CC BY
6
(M
CO
o >
Systematic design of ADRC-based unmanned tracked vehicle trajectory tracking with FPGA-in-the-loop validation
o
> Momir R. Stankovica, Rafal Madonskib
CM o CM
(£
Stojadin M. Manojlovicc
a University of Defence in Belgrade, Military Academy, lju Belgrade, Republic of Serbia,
e-mail: [email protected], corresponding author, O ORCID iD: ©https://orcid.org/0000-0001-8371-9341
b Silesian University of Technology, o Faculty of Automatic Control, Electronics and Computer Science,
Gliwice, Republic of Poland, o e-mail: [email protected],
^ ORCID iD: ©https://orcid.org/0000-0002-1798-0717
> c University of Defence in Belgrade, Military Academy,
;< Belgrade, Republic of Serbia,
e-mail: [email protected], ORCID iD: ©https://orcid.org/0000-0003-1268-5310
doi https://doi.org/10.5937/vojtehg72-49983
FIELD: automatic control, control engineering, mobile robotics ARTICLE TYPE: original scientific paper
Abstract:
w Introduction/purpose: The trajectory tracking control problem in an un-
o manned tracked vehicle (UTV) represents a challenging task, due to un-
known and unmeasurable slippage dynamics which inevitably exists during movement. Therefore, the application of standard industrial control schemes is often limited.
Methods: In this paper, an active disturbance rejection control (ADRC) scheme is proposed for the longitudinal (vehicle longitudinal velocity control) and lateral (vehicle course angle control) control channels of the UTV to collectively handle all the plant modeling uncertainties and acting slippage disturbances.
Results: A step-by-step procedure for applying the ADRC algorithm for the specific case of UTV trajectory tracking is presented. It includes systematic design, discretization as well as performance analysis and validation utilizing FPGA-in-the-loop (FIL) simulations.
ACKNOWLEDGMENT: The authors are grateful for the financial support from the University of Defence in Belgrade, Republic of Serbia, project code: VA-TT/1/21-23.
Introduction
tt
o o
Conclusions: The proposed FIL-based validation method reduces the gap between pure simulation design (which may be too idealized) and implementation on the real vehicle (which maybe time-consuming). The obtained experimental results show the advantages of the proposed control structure over standard industrial PI/PID controllers in different work- £ ing conditions.
Key words: unmanned tracked vehicles, trajectory tracking, active dis- > turbance rejection control, velocity control, PID controller, FPGA-in-the- o loop, hardware validation.
< CD
CL
In recent years, unmanned tracked vehicles (UTVs) have received considerable attention in both civilian and military applications, mostly due to their high mobility and performance in different terrain (Nonami etal., 2013).
The control algorithms for UTVs often take into account the dynamic be- jo havior and construction limits of the vehicle to ensure high movement performance along with sufficient robustness against external disturbances. UTVs are commonly designed without a steering mechanism and the direction of motion is changed by adjusting the angular velocities of the right and left track drive wheels. This provides a fast vehicle response to control signals with a minimum turning radius. However, complex dynamics of the interaction between the tracks and ground causes significant slippage during maneuvering (Janarthanan et al., 2012), leading to the UTV's motion § being largely dependent on the terrain properties (Wong, 2022). The UTV E slip models based on the tracks instantaneous centers of rotation were analyzed, for example, in (Pentzer et al., 2014) and (Lu et al., 2016), where jo it was shown that, besides the terrain characteristics, the important factors o which significantly affect slippage dynamics are path curvature and the ab- ^ solute value of the UTV lateral acceleration.
The most common tasks for unmanned vehicles are point-to-point motion, path following, and trajectory tracking (De Luca et al., 1998). The first is a stabilization problem, where the vehicle, starting from a given initial point, has to reach the desired point in space, regardless of the motion profile. Path following and trajectory tracking require motion along the defined geometric path/trajectory, where the trajectory is additionally constrained by time.
To ensure the realization of appropriate motion tasks for UTVs, different control strategies can be used. The usual choice is the time-tested PI/PID
ro E <u
CO >
"ro
<u
1701
>
o
"J controllers. In (Hu et al., 2019), for example, a torque-speed control strategy was proposed based on low-complexity PID controllers. The authors of (Zou et al., 2018) suggested relying on precise vehicle kinematics and dynamics models and applying control structures with back-stepping PID. Finally, in (Huang et al., 2018b), look-ahead lateral errors were used to design a PI-based path-tracking control system. The main advantages of using the PID-family controllers are their simple design, intuitive procedures for parameter settings, and straightforward implementation. However, their applications to UTVs are often limited in practice by the level of modeling u uncertainties and the influence of large, unmodeled slippage dynamics.
Therefore, the solutions to the problems of UTV path following and trajec-x tory tracking on uneven terrains often have to rely in real life on using more u sophisticated control techniques.
Analyzing the literature on advanced control for UTVs, one can notice three main types of governing algorithms, namely intelligent control, so-called modern control, and observer-based control, with some notable, representative works recalled below.
• The first type is usually based on fuzzy logic and neural networks (Li ^ et al., 2020; Dai et al., 2018) or their combination with classical con-CD trollers, like those seen in (Huang et al., 2018a; Al-Jarrah et al., 2019). g Although these advanced control structures can provide high trajectory tracking performance even in challenging terrains, their hardware
w implementation is often limited due to algorithmic complexity and re-
0 sultant increased computational requirements.
