Professional Documents
Culture Documents
2021 Force - Disturbance - Observer-Based - Force - Control - For - Compliant - Interaction - With - Dynamic - Environment
2021 Force - Disturbance - Observer-Based - Force - Control - For - Compliant - Interaction - With - Dynamic - Environment
Abstract—Disturbances are one of the major challenges that quickly react to collisions is presented in [10]. Furthermore,
should be dealt with when designing high performance force other researchers have investigated using distributed joint
control systems for robots that interact with unknown environ- torque sensors to detect external forces and utilize the infor-
ments. To achieve high performance dynamic interaction, this
mation to control the joints [11], [12].
2021 IEEE International Conference on Mechatronics (ICM) | 978-1-7281-4442-9/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICM46511.2021.9385628
Authorized licensed use limited to: Sejong Univ. Downloaded on January 28,2022 at 00:59:19 UTC from IEEE Xplore. Restrictions apply.
20
Magnitude (dB)
0
-20
-40
-1 0 1 2
10 10 10 10
Authorized licensed use limited to: Sejong Univ. Downloaded on January 28,2022 at 00:59:19 UTC from IEEE Xplore. Restrictions apply.
0
Magnitude (dB)
-20
-40
Fig. 5. The proposed FDOB attached to the open-loop system in Fig. 3. Where 100 102 104
τc , Fnoise , Fmeas , τ̂d , and Q are auxiliary control input, force measurement Frequency (Hz)
noises, force sensor measurement, estimate of τd , and Q-filter, respectively.
Fig. 6. Influence of force measurement noises to the estimated disturbances.
where Q = (ωQ /(s + ωQ ))2 , ωQ is the Q-filter bandwidth,
and T̂d , Tm , and Fmeas are the Laplace transforms of τ̂d , τm , used here and on-wards are given in Table I. It can be
and Fmeas , respectively. observed that the force measurement noise intensity in the
The closed-loop sensitivity, SF DOB , and complimentary estimated disturbance becomes larger as ωQ is increased.
sensitivity, TF DOB , transfer functions are derived from Fig. 5 Hence, the FDOB bandwidth can be increased to 30 Hz,
as follows thereby providing good robustness performance with allowable
1−Q amount of measurement noises.
SF DOB (s) = , (4)
1 − Q + QG−1 n Gol III. C LOSED - LOOP C OMPLIANT F ORCE C ONTROL
QG−1n Gol Fig. 7 shows a block diagram of the two-degree-of-freedom
TF DOB (s) = 1 − SF DOB (s) = . (5)
1 − Q + QG−1 n Gol robust force control structure proposed in this study. It consists
At low frequencies, i.e., Q = 1, SF DOB → 0 and TF DOB → of the feedback controller, Cf b , a feedforward controller, Cf f ,
1, whereas at high frequencies, i.e., Q = 0, SF DOB → 1 and the FDOB which was presented in the previous section.
and TF DOB → 0. Therefore, (4) and (5) show that the The LPF is given by LP F (s) = N −1 ωL /(s + ωL ), where ωL
FDOB-based controller can precisely suppress disturbances is its bandwidth.
and noise at asymptotic frequencies. Further, the close-loop A. Design of Feedforward and Feedback Controllers
transfer functions of the FDOB loop in Fig. 5, are given by Each of the controllers is designed by utilizing the nominal
Gol model, Gn , in (2). Cf b is designed as a propotional-integral-
Tτc →F (s) = , (6)
1 − Q + QG−1 n Gol derivative (PID) controller whose gains are calculated by pole
(1 − Q)Gol cancellation technique based on the nominal plant model.
Tτd →F (s) = , (7) Whereas Cf f is designed as Cf f = G−1
1 − Q + QG−1 n Gol n × LP F , where the
LPF is used to make the controller proper.
where T◦→• represents the closed-loop transfer function from
◦ to •. It is observed from (6) and (7) that, at low frequencies, B. Analysis of LPF Bandwidth
i.e., when Q = 1, Tτc →F is normalized to Gn by the FDOB, There is a trade-off between noise attenuation and dis-
and Tτd →F → 0, hence, the influence of the force disturbances turbance estimation in the proposed force control structure.
