See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/269475573

Nonlinear PD controllers with gravity compensation for robot manipulators

Article in Cybernetics and Information Technologies · April 2014

DOI: 10.2478/cait-2014-0011

CITATIONS READS

11 1,082

3 authors, including:

Jun Ye
Ningbo University
331 PUBLICATIONS 14,777 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jun Ye on 09 May 2015.

The user has requested enhancement of the downloaded file.

BULGARIAN ACADEMY OF SCIENCES

CYBERNETICS AND INFORMATION TECHNOLOGIES • Volume 14, No 1

Sofia • 2014 Print ISSN: 1311-9702; Online ISSN: 1314-4081

DOI: 10.2478/cait-2014-0011

Nonlinear PD Controllers with Gravity Compensation

for Robot Manipulators

Jianfeng Huang, Chengying Yang, Jun Ye∗

Department of Electrical and Information Engineering, Shaoxing University, 508 Huancheng West
Road, Shaoxing, Zhejiang 312000, PR China
Emails: [email protected] [email protected], *[email protected]

Abstract: A Nonlinear Proportional-Derivative (NPD) controller with gravity

compensation is proposed and applied to robot manipulators in this paper. The
proportional and derivative gains are changed by the nonlinear function of errors
in the NPD controller. The closed-loop system, composed of nonlinear robot
dynamics and NPD controllers, is globally asymptotically stable in position control
of robot manipulators. The comparison of the simulation experiments in the
position control (the step response) of a robot manipulator with two degrees of
freedom is also presented to illustrate that the NPD controller is superior to the
conventional PD controller in a position control system. The experimental results
show that the NPD controller can obtain a faster response velocity and higher
position accuracy than the conventional PD controller in the position control of
robot manipulators because the proportional and derivative gains of the NPD
controller can be changed by the nonlinear function of errors. The NPD controller
provides a novel approach for robot control systems.
Keywords: Nonlinear proportional-derivative controller, proportional-derivative
controller, robot manipulator, position control.

∗
Corresponding author. Tel.: +86 575 88327323
E-mail address: [email protected] (Jun Ye).
This work was supported by the undergraduate science and technology innovation project of Zhejiang
province, PR China (No 2012R426009).
141
1. Introduction
Robot dynamics are highly nonlinear because of the coupling between joints. Due
to the parametric uncertainties in the system dynamics, it is difficult to derive the
exact description of the system. The position control (also called a regulation
problem) is one of the most relevant issues in the operation of robot manipulators.
This is a particular case of the motion control or trajectory control. The primary
goal of the motion control in the points space is to make the robot joints track a
given time-varying desired joint position. T a k e g a k i and A r i m o t o [1],
A r i m o t o and M i y a z a k i [2] showed that simple controllers, such as the
Proportional-Derivative (PD) controller and Proportional-Integral-Derivative (PID)
feedback controller are efficient for general control, despite the nonlinearity and
uncertainty of the robot dynamics. In recent years, various linear PD- or PID-type
control schemes have been extended to a nonlinear PID control strategy. A PD
controller with stability robustness in the presence of parametric uncertainty in the
gravitational torque vector was presented by H s i a [3]. A class of nonlinear PD-
type controllers for robot manipulators was proposed by K e l l y and C a r e l l i [4].
S e r a j i [5] presented the analysis and design of a nonlinear PID control with an
extension to tracking. B u c k l a e w and L i u [6] also proposed a nonlinear gain
structure for PD-type controllers in robotic applications. Furthermore, R e y e s and
R o s a d o [7] proposed a polynomial family of PD-type controllers for robot
manipulators. However, these PID controllers are difficult to determine the
appropriate PID gains in case of nonlinear and unknown controlled plants, and then
the PID controller with fixed parameters may usually deteriorate the control
performance. Therefore, various types of PID control have been developed by
means of neural networks [8-12]. However, neural networks may be difficult to
reach the real-time control of robot systems due to quick learning problems on line.
The strategy of PID control has been one of the most sophisticated and most
frequently used methods in industry. This is because the PID controller has a simple
form and strong robustness under broad operating conditions. However, the
conventional PID controller may usually deteriorate the control performance in
nonlinear control systems. The nonlinear PID-type controllers have been proved to
be a promising approach to solve nonlinear control problems and are adapted to the
control of robot manipulators, because the nonlinear PID controllers have the
nonlinear characteristics and advantages of PID controllers. Hence, the aim of this
paper is to propose a Nonlinear Proportional-Derivative (NPD) controller with
gravity compensation to control robot manipulators, which leads to global
asymptotic stability of the closed-loop system (dynamics model of a robot
manipulator plus controllers). This paper is organized as follows. Section 2
discusses the robot dynamics. In Section 3 a NPD controller with gravity
compensation is presented based on variable proportional and derivative gains
corresponding to the error. Section 4 contains the simulation experimental
comparison between the NPD controller and the conventional PD controller on a
robot arm with two degrees of freedom. The discussion of the experimental results

142
is given in Section 5. Finally, some conclusions and future work are offered in
Section 6.

