Active Disturbance Rejection Control

Applied to a Low-Cost Educational

Prototype
∗
J. Luis Luna Jessica Maldonado Rubén Garrido
∗
Departamento de Control Automático, CINVESTAV-IPN, Av. IPN
2508 San Pedro Zacatenco, Mexico City, MEXICO (e-mail: garrido,
jluna, jmaldonado @ctrl.cinvestav.mx ).

Abstract: The goal of this work is to implement Active Disturbance Rejection Control
(ADRC) algorithms in a low-cost educational prototype based on a Radio Control (RC)
servo, and to show the viability of this device for testing advanced control laws. An RC
servo exhibits high levels of measurement noise and mechanical friction, and these features
represent a challenge and an opportunity to develop a solid foundation in Automatic Control
theory by applying advanced control algorithms. Experiments in a laboratory prototype show
the performance of an RC servo when applying two widely known ADRC algorithms, the
Disturbance observer (DOB) and the Generalized Proportional Integral (GPI) observer.

Keywords: ADRC, GPI, DOB, control education, servodrive.

1. INTRODUCTION phenomena inside this loop thus affecting performance.

The estimation of disturbances, with the objective of
approximately canceling their effects through a control
Assessment of control algorithms is an important algorithm, has been a subject widely studied in the
issue when teaching Automatic Control, Robotics literature dealing with Active Disturbance Rejection
and Mechatronics. This is usually carried out by Control (ADRC) [Han (2009)].
means of numerical simulations or through real-time Disturbance observers (DOB) perform robust control
experiments using laboratory prototypes. Using software based on the estimation and on-line cancellation
like MatLab-Simulink or web page-based numerical of disturbances, they relies on input and output
simulations [R. Pastor and Dormido (2003)] may be measurements, and on a nominal model of a
cheaper compared with acquiring and putting to work perturbed plant. Subsequently, disturbance rejection
physical prototypes. Nevertheless, limiting the students is accomplished by using the disturbances estimate to
only to these resources may overlook some issues construct an inner feedback controller [Ohishi et al.
like measurement noise, unmodeled dynamics, and (1988); Ohnishi et al. (1996)]. Another method that
disturbances, which appear in real-life applications. addresses the same problem is based on Generalized
Real-time prototypes allow the students a direct Proportional Integral Observers (GPI), which have
interaction with a real system improving their skills in the ability to simultaneously estimate on-line the
Automatic Control theory, Electronics, and programming time-derivatives of the output and a disturbance term.
languages. In this regard, there exist several excellent The latter may depend on unmodeled terms depending on
prototypes based on a DC motor [Apkarian and Astrom the plant state, or on external disturbances [Sira-Ramı́rez
(2004), Gunasekaran and Potluri (2012), Bernstein et al. (2010)].
(2005)] that are adequate to this end. However, their By its very nature, the use of DOB and GPI observers
cost may be too expensive for low-budget universities for controlling an RC servo is very attractive because
when bought in large numbers. the friction effects, which may be difficult to model,
Motivated by the above comments, this work uses may be considered as disturbances. On the other hand,
a low-cost educational prototype based on a radio the presence of large levels of measurement noise would
control (RC) servo. This choice is made based upon its preclude the use of large values of observer gain.
wide availability and low-cost. The original in-house The objective of this work is to report an experimental
electronics of the RC servo is replaced and the prototype study on the application of a GPI observer plus a classic
takes advantage of the mechanical structure of the RC Proportional Derivative (PD) control law for the position
servo including its gear train and feedback potentiometer. control of an RC servo in a low-cost educational platform,
It is worth noting that having a potentiometer as a comparing its performance with the one produced by a
position sensor produce large levels of measurement DOB plus a PD controller.
noise. The RC servo gear train introduces large levels The remaining of the work is as follows. Section 2
of mechanical friction, and its output shaft is directly describes the educational prototype, its components and
coupled to the feedback potentiometer. Hence, closing the key aspects of its design and implementation. Section
the loop through the potentiometer leaves the friction
 PD CONTROLLER SERVODRIVE
d

 e   1 u 1 q 1 q
r Kp

b
  d̂ b s s

b   s
Kd s s

DISTURBANCE OBSERVER

Fig. 1. The educational prototype.

3 is devoted to the mathematical model of the prototype. Fig. 2. PD+DOB controller block diagram.
Sections 4 and 5 describe respectively a DOB and a GPI Model (1) has the next alternative writing
observer applied to the educational prototype. In Section
6 shows the experimental results. The work ends with q̈(t) = −f (q̇) + bu + d¯ (2)
some concluding remarks. where b = k/J, f (q̇) = F (q̇)/J and d¯ = d¯m /J. If the
friction torque are assumed unknown, then it is grouped
2. EDUCATIONAL PROTOTYPE with the unknown term d in (2)
q̈(t) = bu + d (3)
2.1 Description ¯
with d = d − f (q̇).

