life but not pre-
programmed to humanoid robots should be taught to them
naturally, which resembles the learning process of children.
For inspiration, we look to successful strategies adopted by
people, such as imitation, where people observe the behavior
of another, and reproduce it. [3] After human demonstrations,
imitation of this behavior is usually more convenient for
humanoid robots than manually programmed controllers. [4]
Hence in this regard, human imitation is of increasing im-
portance: human-like behavior, emotionally and aesthetically
pleasing action, and expressive motion are fundamental. [5]
There are several related works over the past few years.
Marcia Riley proposed a framework where external cameras
and head-mounted cameras are used to capture body postures
and joint angles are computed through a fast full-body inverse
kinematics (IK) method. [3] The full-body IK problem is
divided into many sub-problems to realize real-time imitation
on a Sarcos humanoid robot with 30 degrees of freedom
(DOF). Besides, several articles ( [1], [6], [7]) utilize Kinect
for gesture recognition and then perform similar actions on
robots through different algorithms. Additional to vision, other
sensory methods are also adopted. Abhay Bindal fixes IR
sensors and accelerometer motion sensors to human legs and
achieve real-time control of gaits through an Arduino ATmega
2560 based microcontroller on a biped humanoid robot. [8]
I. INTRODUCTION
With the rapid development of technology, robots have
experienced great changes in many aspects since the invention
of the first industrial robot in 1959. Faced with different
missions, various robots have been developed, ranging from
sweeping robots to educational robots. To improve the coop-
eration between human and robots, human-robot interaction
(HRI) has risen as a research area, attracting attention from
academics and industry. [1] In 2017, the National Natural
Science Foundation of China (NSFC) has launched the Tri-Co
Robot (i.e. the Coexisting-Cooperative-Cognitive Robot) ma-
jor research program to enhance robot interaction capacities,
one of which is HRI capacity. [2] These three features ensure
the ubiquitous applications of robots, effective coordination
with other agents and adaptability to uncertain and dynamic
environments respectively.
Human society as well as most scenes in the world are
constructed according to our scales, demands and capacities,
which means if parts of, or even complete human functions
1School of Mechanical Engineering, Shanghai Jiao Tong University, Shang-
hai, China
2Rui Jin Hospital, Shanghai Jiao Tong University School of Medicine,
Shanghai, China
Corresponding author: gongliang mi@sjtu.edu.cn
This study was supported by the Natural Scientific Foundation of China
under Grant NO.51775333 and the Foundation of Cross-disciplinary Research
on Medicine and Engineering under Grant NO. YG2016MS64.
978-1-5386-7066-8/18/$31.00 ©2018 IEEE
781

2018 3rd International Conference on Advanced Robotics and
Mechatronics (ICARM)
Akif DURDU attaches potentiometers to human joints and
the collected data are classified with the assistance of ANN
to perform movements on the robot. [4] However, there are
some demerits in existing works. First, conventional vision-
based motion recognition methods are not completely reliable
due to their sensibility to lighting conditions and cluttered
backgrounds. Numerous cameras can indeed increase the reli-
ability but also the expenses. Through wearable sensors, more
accurate motion can be captured. Second, the real-time feature
will be negatively affected if the full-body IK problem is
required to be solved. Hence an algorithm with less calculation
amount is desired. Third, motion imitation has rarely been
applied to humanoid robots with human dimensions due to
the complicated structure and the great difficulty of control.
In this paper, we present a human-in-the-loop system where
a life-size humanoid robot established based on InMoov, an
open-sourced 3D printing humanoid robot, can imitate real-
time motion of human’s upper limber. The humanoid robot
has a similar structure with human and is equipped with 29
DOF, 22 of which are controlled in this system. The process
of motion imitation is illustrated as follows. First, a set of
wearable sensors is adopted to capture the motion of human
and the collected motion data are presented in BVH (BioVision
Hierarchy) format. Then these data are transmitted to an
industrial PC with Ubuntu running on it through TCP/IP and
parsed according to BVH format. Next, the parsed euler angles
are converted to corresponding joint angles with mathematical
techniques and encapsulated in a communication protocol. At
last, the industrial PC sends joint angles to the low-level slave
controller to control the robot. This human-in-the-loop system
has paved a novel, real-time and accurate way for human
motion imitation on humanoid robots.