2. Robot dynamics
The dynamics of a serial n-link rigid robot can be written as [7]:
(1) M (q)q + C (q, q ) + G (q) + F (q ) = τ ,
where q, q , q are the n×1 vectors of the joint displacement, velocity, and
acceleration; τ is the n×1 vector of input torques; M(q) is the n×n symmetric
positive definite manipulator inertia matrix; C (q, q ) is the n×n matrix of centripetal
and Coriolis torques; G(q) is the n×1 vector of gravitational torques obtained as the
gradient of the robot potential energy due to gravity; F (q ) is the n×1 vector for the
friction torques. The matrix C (q, q ) and the time derivative M  (q ) of the inertia
matrix satisfy:
⎡1  ⎤
(2) q T ⎢ M (q ) − C (q, q ) ⎥ q = 0.
⎣2 ⎦

3. NPD controllers
We introduce the design ideas for a NPD controller in the dynamic process of a
control system, described as follows.
For the proportional gain kp, when the control error is increased, kp is increased
under keeping the response velocity without the overshoot; while the control error
is decreased, kp is decreased to decrease the overshoot and to rapidly reach a stable
point under an adequate kp. According to the requirements, we select the shape of
the gain kp with respect to the change of the control error e as shown in Fig. 1.

kp

ap
0 e

Fig. 1. Variable curve of the proportional gain kp corresponding to the control error e

Based on Fig. 1, we present the following nonlinear function:

(3) k p (e) = a p + b p e 2p ,
143
where ap and bp are parameters, which can change the kp curve.
For the derivative gain kd, when the control error is increased, kd is decreased
to keep the response velocity without the overshoot; while the control error is
decreased, kd is increased to decrease the overshoot. According to the requirements,
we select the shape of the gain kd with respect to the change of the control error e,
as shown in Fig. 2.
kd

0 e

Fig. 2. Variable curve of the derivative gain kd corresponding to the error e

Based on Fig. 2, we present the following nonlinear function:

ad
(4) k d (e ) = ,
0.001 + e 2
where ad is a parameter. The denominator takes the smaller value 0.001,when e = 0
in (4) to overcome the undefined function.
The NPD controller for n degrees of freedom for the robot arm is presented by
the following control scheme with gravity compensation:
(5) τ = K p (e )e − K d (e )q + G (q ) + F (q ) ,
where e ∈ Rn×1 is the position error vector which is defined as e = qd – q, qd ∈ Rn×1
representing the desired joint position, Kp ∈ Rn×n is the proportional gain which is
diagonal matrix, and Kd ∈ Rn×n is the derivative gain which is diagonal matrix.
For the nth joint of the robot arm, the following controller can be obtained
from (5):
τ NPDn = k pn (en )en − k dn (en ) q n + g n ( q ) + f n ( q ) =
(6) adn
= ( a pn + b pn en2 ) en − q n + g n ( q ) + f n ( q ),
0.001 + en2

where τNPDn is the output torque of the nth NPD controller, which drives the nth
joint in the robot arm.

144
4. Simulation example
In this section, to verify the control efficiency of the proposed NPD controllers for
the robot arm, the proposed NPD controllers are employed in the position control of
a two-link robot as shown in Fig. 3.