The low-cost educational prototype is built around

3. DOB APPLIED TO A SERVODRIVE
a HITEC RC servo model HS-485HB. Its electronics
including the control card are taken out and replaced by
a low-cost external linear amplifier built around a power Disturbance observers are based on a linear model of
operational lineal amplifier Texas Instruments LM675. A a plant, and are used to reject internal and external
power supply consisting of standard voltage regulators disturbances. The unknown dynamics and disturbances
LM7805 and LM7905, a diode-based rectifier, capacitors, are actively estimated and compensated in real-time,
and an AC transformer supplying 12 V and 0.5 A, feeds which makes the closed-loop system robust and less
the amplifier, and its input range is ±4.8V . Fig. 1 shows dependent on a detailed mathematical model [Gao
the educational prototype. (2006)].
The potentiometer works as a voltage divider and its Consider the next model of a plant under the effect of a
output is directly connected to a data acquisition. The disturbance
cost of the educational prototype including the power y = P (s)(u + d) (4)
supply is less than $50 US. This price may be dropped
by buying the electronics parts in large quantities and by The basic idea behind the DOB is to use measurements
using cheaper RC servos. The low-price and portability of the input and output of the plant to reconstruct the
of the prototype make possible to use it at the classroom disturbance. The above could be achieved by computing
or for private learning. d as
d = P (s)−1 y − u (5)
2.2 Mathematical model Where P (s) is the plant model, which must be of
minimum phase, y is the measurable output of the plant,
Consider a servodrive composed of a DC motor, an and u is the control signal.
angular position sensor, and a power amplifier. It is In order to be able to compute an estimate of the
assumed that the DC motor armature inductance has a disturbance, all that is required is to know the order of
very low value compared with the value of its armature the system and the value of the input gain.
resistance. The above assumption allows considering A problem with this estimation scheme is that using the
a proportional relationship between the input voltage inverse plant model P (s)−1 would require measurements
to the power amplifier and the electromagnetic torque of the time derivatives of the output y, which may not be
generated by the DC motor. A model of this system is available in practice. In order to avoid the above problem,
disturbance estimation is performed as follows
J q̈(t) + F (q̇) = ku + d¯m (1)
dˆ = P (s)−1 y − u F (s)
 
Variables q, q̇ and q̈ are respectively the angular position, (6)
velocity and acceleration of the servodrive, u is the The transfer function F (s) is a strictly proper stable
control input voltage, J the servomotor and load lumped filter. If the filter order is equal or higher than the
inertia, f (q̇) is a nonlinear friction torque that may order of the characteristic polynomial of P (s), then this
include viscous and Coulomb friction torques, k is a gain condition guarantees a proper or strictly proper transfer
depending on the power amplifier gain and on the DC function P (s)−1 F (s). The design of F (s) defines several
motor torque constant, and the term d¯m is an external characteristics of DOB-based controllers [Choi et al.
disturbance. (2003).
Fig. 2 depicts the block diagram of a Proportional
Derivative controller plus a Disturbance Observer PD CONTROLLER SERVODRIVE
d
(PD+DOB) applied to a servodrive. The term e =