is fully regulated by the proposed FDOB. To analyze this phenomenon, the sensitivity function of the
3) Q-filter Bandwidth Analysis: The force sensor measure- FDOB-based force control is derived from Fig. 7 as follows
ment noise, Fnoise , and the lumped disturbances, τd , are 1−Q
independent and uncorrelated since Fnoise is a force sensor SF DOB (s) = . (9)
1 − Q + Gol (LP F × Cf b + QG−1n )
characteristic. In spite of this fact, Fnoise can affect the FDOB
performance. Therefore, the influence of force measurement The frequency response characteristics of (9) are plotted
noises on the estimated disturbances is determined by the in Fig. 8 at varying values of ωL when ωQ = 30 Hz.
following transfer function It is observed that, robustness performance deteriorates as
ωL increases, an indication that the system becomes more
−QG−1
TFnoise →τ̂d (s) = n
. (8) sensitive to noises. Therefore, the reasonable value of ωL
1 − Q + QG−1n Gol can be determined from Fig. 8 as ωL = 10 Hz where
Bode plot of (8) is given in Fig. 6 which shows the amount both noise attenuation and accurate disturbance estimation and
of Fnoise in τ̂d at different ωQ values. The other parameters suppression performances are maintained.
TABLE I C. Reference Tracking Performance
PARAMETER VALUES
From Fig. 7, the tracking performance of the closed-loop
Parameter Value Parameter Value system is derived as follows
Jm 7.441e-4 kg·m2 Ml 11.257 kg
Bm 2.1e-3 Nm·s/rad Bl 0.55 N·s/m Gol (Cf f + Cf b )
Tτref →F (s) = . (10)
K 1 N/m N 1.923 1 − Q + Gol (LP F × Cf b + QG−1
n )
Authorized licensed use limited to: Sejong Univ. Downloaded on January 28,2022 at 00:59:19 UTC from IEEE Xplore. Restrictions apply.
Fig. 7. Block diagram of overall robust force control structure. τref , Cf f , and Cf b are reference torque input, feedforward controller, and feedback controller,
respectively. The rest of the parameters are as defined before.
50 40
Magnitude (dB)
Magnitude (dB)
0 20
-50
0
-100
-20
-1 0 1 2 3
-150 10 10 10 10 10
0 2 4 Frequency (Hz)
10 10 10
Frequency (Hz) Fig. 9. Tracking performance analysis.
Fig. 8. Bode-plot of the closed-loop sensitivity transfer function.
Thus, (13), (14), and (15) can be combined by the relationship
It is observed from (10) that the FDOB affects the behavior of F = Φ1 (Tref + Φ2 Fnoise + Φ3 Td ), (16)
the closed-loop control system. This is because of the presence
of the QG−1 term in the denominator. Fig. 9 is the Bode which expresses the influence of noise and disturbances on the
n
diagram of (10) when ωQ = 30 Hz and ωL is varying. There reference tracking. In (16), F, Tref , Fnoise , and Td are the
are highest resonances at low values of ωL , which deteriorates Laplace transforms of F , τref , Fnoise , and τd , respectively.
stability and robustness, resulting in bad tracking performance. Further, the frequency characteristics of (16) are plotted in
As ωL increases, the level of resonances reduces and hence Fig. 10 in red colored lines while gray lines show those of
improved stability. However, this allows more high frequency when FDOB is not implemented (conventional force control
noises into the system. with force sensor feedback only). Implementation of FDOB
shows that there is less effect of low frequency disturbances on
D. Influence of Fnoise and τd on Reference Tracking the reference tracking as compared to when the FDOB is not
The closed-loop transfer functions from Fnoise and τd to utilized (dash-dot lines), whereas the influence of measurement
the output force F are derived from Fig. 7 as noises is similar in both cases (dashed lines). Comparing the
−Gol (LP F × Cf b + QG−1 reference tracking, i.e., Φ1 for both cases, (solid lines), the
n )
TFnoise →F (s) = , (11) proposed control scheme exhibits a lower resonance than the
1 − Q + Gol (LP F × Cf b + QG−1n ) conventional method, which is an indication of better control
(1 − Q)Gol performance of the FDOB-based force control method.
Tτd →F (s) = . (12)
1 − Q + Gol (LP F × Cf b + QG−1
n )
IV. E XPERIMENTAL V ERIFICATION
If Tτref →F (s) in (10) is represented by Φ1 , i.e., In this section, experiments are conducted to validate the
Φ1 (s) = Tτref →F (s), (13) performance of the proposed force control scheme. Moreover,
for comparison purposes with the proposed FDOB-based
the influence of Fnoise and τd on τref can be calculated from
control, force control without the FDOB is also carried-out.