m2
yy

l2 q2

m1

l1
q1

xx
Fig. 3. A two-link robot

In Fig. 2 m1 and m2 are masses of arm 1 and arm 2, respectively; l1 and l2 are
lengths of arm 1 and arm 2; τ1 and τ2 are driven torques on arm 1 and arm 2; q1 and
q2 are positions of arm 1 and arm 2. The dynamics model of the two-link robot is
the same as (1).
Let
q = [q1 , q2 ]T , q = [q1 , q2 ]T , q = [q1 , q2 ]T , τ = [τ 1 , τ 2 ]T ,
ci = cos(qi), si =sin(qi), cij = cos(qi+qj), sij = sin(qi+qj),
then M, V, G in (1) can be described as
⎡m l 2 + m 2 (l12 + l 22 + 2l1l 2 c 2 ) m 2 l 22 + m 2 l1 l 2 c 2 ⎤
M (q ) = ⎢ 1 1 ⎥,
⎣ m 2 l 22 + m 2 l1 l 2 c 2 m 2 l 22 ⎦
⎡ −m2l1l2 s2 q22 − 2m2l1l2 s2 q1q2 ⎤
C (q, q ) = ⎢
 ⎥,
⎣ m2l1l2 s2 q12 ⎦
⎡m l gc + (m1 + m 2 )l1 gc1 ⎤
G (q ) = ⎢ 2 2 12 ⎥.
⎣ m 2 l 2 gc12 ⎦
In this case the parameters of the two-link robot are m1 = 10 kg, m2 = 3 kg and
l1 = 1.1 m, l2 = 0.8 m.
To support the theoretical developments, this section presents an experimental
comparison of the two position controllers on a two-degree-of-freedom direct drive
robot manipulator, where the servo motors directly drive the joints without gear
reduction. The advantages of this type of a direct-drive actuator include freedom
145
from a backlash and significantly lower joint friction compared to the actuators
composed by gear drives [7]. To investigate the performance between controllers,
they are classified as τNPD for NPD controller andτPD for the conventional PD
controller. The applied torques of the actuators for joints 1 and 2 are chosen so that
τ1max ≤ 600 N.m and τ2max ≤ 200 N.m, respectively, by practical considerations,
because it can also produce torque saturation of the actuators.
An experiment of the position control is designed to compare the performance
of the controllers in a direct-drive robot. The experiment consists of moving the
end-effector from its initial position to a desired target (step response). For the
present application the desired point positions are chosen as: [qd1, qd2]T = [1, 1.5]T
radians, the initial positions and velocities are set to zero (for example, at home
position). The friction phenomena and disturbances are not modeled for
compensation purposes. That is, all the controllers do not show any type of friction
and disturbance compensations. Therefore, they consider the friction and
disturbance as unmodelled dynamics. The friction forces of the joints and the
disturbances are assumed (in N.m) as
⎡ 3q + 0.5sign(q1 ) ⎤ ⎡ 5cos(q1 ) ⎤
F (q ) = ⎢ 1 ⎥ , Td (q, q ) = ⎢ ⎥.
⎣ 2q2 + 0.5sign(q2 ) ⎦ ⎣5cos(q2 ) ⎦
Simulation experiments are carried out by using the NPD controller and the
conventional PD controller to select their gains according to the method in [7], such
that the best time response without an overshoot and a minimal steady-state position
error are obtained without going into the saturation zone of the actuator’s torques.
The final values of all simulation parameters are shown in Table 1.
Table 1. Simulation parameters of the position control for a robot manipulator with two-joints
Joint 1 Joint 2
Controllers
ap1 bp1 ad1 kp1 kd1 ap2 bp2 ad2 kp2 Kd2
NPD 90 560 14 / / 20 950 0.6 / /
PD / / / 460 160 / / / 115 28

4.1. Simulation experiment of the NPD controllers

The position control of a two-degree-of-freedom direct drive robot manipulator uses
the following NPD controllers:
ad 1
(7) τ NPD1 = (a p1 + bp1e12 )e1 − q1 + 23.52 cos(q1 + q2 ) + 140.14 cos(q1 ) ,
0.001 + e 2
ad 2
(8) τ NPD 2 = (a p 2 + b p 2e22 )e2 − q 2 + 23.52 cos(q1 + q2 ) ,
0.001 + e22
where τNPD1 and τNPD2 represent the applied torques for the two joints. The
experimental parameters are shown in Table 1.
The experimental results of the step response of joint 1 and joint 2 are shown
in Fig. 4 under NPD control (τNPD1 and τNPD2) without the overshoot. On the other
hand, the position errors of the NPD controllers (7) and (8), corresponding to two
joints are depicted in Fig. 5, which demonstrates the convergence properties for
each controller.
146
 1.6

1.4
q2

1.2

1
q1,q2(rad)

0.8
q1

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
t(s)

Fig. 4. Step response of the NPD control

1.5

e1

1
e1,e2(rad)

e2

0.5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

t(s)

Fig. 5. Position errors of the NPD control

147
4.2. Simulation experiment of the conventional PD controllers

For the simulation experiment of conventional PD controllers, the desired position

and initial conditions are the same as in the previous experiment. The PD
controllers for the robot arm (n = 2) are represented by the following equations:
(9) τ PD1 = k p1e1 − kd 1q1 + 23.52 cos(q1 + q2 ) + 140.14 cos(q1 ) ,
(10) τ PD 2 = k p 2e2 − kd 2 q2 + 23.52 cos(q1 + q2 ) ,
where τPD1 and τPD2 represent the applied torques for joints 1 and 2, respectively.
The experimental parameters are shown in Table 1.
The step response for joints 1 and joint 2 is shown in Fig. 6 under PD control
without the overshoot. Fig. 7 contains the experimental results of the position
errors.