r − q stands for the position error and r is a desired  e   1 u 1 q 1 q
r Kp b
angular position. Note that the disturbance estimate dˆ    d̂ b s s
is injected to the input to counteract the effects of the
real disturbance d. GPI
The DOB filter is defined as Kd OBSERVER
β x3e x2e x1e
F (s) = (7)
s+β
with cutoff frequency β > 0. Since mechanical systems
with large inertias are not fast, then the bandwidth β
may be set higher than the bandwidth of the mechanical
system [Choi et al. (2003)]. Fig. 3. PD+GPI controller block diagram.
The proposed PD+DOB controller relies on angular
velocity measurements and assumes knowledge of the 1
servodrive input gain b. This control law has the following [Kp (r − x1 ) − Kd x2e − x3e ]
uGP I = (15)
b
mathematical expression
Fig. 3 depicts the block diagram of a Proportional
1h i
uDOB = Kp (r − q) − Kd q̇ − dˆ (8) with Derivative action controller plus a Generalized
b Proportional Integral observer (PD+GPI) applied to a
servodrive.
4. GPI OBSERVER APPLIED TO SERVODRIVE
Before ending this section it is worth pointing out that
the PD+DOB and PD+GPI controllers share the same
In the GPI observer applied to Active Disturbance structure. The nominal control law corresponds to a
Rejection Control (ADRC), the state of the plant model PD controller, and the disturbance estimates produced
is extended by defining the disturbance term and its by GPI observer and the DOB seek to counteract
time-derivatives as new state variables. The on-line the effect of the lumped disturbance d. On the other
estimation of this new state is performed using the hand, the DOB only produces a disturbance estimate
GPI observer, which is also called an extended State whereas the GPI observer is able to generate velocity and
Observer (ESO) [Parvathy and Daniel (2013)]. To apply disturbance estimates. It could be an advantage for the
the GPI observer the model of the servodrive (1) must be GPI observer since only one parameter is needed to tune
expressed in its state variable form by choosing the observer, and the PD+DOB controller would need
x1 = q Position a velocity estimator. However, the fact that high values
(9)
x2 = q̇ Velocity of θ could lead to peaking phenomena and measurement
The above definitions allows obtaining the next state noise amplification in the GPI observer would balance
space expresion for the servodrive model the advantages of the PD+GIP controller compared with
ẋ1 = q̇ = x2 the PD+DOB. Moreover, the velocity estimator needed
(10) in the DOB is tuned without regard on the tuning of the
ẋ2 = q̈ = bu + d
DOB, thus reducing the possibility of measurement noise
The dynamics (10) are rewritten considering d as a new amplification.
state, and assuming that its time deriavtive exists
ẋ1 = x2
ẋ2 = bu + x3 (11) 5. EXPERIMENTS RESULTS
ẋ3 = ϕ(t)
with x3 = d y ϕ(t) = d. ˙ It is assumed that x3 and ϕ(t) 5.1 Experimental setup
are bounded
|d| < ∞ Fig. 4 depicts the setup used in the experiments. The
(12)
|ϕ(t)| < ∞ experimental platform is composed of a personal
Thus, the GPI observer takes the form: computer endowed with a Sensoray model 626
ẋ1e = x2e + k1 (x1 − x1e ) data acquisition card. It has analog-to-digital and
ẋ2e = b̂u + x3e + k2 (x1 − x1e ) (13) digital-to-analog converters, and an output voltage range
ẋ3e = k3 (x1 − x1e ) of ±10 V. The programming and real-time execution
Where x1e , x2e and x3e are the estimates of x1 , x2 of the control algorithms applied to the educational
and x3 , respectively. With the help of the GPI observer prototype are performed using MatLab/Simulink from
characteristic polynomial the gains k1 , k2 and k3 are MathWorks and QUARC from Quanser Consulting.
computed The sampling period is set to 1 ms and the numerical
s3 + 3θs2 + 3θ2 s + θ3 (14) integration algorithm is the Euler01 method. The gain
in model (3) is b = 253.329. In the case of the PD+DOB
Therefore, k1 = 3θ, k2 = 3θ2 , k3 = θ3 . controller the DC motor angular velocity q̇ is estimated
A Proportional Derivative controller plus a GPI from position measurements through the next filter
observer-based disturbance rejection control (PD+GPI) 160s 160
is described by Gv (s) = (16)
s + 160 s + 160
 Control
Computer Reference

Position (Degrees °)
80 PD+DOB 5
PD+DOB 10
PD+DOB 15
60

40

20
Connection
panel for the 0
data acquisition
card 10 11 12 13 14 15 16 17 18
Time (s)

Position
Fig. 5. Responses for the PD+DOB controller.
feedback

Power
DC Motor Position
amplifier
sensor
80
PD+DOB 5

Position Error (Degrees °)

60 PD+DOB 10
PD+DOB 15
Control 40
Signal ± 10 V
20

-20

-40

Fig. 4. Experimental setup. -60

-80
10 11 12 13 14 15 16 17 18
Time(s)

Fig. 6. Error signals for the PD+DOB controller.

5.2 Experiments with the PD+DOB controller

The following experiments show the performance of the

PD+DOB using the control law (8). 0.8
The reference is a filtered step of 90◦ . The performance PD+DOB
PD+DOB
5
10
0.6
is measured using the Integral Squared Error (ISE). In PD+DOB 15
Control Signal (V)

addition to the ISE index (17), performance is assessed 0.4

through the Integral of the Absolute value of the Control 0.2

(IAC) and the Integral of the Absolute value of the
0
Control Variation (IACV ) indexes, all defined as
-0.2
Z T -0.4
2
ISE = 100 [e(t)] dt (17)
0 -0.6
10 11 12 13 14 15 16 17 18
Time (s)

Fig. 7. Control signals for the PD+DOB controller.