Fig. 1. An example of BVH format
on Windows OS for calibration and management. Besides, a
skeleton model is visualized in ANP to reflect real-time human
motion. Another important feature of ANP is to broadcast
BVH data through TCP so that other programs can obtain
and analyze these data through specified IP and port using
SDK provided by Noitom Technology Ltd.
B. Human Motion Retargeting
Motion retargeting is a classic problem which aims to
retarget motion from one character to another while keeping
styles of the original motion. [9] With this method, real-
time human motion can be reflected on the skeletal model in
ANP through BVH data, developed by the BVH company for
motion description. It is a universal human feature animation
file format and usually adopted in skeletal animation models. It
can store motion for a hierarchical skeleton, which means that
motion of the child node is directly dependent on the motion
of the parent one. [10] A normal BVH file will consist of
several parts as follows.
This paper is organized as follows. In section 2, the motion
capture system is introduced. Section 3 discusses the setup
of the humanoid robot. Section 4 presents the realization of
real-time motion imitation on the humanoid robot. Section 5
performs several experiments on complicated gesture imitation
through our proposed method. Finally, section 6 deals with the
conclusion about our work.
II. MOTION CAPTURE SYSTEM
• HIERARCHY signifies the beginning of skeleton defini-
tion.
• ROOT gives the first defined joint which is also the root
This section gives an introduction about the motion capture
system. This system is composed of a motion sensor to capture
real-time human motion and a human motion retargeting
method which enables us to map these data to a simplified
skeletal model.
of the whole skeleton.
• OFFSET specifies the deviation of the child joint from the
parent joint, which remains constant due to the unchanged
lengths of human limbers.
A. Motion Sensor
• CHANNELS contains several parameters. The first pa-
rameter, which is 6 in Fig. 1, indicates the number of
DOF. Usually only the root joint has both position data
and rotation data. The rest only contains rotation data
because positions of other joints are obtained from the
fixed values of OFFSET. Besides, these rotation data
are Euler angles and the sequence of rotation hinges on
the sequence mentioned in CHANNELS. In Fig. 1 , the
rotation is carried out in YXZ order.
A modular system composed of 32 9-axis sensors is adopted
as the motion sensor. It is a set of wearable sensors designed by
Noitom Technology Ltd to deliver motion capture technology.
Characteristics of the whole modular system include small,
adaptive, versatile and affordable. It contains 32 IMU (Inertial
Measurement Unit), each of which is composed of a 3-axis
gyroscope, 3-axis accelerometer and 3-axis magnetometer.
The system is used with Axis Neuron Pro(ANP) running
978-1-5386-7066-8/18/$31.00 ©2018 IEEE
782

2018 3rd International Conference on Advanced Robotics and
Mechatronics (ICARM)
• End Site is only tagged in the definition of an end-effector
and describes the lengths of the bone through OFFSET.
• MOTION represents the beginning of another section
which describes states of each joint at each moment.
• Frames stands for the current order of frames and Frame
Time is the duration of each frame. The rest data are
real-time states of each joint described sequentially in
the HIERARCHY section. The number of these data is
equal to the total number of channels defined in the
HIERARCHY section.
We adopt BVH with no position channels and hence three
rotation data are obtained for each joint since position data
will keep constant. Then we can figure out human gestures
through these three rotation angles based on the assumption
that wearable sensors are fixed with respect to human body.
