Interpolation of Missing Data

for Wireless Operation of Industrial Robot

Yosuke TAKAGI Chedlia BEN NAILA, Hiraku OKADA
Dept. Information and Communication Engineering and Masaaki KATAYAMA
Nagoya University Institute of Materials and Systems for Sustainability
Furo-cho, Chikusa-ku, Nagoya 464-8603, Japan Nagoya University
Email: [email protected] Furo-cho, Chikusa-ku, Nagoya 464-8603, Japan
IECON 2023- 49th Annual Conference of the IEEE Industrial Electronics Society | 979-8-3503-3182-0/23/$31.00 ©2023 IEEE | DOI: 10.1109/IECON51785.2023.10312364

0000-0002-0556-2745, 0000-0003-2276-6421,
0000-0002-7612-390X

Abstract—This paper proposes an industrial robot system that order to improve the operation performance of a remote-
interpolates data loss due to imperfect communication. Most control system, improving communication quality by using
wireless communication standards adopt Carrier Sense Multiple more sophisticated schemes such as multi-channel transmis-
Access protocol. This protocol may introduce intentional delay
in packet transmissions to avoid interference, on top of the delay sion may be one solution. However, this work introduces
caused by transmission failure and retransmission. This delay can another approach from the control software configuration,
vary the reception timing of commands to the robot, and affect which does not require changing communication hardware.
the behavior of the robot. To overcome this problem, we develop Wireless communication channels sometimes fail to trans-
an algorithm that interpolates the data not received appropriately mit data at the appropriate timing due to signal interference
at the robot side. Our findings demonstrate that the proposed
method reduces unnecessary behavior (vibration) compared to and the behavior of communication protocol. Under such
the conventional one, thus improving the operability robot. communication, the robot may not behave as operated, i.e.,
Index Terms—Industrial Robot, Wireless Control, Operability, impaired control performance and operability. However, in
6DoF Robot, Control Communication, Interpolation industrial areas where many electronic devices are used, it may
be very difficult to completely suppress the signal interference
I. I NTRODUCTION completely.
An industrial robot is a multi-axis robot arm that performs Therefore, this paper developed a system for wireless opera-
tasks considered monotonous or unsafe for humans [1, 2]. tion of industrial robots, taking into account the imperfection
Though the industrial robot is designed to execute operations of wireless communication to improve operability.
automatically by precoded programs, there are some tasks
II. SYSTEM ARCHITECTURE
where manual control is still necessary to operate the robot
[3]. In such cases, it is often needed to confirm the positions A. Hardware configuration
and orientations of the robot and its task objects from multiple Fig. 1 and Table I show the configuration and specifications
perspectives walking around the robot. To achieve this, wire- of the robot system used in the experiments. As shown, the
less control is a solution to improve the operator’s freedom joy controller sends control inputs which consist of open/close
and efficiency of work. command and three-dimensional movement and rotation of
Regarding the wireless operation of industrial robots, several the gripper to the operator computer (OC). The OC sends the
studies have proposed different control systems to command command data to the robot computer (RC) connected with OC
robots based on the detection of human motion [4]–[6]. via the network emulator that emulates a wireless network.
For instance, a robot control system using either a sensor- The RC calculates and sends the control data to the robot
embedded band [4] or accelerometer [5] attached to the controller.
operator’s arm has been proposed. In [6], a robot control
system that employs a Kinect sensor and a control system B. Software configuration
that detects arm movements using an accelerometer built into We built a control system based on Robot Operating System
an Android smartphone have been proposed [7]. In [8]–[10], (noetic). The software configuration of the OC and RC is
the authors have also presented master-slave control systems
using wireless communication. Whereas, the authors in [11,
TABLE I: system specifications
12] have developed a wireless control system for multi-axis
robots with grippers that implement wearable interfaces. In Industrial robot Kawasaki RS007N
Robot controller Kawasaki F60
This research is subsidized by New Energy and Industrial Technology Gripper Robotiq 2F-140
Development Organization (NEDO) under a project JPNP20016. This paper Network emulator Apposite Netropy N61
is one of the achievements of joint research with ROBOT Industrial Basic Joy controller DualSense wireless controller
Technology Collaborative Innovation Partnership (ROBOCIP). CFI-ZCT1J