(11), (12), and (13) as given below
The apparatus in Fig. 2 is utilized in experiments while the
TFnoise →F (s) −(LP F × Cf b + QG−1
n ) parameter values are given in Table I.
Φ2 (s) = = , (14)
Tτref →F (s) Cf f + Cf b A. Reference Tracking Performance Experiments
Tτd →F (s) 1−Q
Φ3 (s) = = . (15) The first experiment is when τref is supplied as: 0 Nm
Tτref →F (s) Cf f + Cf b from 0 s to 1 s, 1 Nm magnitude sine signal from 1 s to
Authorized licensed use limited to: Sejong Univ. Downloaded on January 28,2022 at 00:59:19 UTC from IEEE Xplore. Restrictions apply.
50 2
Torque (Nm)
Feedback with FDOB Ref. torque input
1
Magnitude (dB) 0 0
-1
-50
0 1 2 3 4 5
Time (s)
-100
(a)
2
Torque (Nm)
-150 Feedback with no FDOB Ref. torque input
1
-200 0
10-2 10 0
10 2
10 4 -1
Frequency (Hz) 0 1 2 3 4 5
Fig. 10. Bode-plot showing the influence of measurement noise and distur- Time (s)
bances to the overall force control performance. (b)
Variance (Nm2 )
0.2
Feedback with no FDOB
4 s, and 0 Nm from 4 s to 5 s. As expected, force control Feedback with FDOB
with FDOB shows better reference tracking performance than 0.1
when FDOB is not utilized, as shown in Fig. 11(a) and (b).
To compare the performance, variances for the tracking errors 0
are calculated and plotted in Fig. 11(c) where large error 0 1 2 3 4 5
Time (s)
variances are observed when FDOB is not utilized in force
(c)
control. Moreover, the root mean square error (RMSE) values
are calculated as 0.0796 Nm for feedback with FDOB and Fig. 11. Reference tracking performance. (a) Force control with FDOB. (b)
Force control with out FDOB. (c) Tracking error variances.
0.2879 Nm for feedback with out FDOB, which confirm the
superior performance of the proposed method.
In the second experiment, τref , which is a swept-sine signal 1
Torque (Nm)
Fig. 12(a) and feedback without the FDOB in Fig. 12(b). The
force feedback signal with FDOB is observed to be superior to -1
0 2 4 6 8 10
the direct feedback control. Moreover, much as performance
Time (s)
is deteriorated as frequency of the reference input increases,
(a)
the proposed method outperforms the conventional method.
1
Torque (Nm)
Authorized licensed use limited to: Sejong Univ. Downloaded on January 28,2022 at 00:59:19 UTC from IEEE Xplore. Restrictions apply.
Reference tracking performance Slow push and pull Slow push and pull
0.15 2 1
Applied external Feedback
Torque (Nm)
Torque (Nm)
Torque (Nm)
input 0
0.05 Feedback with no FDOB Ref. torque input 0 Ref. torque
Feedback with FDOB input
Feedback -1
0 with FDOB
0 2 4 6 8 10 -1 Applied external
torque
Time (s)
(a) -2 -2
17 17.5 18 18.5 19 6 7 8 9
Time (s) Time (s)
Disturbance suppression performance
0.2
Estimated disturbance Actual disturbance input Fast push and pull Fast push and pull
Torque (Nm)
1.5 0.5
0 1 0
Torque (Nm)
Torque (Nm)
0.5 -0.5
-0.2 0 -1
0 2 4 6 8 10
Time (s) -0.5 -1.5
(b)
-1 -2
Fig. 13. Robustness to perturbation at the motor input when the reference 8.8 9 9.2 9.4 15.5 16 16.5 17
force is supplied. Time (s) Time (s)
the proposed FDOB-based force control is implemented than Fig. 14. Zero force control experimental results.
the conventional direct force sensor feedback control.