Fig. 2. DOF of the humanoid robot (DOF of fingers are not displayed)
III. SETUP OF HUMANOID ROBOT
To realize real-time motion imitation on robots, a humanoid
robot is set up since they possess human-like design and
are able to mimic human motion. [11] However, due to
complicated structures and various constraints of conventional
manufacturing methods, it is difficult to fulfill an elegant
design of a dexterous humanoid robot. Fortunately, with the
rapid advancement in 3D printing technology, 3D printing
turns to be more cost-effective, and the printing element is also
becoming more accurate, complex and stronger. 3D-printed
humanoid robots like InMoov, Flobi and iCub have been
created to serve as experiment platforms where research on
HRI is conducted.
In this paper, a 3D-printed life-size humanoid robot is
established based on InMoov initiated by Gael Langevin, a
French sculptor in 2012. [12] The whole structure as well
as other necessary backgrounds have been illustrated in the
previous work. [13] 22 out of 29 DOF are controlled during
motion imitation, including 5 DOF for each hand, 4 for each
arm, 3 for each shoulder and 2 for the neck, as shown in
Fig. 2. With regard to control, the low-level slave controller
is composed of 4 small Arduino Nano control core boards,
each of which can drive 6 servos with corresponding angles
through PWM wave, and an Arduino Mega 2560 master board
which communicates with the aforementioned nano nodes via
485 Hub based on the Modbus RTU control.
Fig. 3. Whole structure of the proposed method
A. Data Transmission
Nodes, which are executables after compilation, can sub-
scribe to or publish new messages to a topic in ROS (Robot
Operating System). [14] This publish-subscribe messaging
mechanism is one of the most important features of ROS.
Data streams are enabled mainly through topics in this paper.
The whole data stream is visualized in Fig. 4, where ellipses
stand for nodes and squares represent topics.
• rosserial server socket node connects with the win32
console through TCP/IP and then advertises the topic,
perception neuron/data 1
• perception neuron one topic talk node subscribes to
the previous topic and then converts euler angles in BVH
data to joint angles, which are then published to another
topic called Controller joint states.
• joint state publisher subscribes to the previous topic and
realizes the real-time simulation of robot model.
• perception serial will send joint angles to the low-level
slave controller through a serial port after obtaining them
from Contoller joint states
IV. REAL-TIME IMITATION OF HUMAN MOTION
Several functional components are developed to realize
real-time human motion imitation. The whole structure of
the proposed method is shown in Fig. 3. First, the publish-
subscribe messaging mechanism and the designed communi-
cation protocol ensures the safety of data transfer. Second, a
key point to the accomplishment of motion imitation is the fast
mapping algorithm which converts euler angles in BVH data
into corresponding joint angles. Third, visualization terminals
make it possible to make comparisons between different but
simultaneous motion systems.
978-1-5386-7066-8/18/$31.00 ©2018 IEEE
783

2018 3rd International Conference on Advanced Robotics and
Mechatronics (ICARM)
Fig. 4. Visualized data stream through ROS publish subscribe messaging
Fig. 5. Designed Communication Protocol
Fig. 7. Elbow conversion from 2 DOF to 1 DOF
no universal algorithm exists for each joint and we need to
specify the algorithm for certain cases. The first case is the
conversion from 3 human DOF to 3 robot DOF. Take shoulder
as an example. There are 3 DOF on each shoulder part of
InMoov and their axes of rotation can be approximately
treated as perpendicular to each other. Assume the joint
angles of 3 shoulder joints are relatively α, β, γ and the
rotation matrix of the arm link with respect to the shoulder
link can be similarly expressed as
Fig. 6. Three motion systems with different constraints
While topics are mainly adopted in the data transfer process
on Ubuntu, to control the robot, we still need to transmit joint
angles to the low-level slave controller through a serial port. In
case packet loss or data dislocation takes place during trans-
mission, a specific communication protocol is designed, as
shown in Fig. 5. Before transmission, all these data including
time stamp and joint angles are converted to integers. The
communication protocol contains 2 bits time stamp data, 22
bits position data corresponding to each joint, and two bits of
CRC16 check code which are generated according to prior 27
bits to ensure the safety of data transfer.