Authorized licensed use limited to: Makerere University Library. Downloaded on December 29,2023 at 15:34:10 UTC from IEEE Xplore. Restrictions apply.
 (a) arrivals from OC (b) output to RC
Fig. 2: Example of the transmission and rception timing of
position data with group arrival
Fig. 1: Configuration of the robot system

illustrated in Fig.1.
In OC, the posture calculator [see OC of Fig. 1] calculates
the position and orientation of the robot based on the joy
controller output and the feedback from the RC. The gripper
controller [see OC of Fig. 1] determines the open/close com- (b)
(a)
mand of the gripper based on the the joy controller output.
The communication block [see OC of Fig. 1] in the OC then Fig. 3: Behavior of robot with conventional method (a) and
sends these commands to RC using User Datagram Protocol proposed method (b)
(UDP) packet.
In RC, the communication block [see RC of Fig. 1] receives
UDP packets and extracts position and orientation of the follows the procedure above. In addition, if the user receives
robot and open/close commands. This study proposed the new packets to the same destination before the retransmission
introduction interpolation block [see RC of Fig. 1] to mitigate of a packet and if the size of the packets is small enough,
the effects of imperfect wireless communication. Interpolation then the transmitter may transmit multiple packets in a single
block refers to the position and orientation and stores them for radio frame. As the results of the delays of CSMA and
a certain period. If the interpolation block detects any packet retransmission, together with the group arrival of concatenated
loss, it generates interpolated data. The detailed functions of packets, the control packets sent at a constant interval from OC
interpolation block are described in Section 3. Robot joint may arrive at RC at random intervals.
manager calculates the posture of the robot based on the Fig. 2 (a) shows an example of arrival tumings of packets
received position and orientation data and the current value carrying the robot gripper position and orientation data at the
of each axis. Then, robot joint manager sends the calculated transmitter from OC. Fig. 2 (b) shows that the packet x[1]
values to the robot controller. is transmitted immediately after arrival to the transmitter and
At the same time, the gripper manager [see RC of Fig. 1] detected correctly at the receiver. In this case, there is only
refers to the open/close command of gripper. The gripper a negligible small transmission delay of x[1] from OC and
manager determines the state of the gripper based on referred RC. On the other hand, the packets x[2] and x[3] are not
command and current status. Then gripper manager calculates transmitted upon their arrival to the transmitter because of
control input of the gripper and sends it to the gripper. channel congestion or are corrupted by jamming or noise at
the receiver. In this figure, these packets are transmitted and
III. C ONTROL WITH THE INTERPOLATION BLOCK received correctly and simultaneously with the packet x[4].
A. Issues and solution Fig. 3 (a) shows the behavior of the robot with the conven-
Wireless communication systems, including IEEE 802.11, tional system without the interpolation block in Fig.1. The
often adopt Career Sense Multiple Access with Colision robot stops operating after receiving x[1] until it receives
Avoidance protocol (CSMA/CA). With this protocol, a user the next data. When the robot receives multiple control data
station senses the channel status before transmitting a packet. (x[2], x[3], x[4]) simultaneously after the pause, then it oper-
If the channel is idle, i.e., no other user is sending a packet, ates as per the latest data x[4]. This strategy of using the data
then the user sends its packet. If the channel is in use, the may cause a sudden movement in the robot and may result in
user suspends and schedules the packet transmission at a later unpredicted vibration of robot movement.
time according to some delay distribution. At the end of the In this study, we introduced a method for RC to reduce the
delay interval, the user again senses the channel and repeats degradation of the control performance and the operability.
the same steps. After transmitting a packet, if the user does Fig. 3 (b) shows the behavior of the robot with respect to the
not receive an acknowledgment packet for the transmitted proposed method. With this method, the received control data
packet, then the user schedules the retransmission later and is stored for a fixed period of time, ∆T , instead of immediately

Authorized licensed use limited to: Makerere University Library. Downloaded on December 29,2023 at 15:34:10 UTC from IEEE Xplore. Restrictions apply.
sending it to the robot. Within the period, if the RC realizes
that some packets are missing, then it regenerates the data of
the missing packet by interpolation.