1.6

1.4

q2

1.2

1
q 1 ,q 2 (rad)

q1
0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
t(s)

Fig. 6. Step response of the PD control

148
 1.5

e2

e1
e1,e2(rad)

0.5

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

t(s)
Fig. 7. Position errors of the PD control

5. Discussion
Through position control a two-degree-of-freedom direct drive robot manipulator is
obtained by using the NPD controllers for the two joints. We can see from the
experimental results of Figs 4 and 5 that the settling times of joint 1 and joint 2 are
ts1 = 0.33 s and ts2 = 0.2 s, respectively, when the time required for the system to
settle 2% of the step input amplitude and the final steady state errors are
[e1, e2]T = [0.003, 0.001]T radians. The steady-state position errors are presented due
to the presence of frictions and disturbances at the joints and the lack of friction and
disturbance compensations in the controllers. It is important to note that despite the
presence of unmodelled friction and disturbance phenomena, these joint position
errors are acceptably small.
Then, through position control a two-degree-of-freedom direct drive robot
manipulator is carried out by using the conventional PD controllers for the two
joints. We can see from the experimental results of Figs 6 and 7, that the settling
times of joint 1 and 2 are ts1 = 0.53 s and ts2 = 0.52 s, respectively, when the time
required for the system to settle 2% of the step input amplitude and the final steady
state errors are [e1, e2]T = [0.0054, 0.0035]T radians.
The above results show that PD controllers are relatively slower in the step
response and have larger final errors than the NPD controllers. It is worth noticing
that the response velocity of the proposed NPD controller is very fast. Therefore,
the proposed NPD controllers can improve the control performance for the position

149
 control problem of robot manipulators because the gains of the NPD controller can
be changed by the nonlinear function of control errors.

6. Conclusion
This paper proposed a NPD controller with gravity compensation for robot
manipulators. The proportional and derivative gains of the NPD controller can vary
as the error varies. The advantage is that the NPD controller has a faster response
and smaller position errors compared to the conventional PD controllers in the
position control of the robot arm. Therefore, the NPD controller is superior to the
conventional PD controller in the position control system and provides a novel
approach for robot control systems. In future, our further work will investigate the
control performance of the NPD controller in various nonlinear control systems.

Acknowledgements: This work was supported by the undergraduate science and technology
innovation project of Zhejiang province, PR China (No 2012R426009).

References
1. T a k e g a k i, M., S. A r i m o t o. A New Feedback Method for Dynamic Control of Manipulator.
– Journal of Dynamic Systems. Measurement and Control, Vol. 102, 1981, 119-125.
2. A r i m o t o, S., F. M i y a z a k i. Stability and Robustness of PD Feedback Control With Gravity
Compensation for Robot Manipulator. – In: F. W. Paul, D. Youcef-Toumi, Eds. Robotics:
Theory and Practice, Vol. 3, 1986, 67-72.
3. H s i a, T. C. Robustness Analysis of a PD Controller with Approximate Gravity Compensation for
Robot Manipulator Control. – Journal of Robotic Systems, Vol. 11, 1994, 517-521.
4. K e l l y, R., R. C a r e l l i. A Class of Nonlinear PD-Type Controllers for Robot Manipulators. –
Journal of Robotic Systems, Vol. 13, 1996, 793-802.
5. S e r a j i, H. A New Class of Nonlinear PID Controllers with Robotic Applications. – Journal of
Robotic Systems, Vol. 15, 1998, 61-81.
6. B u c k l a e w, T. P., C. S. L i u. A New Nonlinear Gain Structure for PD-Type Controllers in
Robotic Applications. – Journal of Robotic Systems, Vol. 16, 1999, 627-649.
7. R e y e s, F., A. R o s a d o. Polynomial Family of PD-Type Controllers for Robot Manipulators. –
Control Engineering Practice, Vol. 13, 2005, 441-450.
8. Y a m a d a, T., T. Y a b u t a. Neural Network Controller Using Autotuning Method for Nonlinear
Function. – IEEE Transactions on Neural Networks, Vol. 3, 1992, 595-601.
9. Y e, J. Analog Compound Orthogonal Neural Network Control of Robotic Manipulators. – In: Y.
L. Wei, K. T. Chong, T. Takahashi et al., Eds. Proceedings of the 3th International
Conference on Mechatronics and Information Technology, Chongqing, China, Vol. 6042,
2005, Part Two, 60422L-1-60422L-6.
10. T h a n h, T. D. C., K. K. A h n. Nonlinear PID Control to Improve the Control Performence of 2
Axes Pneumatic Artificial Muscle Manipulator Using Neural Network. – Mechatronics,
Vol. 16, 2006, 577-587.
11. Y e, J., Y. P. Z h a o. Application of an Analog Compound Orthogonal Neural Network in Robot
Control. – In: W. Shi, S. X. Yang, S. Liang et al., Eds. Proceedings of the International
Conference on Sensing, Computing and Automation, Chongqing, China, 2006, 455-458.
12. Y e, J. Adaptive Control of Nonlinear PID-Based Analog Neural Networks for a Nonholonomic
Mobile Robot. – Neurocomputing, Vol. 71, 2008, 1561-1565.

150

View publication stats