⎛
⎞ ⎛
⎞ ⎛
⎞
cosα −sinα
0
0
1
cosβ
0
−sinβ
0
1
0
sinβ
0
cosβ
1
0
0
0
0
shoulder
arm
⎝
⎠ ⎝
⎠ ⎝
⎠
R
=
sinα cosα
0
cosγ −sinγ
(2)
0
sinγ cosγ
Then we only need to equal ϕ, θ, ψ obtained from the
motion sensor and α, β, γ used to control the robot but one
thing needs to be noticed. Since the mechanical structure of the
robot has determined the rotation order of three joints between
shoulder and arm, the rotation order of euler angles in BVH
should be set to be the same (in our case, ZYX), or this method
will be invalid.
B. Mapping Algorithm
To make the robot imitate human motion, the key is to send
corresponding joint angles computed from BVH data. BVH
has provided us with three euler angles for each node, which
enables us to ascertain the rotation matrix between child and
parent links. Assume euler angles with a rotation order of ZYX
can be expressed as ϕ, θ, ψ, the rotation matrix of child frame
with respect to parent frame is
The second one is conversion from 2 human DOF to 1 robot
DOF. The example is the elbow. Human elbows are able to
bend and rotate while those of the robot can only bend. Then
we need to compute the joint angle for bending, which is Ω,
as shown in Fig. 7. With the assumptions that sensors are fixed
with respect to human body and the x-direction is along the
links, we can derive the following equations with the rotation
⎛
⎞ ⎛
⎞ ⎛
⎞
cosϕ −sinϕ
0
0
1
cosθ
0
−sinθ
0
1
0
sinθ
0
cosθ
1
0
0
0
0
parent
child
⎝
⎠ ⎝
⎠ ⎝
⎠
R
=
sinϕ cosϕ
0
cosψ −sinψ
(1)
0
sinψ cosψ
matrix (1). R21 stands for the rotation matrix of frame x2y2z2
1
Fig. 7 shows the mapping problem. Human, with biological
constraints, cannot have 3 rotational DOFs at each joint
and some of them are not completely independent. With
mechanical constraints, many joints of humanoid robots are
also unable to rotate in three independent directions. The
main difficulty is to eliminate these differences. Therefore,
with respect to x1y1z1. xˆ is the description of unit vector
1
of x1 in frame x1y1z1.
2
xˆ = (1, 0, 0)T
(3)
(4)
2
1
2
xˆ = R1xˆ = (cosθcosψ, cosθsinψ, −sinθ)T
2
2
2
978-1-5386-7066-8/18/$31.00 ©2018 IEEE
784

2018 3rd International Conference on Advanced Robotics and
Mechatronics (ICARM)
Fig. 8. Different visualization terminals for different motion systems
Fig. 9. Experiments of different gestures with arms and head
1
1
< xˆ , xˆ >= arccos(cosθcosψ)
(5)
2
1
1
1
Ω = π− < xˆ , xˆ >= π − arccos(cosθcosψ) (6)
2
1
The last case is conversion from 3 human DOF to 2 robot
DOF like the neck joint. The solution to this case resembles
that for the shoulder joint and we only need to take 2 of 3
euler angles in the corresponding order.
C. Different Visualization Terminals
On one hand, to display human motion, ANP can visualize
a skeletal model where each joint has three rotational DOF
regardless of human’s physiological structures as well as the
humanoid robot’s DOF, as shown in Fig. 8. The skeletal model
mainly utilizes the euler angles to reflect the actual human
gestures and the length of each link is fixed for one chosen
body size.