Z T
IAC = |u(t)| dt (18)
0

5.3 Experiments with the PD+GPI controller

Z T  
 du(t) 
IACV = The following experiments show the performance of the
 dt  dt (19)
 
0 PD+GPI using the control law (15). Fig. 8 displays the
angular position of the DC motor using different values
which are evaluated at T = 2s. Fig. 5 shows the of θ but maintaining the same values of the proportional
response of the system using the PD+DOB controller and derivative gains Kp and Kd as in the case of the
with different values of β, in Fig. 6 is the error signal PD+DOB controller. The error and control signal are
and the control signal is shown in Fig. 7. shown in Fig. 9 and Fig. 10 respectively.
 Reference Reference
Position (Degrees °)

Position (Degrees °)
80 PD+GPI =33 80 PD+DOB
PD+GPI =66 PD+GPI
PD+GPI =100
60 60

40 40

20 20

0 0

10 11 12 13 14 15 16 17 18 10 11 12 13 14 15 16 17 18
Time (s) Time (s)

Fig. 8. Responses for the PD+GPI controller. Fig. 11. Responses for PD+GPI and PD+DOB
controllers.
Controller +ess −ess Gain ISE IACV IAC
80
PD+GPI =33 1.11 0.82 β =5 8379 3.943 0.15
Position Error (Degrees °)

60 PD+GPI =66
PD+DOB 0.11 0.09 β =10 7777 4.162 0.1514
PD+GPI =100
40 0.07 0.06 β =15 7546 5.063 0.1609
20 1.49 0.83 θ =33 8580 1.155 0.1491
PD+GPI 0.12 0.11 θ =66 7792 3.497 0.1489
0
0.07 0.06 θ =100 7555 8.433 0.1725
-20
Table 1. Experimental results for the
-40 PD+DOB and PD+GPI controllers using
-60 Kp = 124.1312 and Kd = 31.2101.
-80
10 11 12 13 14 15 16 17 18
Time(s)

Reference
Fig. 9. Error signals for the PD+GPI controller.
Position (Degrees °)

80 PD+DOB
PD+GPI

60

1
PD+GPI =33 40
PD+GPI =66
PD+GPI =100
Control Signal (V)

0.5 20

0
0
10 11 12 13 14 15 16 17 18
Time (s)
-0.5

Fig. 12. Responses for PD+GPI and PD+DOB

-1 controllers using b = 180.
10 11 12 13 14 15 16 17 18
Time (s)
the nominal gain b = 253.329. Fig. 12, 13 and 14 depicts
Fig. 10. Control signals for the PD+GPI controller. the response, position error and control signals for these
controllers. Note that the PD+GPI controller shows a
small oscillation in its response that appears in Fig. 12
5.4 Comparison between PD+GPI and PD+DOB and 13. This behavior is also observed in the control
controllers signal in Fig. 14 that shows a high level of chattering for
the PD+GPI controller. These results show that the GPI
Fig. 11 shows the response of both controllers using observer amplifies the measurement noise more that the
β = 15 and θ = 100. The behavior is essentially the same. DOB. Table 5.4 depicts the corresponding performance
On the other hand, Table 5.4 displays the performance measures for both controllers. The chatter observed in
measures for both controllers. Note that there is almost the control signal of the PD+GPI controller is reflected
no difference between the performance in terms of the in a high value of the IAV C index of 30.24. The the
ISE and IAC indexes, and the steady state error at PD+DOB controller produced a value of 6.1415, which
90◦ (+ess ) and at 0◦ (−ess ) is similar. On the other is slightly higher than the value obtained when using the
hand, the PD+GPI controller exhibits a higher IAV C nominal value of the input gain of 5.63 . From the above
index value for θ = 100 compared with the corresponding it is clear that the PD+GPI controller exhibits more
value for the PD+DOB controller when using β = 15. sensitivity to uncertainty in the input gain compared
In order the assess the performance of the PD+DOB with the PD+DOB controller. Therefore, users of this
and PD+GPI controllers under uncertainty in the input algorithm must ensure good estimates of the input gain b
gain b, a set of experiments are performed computing the in order to obtain good performance of both controllers,
controllers with b = 180, which corresponds to 71% of in particular when using a GPI observer.
 control signal chattering.
80
PD+DOB Future work includes implementing the algorithms
Position Error (Degrees °)

60 PD+GPI
studied here in a low-cost processors like the Arduino
40 using free software. Moreover, in order to improve
20 portability new approaches for building the prototype
0 will be considered.
-20
ACKNOWLEDGEMENTS
-40