B. Design of the interpolation block

Fig. 4 shows the software configuration of ”Interpolation
block[see RC of Fig. 1]” to realize the proposed method.
In the figure, input[k] denotes the k-th received packet
data consisting of the gripper position and orientation, thus
6 dimensional data[k] and Identif icationN umber[k] that
is used to detect data loss. Though the arrival interval of
input[k] at RC is not constant because of the random delay,
the block outputs output[k] every T seconds, where T is the
transmission interval of the packets.
At the start of the operation, the block stores position and
orientation data for a fixed period ∆T and outputs them after
the period. Thus the data of the starting packet data[0] is
output as dataout [0] at ∆T seconds after the starting time.
At OC, a sequential number is given to each packet that
carries the gripper position and orientation, and this number
is denoted as Identif icationN umber[k] at RC. Using this
data part, RC can detect packet loss. For example, there is no
packet lost between k-th and (k+1)-th received packet if we Fig. 4: Operation of interpolation block
obtain following equation:
Identif icationN umber[k + 1]
(1) gripper when the joy controller sent the inputs to make a linear
−Identif icationN umber[k] = 1 motion.
On the other hand, there are n − 1 lost packets if we obtain Table II shows the parameters used in the experiments.
following inequation: In this experiment, the system adopts a network emulator
instead of wireless communication module between the OC
Identif icationN umber[k + 1]
(2) and RC. It causes transmission errors to reproduce imperfect
−Identif icationN umber[k] > 1 communication like wireless network. To reproduce the delay
Once ”Interpolation block” detects a packet loss, the effector interval stated in Section 3, the network emulator causes burst
data of the lost packet to be output as dataout is calculated by error in which multiple data are lost continuously with a
linear interpolation using the data of the two packets closest in certain probability. The burst length is 10–20 when the burst
time before and after the lost packets. For example, if the block occurrence probability is 2% and 25–35 when the burst occur-
detects one packet loss between data[k] and data[k + 1], the rence probability is 10%. The burst length occurs with equal
6 dimensional output of the block can be expressed as: probability from these values. The burst occurrence probability
denotes the probability that a certain packet becomes the head
dataout = (data[k] + data[k + 1])/2 (3) of a burst error.
Fig. 5, 6, and 7 show the time variation of the gripper’s
position in the direction of movement for each pattern. The
results show that the robot temporarily stops the motion
In the above procedure, choosing the appropriate value of and moves to catch up to the expected location with the
the period ∆T is necessary. If the value of the fixed period is conventional system, resulting in vibration in the behavior
too small, interpolation may often fail with burst packet loss. of the robot. The vibration increases with the increase in the
On the other hand, if the period is too large, the additional burst occurrence probability. In contrast, the system with the
∆T second delay may impair the robot’s operability. Thus, In proposed method can reduce the vibration by interpolation.
this paper, we set the value of ∆T to 0.1 [s], which is almost
imperceptible to humans based on the results of perceptual
psychology. TABLE II: Experiment specifications
IV. E VALUATION OF CONTROL PERFORMANCE Transmission period 0.002 s
[tb] We conducted experiment to evaluate the control per- burst occurrence probability 0, 2, 10 %
burst length 10 - 35 data
formance of the proposed method under imperfect commu- gripper velocity 0.20 m/s
nication. In the experiment, we measured the position of the

Authorized licensed use limited to: Makerere University Library. Downloaded on December 29,2023 at 15:34:10 UTC from IEEE Xplore. Restrictions apply.
Fig. 5: Time variation position of the direction of movement Fig. 7: Time variation position of the direction of movement
for each method at 0% burst occurrence probability for each method at 10% burst occurrence probability

Fig. 10 shows the flow of the evaluation. The participants

first perform the task for five times using the conventional
system under the lossless communication channel to master the
task. They then perform an evaluation that consists of two tasks
per set; the former is performed using the conventional system,
while the latter is performed using either the conventional or
proposed system. The network emulator causes no errors in the
former case, but it exhibits in the latter case. Here, the patterns
of the burst occurrence probability and the burst lengths are the
same as that mentioned in Section 3 (Table 2). The participants
compare the operability of both cases and evaluate the latter
one using a 5-point evaluation shown in Table III. In the latter
case, the participants don’t know which pattern is caused by
Fig. 6: Time variation position of the direction of movement
the network emulator. This set is repeated, and all patterns (4
for each method at 2% burst occurrence probability
patterns) are evaluated three times each.
The experiments were conducted with eight participants
However, in the case of a burst occurrence probability of 10%, (aged 21–25 years). Thus, we evaluated the operability by OS
a little vibration occurs even when the proposed method is and MOS.
applied. In such a case, a larger value of the storage time ∆T
is needed for smaller vibration but with the cost of a lager
control delay. We need to consider the trade-off between these
two control performance degradation factors. Even though the
burst occurrence probability increases, the occurrence of the
vibration is less than that with the conventional one.
These results show that the proposed method has advantages
in terms of control performance.

V. E VALUATION OF THE OPERABILITY

A. Flow of evaluation
The operability of the system was evaluated by conducting
experiments using opinion score (OS) as a measure of evalua-
tion. Fig. 8 and 9 show the environments of the experiment. In
the experiments, the participants perform a task, i.e., picking
a block with the gripper and placing it into a frame. The
participants then evaluate the operability of the task based on
the difficulty of the operation. Fig. 8: The robot system used in the evaluation

Authorized licensed use limited to: Makerere University Library. Downloaded on December 29,2023 at 15:34:10 UTC from IEEE Xplore. Restrictions apply.
 [h] TABLE III: summary of evaluation
Point Impairment
5 Imperceptible
4 Perceptible but not annoying
3 Slightly annoying
2 Annoying
1 Very annoying

TABLE IV: MOS of each method

Burst occurrence probability Conventional method Proposed method
2% (burst length of 10-20) 2.88 4.67
10% (burst length of 25-35) 1.67 3.58

the proposed method is advantageous over the conventional

one in terms of operability.
Here, the operability is impaired from the ideal commu-
nication channel probably due to the delay caused by the
storage period, and the vibration that could not be completely
suppressed.
Fig. 9: Experimental environment