Fig. 10. Comparison between fingers
V. EXPERIMENT
This section presents the experimental results using the
proposed method based on the humanoid robot. The results
can be seen from Fig. 9 -Fig.10. To verify the feasibility
of the system, photos for various poses were token from the
human motion imitation system, including different positions
of two arms, face orientations and movements of fingers. These
gestures are complicated because imitation of these gestures
entails the rotation of most revolute joints at the same time
rather than one or two. Also, the high degree of similarity
between the wearer and the humanoid robot has demonstrated
that the robot has successfully followed the upper limber’s
motion of the wearer., thus proving the feasibility of our
proposed method. Besides, the synchronous latency of less
than 0.5 seconds validates the real-time performance.
To illustrate the accuracy, we take one DOF of the right
shoulder as an example. Fig. 11 shows the comparison between
different trajectories of the same motion on human and the
humanoid robot. The trajectory can be approximated as an arc
and the central angle of the arc represents the accuracy of
the proposed method. The error is nearly 3◦, which is 6.1%
based on the rotation angle of the human arm. The relatively
small error has further demonstrated the high accuracy of our
method.
On the other hand, to visualize the robot model as well
as to make simulation more convenient, another visualization
scheme is provided with the assistance of ROS. A 3D
visualization model is created in URDF (unified robot
description format). URDF is a language based on XML and
designed to describe the robot simulation model universally in
ROS system, including the shape, size and colour, kinematic
and dynamic characteristics of the model. The basic grammars
are mentioned in [14]. However, since Inmoov has a very
complicated structure, it would be arduous to write a URDF
manually. Hence we resort to a powerful tool called Xacro
(XML Macros). Xacro is also an XML macro language
adopted to reuse one structure for two different parts, i.e.
left arms and right arms and to auto-generate a URDF
file. Some fundamental grammars are shown in Table I.
Also, open-sourced *.STL files can be imported in URDF
after scale adjustment. After finishing a URDF file, we
need to invoke RVIZ, a 3D visualization tool in ROS and
a node called joint state publisher subscribing to a topic
which publishes joint angles in sensor msgs/JointState
format. Then the robot model will be visualized in RVIZ
and behave with the computed joint angles, as shown in Fig. 8.
However, there are still some limitations due to several
978-1-5386-7066-8/18/$31.00 ©2018 IEEE
785

2018 3rd International Conference on Advanced Robotics and
Mechatronics (ICARM)
TABLE I
FUNDAMENTAL GRAMMARS OF XACRO
Command Definition
Property <xacro:property name=“pi” value=“3.14” />
Argument <xacro:arg name=“use gui” default=“false”/>
Macro <xacro:macro name=“arm” params=“side”/>
Including <xacro:include filename=“other file.xacro” />
Usage
< · · · value =“ ${2*pi}”· · · />
< · · · use gui:= true · · · />
<xacro:arm side=“left”/>
demonstrated that the system is feasible, real-time and accu-
rate. Future work will lay more emphasis on the development
of mapping algorithms and the accuracy of human motion
imitation on a humanoid robot. Encouraged by Tri-Co Robot
Initiative, we hope this work will contribute to the enhance-
ment of robot interaction capacities.
REFERENCES
[1] E. Yavan and A. Uar, “Gesture imitation and recognition using kinect
sensor and extreme learning machines,” Measurement, vol. 94, pp. 852–
861, 2016.
[2] H. Ding, X. Yang, N. Zheng, M. Li, Y. Lai, and H. Wu, “Tri-co robot:
a chinese robotic research initiative for enhanced robot interaction
capabilities,” National Science Review, p. nwx148, 2017. [Online].
Available: http://dx.doi.org/10.1093/nsr/nwx148
[3] M. Riley, A. Ude, K. Wade, and C. G. Atkeson, “Enabling real-time
full-body imitation: a natural way of transferring human movement
to humanoids,” in IEEE International Conference on Robotics and
Automation, 2003. Proceedings. ICRA, 2003, pp. 2368–2374 vol.2.
[4] A. Durdu, H. Cetin, and H. Komur, “Robot imitation of human arm via
artificial neural network,” in International Conference on Mechatronics
- Mechatronika, 2015, pp. 370–374.