-60 The authors would like to thank the support of

-80 Gerardo Castro Zavala and Jesús Meza Serrano in
10 11 12 13 14 15 16 17 18
setting up the laboratory testbed. The support by
Time(s)
CONACyT-MEXICO and CINVESTAV to the first and
Fig. 13. Error signal for PD+GPI and PD+DOB second authors through a Ph.D. scholarship is also
controllers using b = 180. recognized.
REFERENCES
1
PD+DOB Apkarian, J. and Astrom, K. (2004). A laptop servo for
PD+GPI
control education. IEEE control systems, 24, 70–73.
Control Signal (V)

0.5
Bernstein, D.S. (2005). The quanser dc motor control
0 trainer. IEEE Control Systems Magazine, 3, 90–93.
Choi, Y., Yang, K., Chung, W.K., Kim, H.R., and Suh,
-0.5 I.H. (2003). On the robustness and performance of
disturbance observers for second-order systems. IEEE
-1 Transactions on Automatic Control, 48(2), 315–320.
Gao, Z. (2006). Active disturbance rejection control: a
-1.5
10 11 12 13 14 15 16 17 18 paradigm shift in feedback control system design. In
Time (s) American Control Conference, 2006, 7–pp. IEEE.
Gunasekaran, M. and Potluri, R. (2012). Low-cost
Fig. 14. Control signals for PD+GPI and PD+DOB undergraduate control systems experiments using
controllers using b = 180. microcontroller-based control of a dc motor. IEEE
Transactions on Education, 55, 508–516.
Controller +ess −ess Gain ISE IACV IAC
Gutiérrez-Giles, A., Arteaga-Pérez, M.A., and
PD+DOB 1.11 0.82 β =15 7763 6.415 0.1553
PD+GPI 1.49 0.83 θ = 100 7828 30.24 0.3431
Sira-Ramı́rez, H. (2016). Control de fuerza de robots
manipuladores basado en observadores proporcionales
Table 2. Experimental Results for the
integrales generalizados. Revista Iberoamericana de
PD+DOB and PD+GPI controllers using
Automática e Informática Industrial RIAI.
Kp = 124.1312, Kd = 31.2101 and b = 180.
Han, J. (2009). From pid to active disturbance rejection
control. IEEE transactions on Industrial Electronics,
56(3), 900–906.
6. CONCLUSIONS Ohishi, K., Ohnishi, K., and Miyachi, K. (1988). Adaptive
dc servo drive control taking force disturbance
This work describes the application of advanced suppression into account. IEEE Transactions on
robust control techniques on a low-cost laboratory Industry Applications, 24(1), 171–176.
platform based on a Radio Control servo that uses Ohnishi, K., Shibata, M., and Murakami, T.
potentiometer feedback. The controllers under test (1996). Motion control for advanced mechatronics.
are the Proportional Derivative plus a Disturbance IEEE/ASME Transactions On Mechatronics.
Observer (PD+DOB) controller and the Proportional Parvathy, R. and Daniel, A.E. (2013). A survey
Derivative plus Generalized Proportional Integral on active disturbance rejection control. In
observer (PD+GPI) controller. Automation, Computing, Communication, Control
The main advantage on using a GPI observer is that its and Compressed Sensing (iMac4s), 2013 International
tuning is performed using only one parameter. Another Multi-Conference on, 330–335. IEEE.
advantage is the fact that it simultaneously produce R. Pastor, J.S. and Dormido, S. (2003). Xml-based
velocity and disturbance estimates. On the other hand, framework for the development of web-based
a DOB only generates disturbance estimates. The laboratories focused on control systems education. Int.
experimental outcomes shows that both controllers J. Engng, 19, 445–454.
produce smooth responses without overshoot and display Sira-Ramı́rez, H., Ramı́rez-Neria, M., and
similar performance in terms of the Integral Squared Rodrı́guez-Angeles, A. (2010). On the linear control of
Error (ISE), the Integral of the Absolute value of the nonlinear mechanical systems. In Decision and Control
Control (IAC) and the Integral of the Absolute value (CDC), 2010 49th IEEE Conference on. IEEE.
of the Control Variation (IACV ) indexes. However, White, M.T., Tomizuka, M., and Smith, C. (2000).
the PD+GPI controller exhibits more sensitivity to Improved track following in magnetic disk drives using
uncertainty in the input gain which translates into a disturbance observer. IEEE/ASME Transactions on
measurement noise amplification and high levels of mechatronics, 5(1), 3–11.