(a) (b)
Fig. 11: Relative frequency distribution of OS of the conven-
tional (a) and proposed (b) methods at 2% burst occurrence
probability

(a) (b)
Fig. 12: Relative frequency distribution of OS of the conven-
tional (a) and proposed (b) methods at 10% burst occurrence
probability

Fig. 10: Flow of the operability evaluation

VI. C ONCLUSION
In this study, we proposed and evaluated a method to
B. Results improve the operability of industrial robots under imperfect
Fig. 11 and 12 show the relative frequency distribution of communication channels.
OS for each pattern. Table IV shows the MOS for each pattern. Based on the evaluation of control performance, the vibra-
These results show that the proposed method obtained higher tion of the industrial robot caused by missing data was con-
OS than the conventional one, including that the proposed firmed to be reduced by interpolation. Subjective evaluation
method successfully reduced the impairment of operability showed that the proposed method reduced the impairment of
compared to the conventional one. These results confirmed that operability.

Authorized licensed use limited to: Makerere University Library. Downloaded on December 29,2023 at 15:34:10 UTC from IEEE Xplore. Restrictions apply.
 ACKNOWLEDGMENTS
The authors would like to thank Prof. Takaya Yamazato
of Nagoya University for his guidance and encouragement in
carrying out this research.
R EFERENCES
[1] C. Jiao, H. Li, “ A Trajectory Planning Method for Industrial Robot
Weaving Welding Based on Piecewise Function,” International Confer-
ence on Robotics and Biomimetics (ROBIO), pp.1605-1610, 2021
[2] Y. Li, T. Cheng, L. Pan, “ Research and application of information
Model of industrial Robot Welding System based on OPC UA,” IEEE
2nd International Conference on Big Data, Artificial Intelligence and
Internet of Things Engineering (ICBAIE), pp.1084-1087, 2021
[3] T. Hatano, C. M. Horvath, T. Thomessen and M. Niitsuma, ”A vi-
brotactile navigation aid for remote operation of an industrial robot,”
International Symposium on System Integration (SII), pp. 700-705, 2016
[4] M. Coban, G. Gelen, “ Wireless teleoperationof an industrial robot by
using myo arm band,” 2018 International Conference on Artificial In-
telligence and Data Processing (IDAP), pp.1-6, 2018
[5] P. Neto, J. N. Pires, A. P. Moreira, “ Accelerometer-based control of
an industrialrobotic arm,” RO-MAN 2009 - The 18th IEEEInternational
Symposium on Robot and HumanInteractive Communication, pp.1192-
1197, 2009
[6] H. T. Salamea, A. R. Auquilla, O. AlvaradoCando, C. Urrea Oñate, ”
Control of a Telerobotic System Using Wi-Fi and Kinect Sensor for
Disabled people with an industrial Process,” 2020 IEEE ANDESCON,
pp.1-6, 2020
[7] R. K. Singh, A. Sarkar, D. Chakravarty, P. Goyal, V. Lodhi, A. Sharma,
” Bluetooth communication controlled robot based on gesture recogni-
tion,” 2015 IEEE International Transportation Electrification Conference
(ITEC), pp.1-5, 2015
[8] R. K. Megalingam, S. Boddupalli, K. G. S.Apuroop,” Robotic arm
control through mimicking of miniature robotic arm,” 2017 4th Inter-
national Conference on Advanced Computingand Communication Sys-
tems (ICACCS), pp.1-7, 2017
[9] H. M. T. Salamea, D. D. T. Torres, P. D. U. Cardenas, C. U. Onate, ”
Implementation of the Hamming code for the detection and correction
of errors in a telerobotic system using an industrial communication
protocol,” 2020 IEEE ANDESCON, pp.1-6, 2020
[10] C. Pan, X. Liu, W. Jiang, ” Design and synchronization control of
heterogeneous robotic teleoperation system,” 2017 Chinese Automation
Congress (CAC), pp.406-410, 2017
[11] V. Ryabchevsky, A. Nikonov, G. Nikonova, ” Development of pneumatic
artificial muscles and an anthropomorphic gripper control module,”
2021 International Siberian Conference on Control and Communications
(SIBCON), pp.1-4, 2021
[12] G. Salvietti, Z. Iqbal, I. Hussain, D. Prattichizzo, M. Malvezzi, ”
The co-gripper: A wireless cooperative gripper for safe human robot
interaction,” 2018 IEEE/RSJ International Conference on Intelligent
Robots and Systems (IROS), pp.4576-4581, 2018

Authorized licensed use limited to: Makerere University Library. Downloaded on December 29,2023 at 15:34:10 UTC from IEEE Xplore. Restrictions apply.