[5] S. H. Hyon, J. G. Hale, and G. Cheng, “Full-body compliant human-
humanoid interaction: Balancing in the presence of unknown external
forces,” IEEE Transactions on Robotics, vol. 23, no. 5, pp. 884–898,
2007.
[6] I. J. Ding, C. W. Chang, and C. J. He, “A kinect-based gesture command
control method for human action imitations of humanoid robots,” in
International Conference on Fuzzy Theory and ITS Applications, 2014,
pp. 208–211.
[7] A. Bindal, A. Kumar, H. Sharma, and W. K. Kumar, “Design and
implementation of a shadow bot for mimicking the basic motion of
a human leg,” in International Conference on Recent Developments in
Control, Automation and Power Engineering, 2015, pp. 361–366.
[8] S. Gobee, M. Muller, V. Durairajah, and R. Kassoo, “Humanoid robot
upper limb control using microsoft kinect,” in 2017 International Con-
ference on Robotics, Automation and Sciences (ICORAS), Nov 2017, pp.
1–5.
[9] X. Meng, J. Pan, and H. Qin, “Motion capture and retargeting of fish by
monocular camera,” in International Conference on Cyberworlds, 2017,
pp. 80–87.
[10] H. Dai, B. Cai, J. Song, and D. Zhang, “Skeletal animation based on
bvh motion data,” pp. 1–4, 2010.
Fig. 11. Snapshots for motion trajectory
factors. The first one is the difference of structure between
human and robot. Each of our arm has 7 DOF but the robot
has only 5 and the rotational axes of wrists are not the same.
Also, the rotational angles of some robot joints are limited
due to its mechanical design. (i.e. The arm of the robot
cannot get over its shoulder.) The second one is the mismatch
between the skeletal model visualized through BVH data and
the wearer’s real motion. As we all know, the revolution
of each joint is achieved through skeletons in human body,
while the wearable sensors can only keep fixed to the skin
or the clothes and there exists relative angular displacement
between our skins and skeletons. Hence errors are obvious
when we carry out particular behaviors. Other factors include
the accumulated drift errors and different positions of wearable
sensors relative to human bodies. Nevertheless, there are still
some possible solutions to these limitations. For example,
sensors can be bound tightly to limbs in case of relative
displacement between sensors and skins. Human motion can
be confined to a certain range to achieve a higher accuracy.
Besides, reasonable compensations for errors resulting from
relative angular displacements between skins and skeletons can
render the motion retargeting more reliable.
[11] N. E. N. Rodriguez, G. Carbone, and M. Ceccarelli, “Antropomorphic
design and operation of a new low-cost humanoid robot,” in Ieee/ras-
Embs International Conference on Biomedical Robotics and Biomecha-
tronics, 2006, pp. 933–938.
[12] G. Langevin, “Inmoov,” http://www. inmoov.fr/project, 2014.
[13] L. Gong, C. Gong, Z. Ma, L. Zhao, Z. Wang, X. Li, X. Jing, H. Yang,
and C. Liu, “Real-time human-in-the-loop remote control for a life-
size traffic police robot with multiple augmented reality aided display
terminals,” in 2017 2nd International Conference on Advanced Robotics
and Mechatronics (ICARM), Aug 2017, pp. 420–425.
VI. CONCLUSIONS
In this paper, a novel human-in-the-loop system for human
motion imitation on a humanoid robot has been proposed.
The system enables real-time imitation through an accurate
motion capture system, visualization terminals for different
motion systems, fast mapping algorithms and reliable data
transfer methods. Experiments with different gestures have
[14] Z. Wang, L. Gong, Q. Chen, Y. Li, C. Liu, and Y. Huang, Rapid
Developing the Simulation and Control Systems for a Multifunctional
Autonomous Agricultural Robot with ROS.
Publishing, 2016.
Springer International
978-1-5386-7066-8/18/$31.00 ©2018 IEEE
786