V. C ONCLUSIONS [9] K. S. Kim, A. S. Kwok, G. C. Thomas, and L. Sentis, “Fully omnidirec-
tional compliance in mobile robots via drive-torque sensor feedback,”
Force control utilizing force disturbance observer was pro- in 2014 IEEE/RSJ International Conference on Intelligent Robots and
posed in this paper. This method suppresses the effect of Systems, pp. 4757–4763, IEEE, 2014.
unknown disturbances, which improves the overall control [10] K. S. Kim, T. Llado, and L. Sentis, “Full-body collision detection and
reaction with omnidirectional mobile platforms: a step towards safe
performance, and hence compliance with the dynamic forces human–robot interaction,” Autonomous Robots, vol. 40, no. 2, pp. 325–
from the external environment is improved. Robust force 341, 2016.
control experiments were conducted on an actual robot and [11] A. De Luca, A. Albu-Schaffer, S. Haddadin, and G. Hirzinger, “Collision
detection and safe reaction with the dlr-iii lightweight manipulator arm,”
results showed that the proposed method exhibits superior per- in 2006 IEEE/RSJ International Conference on Intelligent Robots and
formance over the conventional force sensor feedback control. Systems, pp. 1623–1630, IEEE, 2006.
[12] C.-h. Wu, “Compliance control of a robot manipulator based on joint
R EFERENCES torque servo,” The International journal of robotics research, vol. 4,
no. 3, pp. 55–71, 1985.
[1] V. Villani, F. Pini, F. Leali, and C. Secchi, “Survey on human– [13] K. Ohnishi, M. Shibata, and T. Murakami, “Motion control for advanced
robot collaboration in industrial settings: Safety, intuitive interfaces and mechatronics,” IEEE/ASME transactions on mechatronics, vol. 1, no. 1,
applications,” Mechatronics, vol. 55, pp. 248–266, 2018. pp. 56–67, 1996.
[2] S. Haddadin, A. De Luca, and A. Albu-Schäffer, “Robot collisions: A [14] E. Sariyildiz and K. Ohnishi, “A guide to design disturbance observer,”
survey on detection, isolation, and identification,” IEEE Transactions on Journal of Dynamic Systems, Measurement, and Control, vol. 136, no. 2,
Robotics, vol. 33, no. 6, pp. 1292–1312, 2017. 2014.
[3] E. Colgate, A. Bicchi, M. A. Peshkin, and J. E. Colgate, “Safety for [15] T. Murakami, N. Oda, Y. Miyasaka, and K. Ohnishi, “A motion control
physical human-robot interaction,” in Springer handbook of robotics, strategy based on equivalent mass matrix in multidegree-of-freedom
pp. 1335–1348, Springer, 2008. manipulator,” IEEE Transactions on Industrial Electronics, vol. 42,
[4] B. Matthias, S. Kock, H. Jerregard, M. Kallman, I. Lundberg, and no. 2, pp. 123–130, 1995.
R. Mellander, “Safety of collaborative industrial robots: Certification [16] S. Sakaino, T. Sato, and K. Ohnishi, “Force-based disturbance observer
possibilities for a collaborative assembly robot concept,” in 2011 IEEE for dynamic force control and a position/force hybrid controller,” IEEJ
International Symposium on Assembly and Manufacturing (ISAM), transactions on electrical and electronic engineering, vol. 8, no. 5,
pp. 1–6, Ieee, 2011. pp. 505–514, 2013.
[5] D. Zhao, X. Deng, and J. Yi, “Motion and internal force control for [17] S. Kangwagye and S. Oh, “Robotic stage for human balance disorder
omnidirectional wheeled mobile robots,” IEEE/ASME Transactions on assessment,” in 2020 20th International Conference on Control, Automa-
Mechatronics, vol. 14, no. 3, pp. 382–387, 2009. tion and Systems (ICCAS), pp. 447–452, IEEE, 2020.
[6] R. V. Patel, H. A. Talebi, J. Jayender, and F. Shadpey, “A robust
position and force control strategy for 7-dof redundant manipulators,”
IEEE/ASME Transactions on Mechatronics, vol. 14, no. 5, pp. 575–589,
2009.
[7] Y. Li and S. S. Ge, “Human–robot collaboration based on motion in-
tention estimation,” IEEE/ASME Transactions on Mechatronics, vol. 19,
no. 3, pp. 1007–1014, 2013.
[8] W. S. Newman, “Stability and performance limits of interaction con-
trollers,” 1992.
Authorized licensed use limited to: Sejong Univ. Downloaded on January 28,2022 at 00:59:19 UTC from IEEE Xplore. Restrictions apply.