IEEE Robotics & Automation Magazine - December 2022 - 32

the hand of the Baxter robot. A Robotiq FT sensor is
mounted on the robot wrist for measuring the external
force, which is converted to rx for obtaining the external
torque on joint " E1. " Joint " E1 " of the robot's arm from full
flexion to extension is
0-1.8 rad, and the human
elbow joint value hi
from flexion to extension
is 0.7-2.4 rad. A linear
transformation is
used to normalize the
value range of the human
elbow joint angle into
the angle range of joint
" E1 " of the robot's arm.
We also transform the
human joint angle from
the camera coordinates
to robot reference coordinates
and transform
force sensory informagenerates
control input to control the robot by the Baxter
Robot Operating System software development kit
in Ubuntu 14.04 LTS.
The application of our
proposed control can
reduce the delay time,
achieve human-robot
synchronization, and
reduce the interaction force
between human and robot.
tion from force sensor coordinates to robot reference coordinates,
which can be found in our previous work [6].
Two computers are employed in the experiment. One
computer is used to calculate the feedforward input x ff
of NN compensation in (7) by MATLAB Simulink. This
computer is also employed to obtain visual and force
sensory information from the Kinect 2 3D depth camera
and the Robotiq FT sensor and transfer compensation
values and sensory information to the other
computer by User Datagram Protocol communications.
The other computer is used to receive sensory information
from the Baxter robot and the first computer and
then calculates feedback control input
x fb in (6) and
2.5
3
1.5
2
0.5
1
2 468 10
12 14 16 18 20
Time (s)
Filtered Angle
Predicted Angle
Elbow Joint Angle
Figure 5. Elbow joint, predicted and filtered angle.
32 * IEEE ROBOTICS & AUTOMATION MAGAZINE * DECEMBER 2022
Experimental Objectives
As mentioned in the preceding section, a human and a
robot will cooperate to carry an object in this experiment.
Considering human-robot cotransporting, we set the experimental
objectives as follows:
1) Minimize the position error between the human's motion
and the robot's motion during cotransporting, which can
prove the effectiveness of the proposed prediction method
for reducing delay.
2) Minimize the interaction force between human and robot,
which means that the human can pay less force to transport
the object.
In the experimental results, we will analyze the mean
square error (MSE) and interaction force between
human and robot and verify the effectiveness of the proposed
method.
Experimental Results
In this section, we compare our hybrid method proposed
in the section " Control Framework " with other
state-of-the-art methods: the V method, vision-prediction
(VP) method, and VF method. In the V method,
the robot tracks human motion hi under a visual servoing
controller. After obtaining elbow joint angle hi
in the section " Visual Sensing of Human Motion, " we
use a visual servoing torque control method for the robot
to track hi for comparison. This visual servoing controller
() ()
xi hii i
and KvD
vvPh vD
where KvP
KK
=- -- -oo as in (6),
denote proportional and differential
gains. In the VP method, the robot tracks the predicted
angle hoit
ii ax=oo
dh
r
tt
under the visual servoing controller. In the VF
is not involved, which means
method, our proposed hybrid controller (6) is employed,
but the predicted value hoit
that we design the update law of the desired trajectory di as
() ()
Composite Filter Evaluation
In this part, we evaluate the effectiveness of the composite
filter proposed in the section " Signal Filtering. "
From Figure 5, we can see that, under our prediction
method proposed in the section " Human Motion Prediction, "
the curve of the predicted angle (light blue) is
similar to the curve of the actual one, but there exist
noises in the predicted signals. After filtering by our
filters proposed in the section " Signal Filtering, " the filtered
signals (deep blue) become smoother. The predicted
angle shows a prediction of the actual elbow joint
angle (red).
Comparison Between VP Method and V Method
We compare the VP and V methods to show the effectiveness
of our proposed prediction method. As shown
Angle (rad)
IEEE Robotics & Automation Magazine - December 2022

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