• The so-called modern control approaches are mostly based on model predictive control (Mitsuhashi et al., 2019; Tao et al., 2021; Burke, 2012), adaptive control (Tang et al., 2021; Gonzalez et al., 2010; Hi-ramatsu et al., 2019), nonlinear control (Hong et al., 2009), or sliding mode control (Sabiha et al., 2022). Those methods, even though effective, rely either on precise modeling of UTV motion and slippage disturbance (Hong et al., 2009; Mitsuhashi et al., 2019) or utilizing a dedicated, separate function block for model parameters identification and/or slippage dynamics estimation.
• The third type of advanced control scheme for UTVs is observed-based control. In this category, an example extended Kalman filter was applied in (Tao et al., 2021) to estimate the coefficients of road resistance, thus strengthening the model predictive control's adapt-
ability to uncertain road conditions. Another example can be (Burke, cn 2012), where the compensation of slip disturbance was addressed "" using an adaptive least squares slippage estimator. In (Hiramatsu ^ et al., 2019), a structure composed of an adaptive controller and dis- d. turbance observer was proposed for UTV angular and linear velocity estimation, where the observer was responsible for estimating the effects of slippage which in return improved the tracking performance §■ without making the model unbearably complex. Having analyzed the literature on UTV control, it seems that the current method landscape is missing an approach that would be a compro- ^ mise between effectiveness, overall complexity as well as knowledge about 2 the controlled system and the environment affecting it. Therefore, in this work, we propose the use of an active disturbance rejection control (ADRC) scheme (Gao, 2006), which is a general methodology known to provide robust performance, straightforward tuning and implementation, and operation not being overly dependent on the plant's mathematical model. The ADRC's recent successful industrial applications (discussed in the overview paper (Zhang et al., 2021)) are the main motivation for using it in this work for the UTV trajectory tracking control system. In fact, this is not the first time ADRC has been considered for UTVs. In (Chen et al., 2019), the authors have introduced back-stepping ADRC as their lateral control for the path-following problem, which featured an extended state observer (ESO) for jointly estimating internal model uncertainties and slip- | page disturbances.
The contribution of our work is the extensions provided to (Chen et al., 2019). Firstly, our research deals with the UTV trajectory tracking task, where the trajectory is additionally constrained with time. Secondly, this £ paper extends the scope to designing both controllers: lateral and longi- < tudinal. Finally in contrast to (Chen et al., 2019), we go beyond just a simulation study and perform FPGA-in-the-loop (FIL) verification. Such a methodology reduces the gap between a pure simulation design (which may be too idealized) and implementation on the real vehicle (which may be resource-consuming). With this paper, we want to show a comprehensive application of ADRC to UTVs (with all its unique characteristics and intricacies), including systematic design, performance analysis, and validation.
k c a
le icl
hi
e v d e k c ra tr
d e
a m e t
s y
S
1703
v o k n a t
S
(M
O >
O
The paper is organized as follows. The UTV model is briefly recalled in the UTV motion model section. Then, the design of lateral and longitudinal controllers is shown in the Lateral controller design and Longitudinal controller design sections. The verification of the proposed control system is described in the Control system verification. The section entitled Conclu-
§ sion summarizes the work.
CM
yy 0£
8 The kinematic model of the considered UTV can be described as:
UTV motion model
o /X(t)\ /cos d(t) 0
I 0 1
sin m 0, ■ , (1)
^ where x(t) and y(t) denote vehicle coordinates in the inertial coordinate system, 9(t) is the course angle of UTV, v(t) and u(t) are the UTV longitudinal and angular velocity, respectively, both considered to be the UTV control inputs. The uncertainties in the longitudinal and angular velocity, caused by unknown track friction, i.e. track slippage, are represented with cd vd(t) and wd(t), respectively. g Including the dynamic model of the UTV, the relation between the control
inputs v(t) and w(t) and the angular velocities of the right and left track drive
w wheel QR(t) and QL(t), are given respectively as:
o
r
v(t) = 2(fiR(t)+fiL(t)) , (2)
^ u>(t) = B (nR(t) - ^L(t)), (3)
where r is the drive wheel radius and B is the normal distance between the right and left track.
In the presence of the track slippage, (2) and (3) can be modified respectively as:
r( )
v(t) + vd(t) = ^ (^R(t) • aR(t) + fiL(t) • aL(t^ , (4)
u(t) + Ud(t) = B (QR(t) • aR(t) - ^L(t) • aL(t)) , (5)
where aR(t) and aL(t) are the unknown friction coefficients of the right and left track, which are in the range [0,1].
One can notice that the UTV motion control involves designing two con- cn trol subsystems. The first one is for angular velocity control and involves "" a so-called lateral controller. The second subsystem is for longitudinal velocity control and deals with constructing a longitudinal controller. ¿.
Lateral controller design
To define the lateral control model, let us consider the UTV reference trajectory tracking problem shown in Figure 1, where the reference trajectory is defined by the kinematic model of a virtual target:
(xr(t)\ = (cos dr(t)\ v
\Ut))-\sin9r(t)J • Vr- (6)
where xr(t) and yr(t) are the coordinates of the virtual target in the inertial coordinate system (X,Y), vr > 0 is the virtual target longitudinal velocity and 9r(t) is the virtual target course angle. The path along track error es(t) and the cross-track error ed(t) are defined in the vehicle coordinate system (Xv, Yv) as:
cos 0(t) sin 0(t)\ fxr(t) - x(tf
— sin e(t) cos e(t\
By denoting the course angle error as:
fes(t)\fcos e(t) sin e(t)\ fxr(t) - x(t)\
\ed(t)J V- sin e(t) cos e(t)J '\yr(t) - y(t)J ■ ()
and substituting (1) and (6) in the derivative of (7), one can obtain trajectory tracking and course angle errors dynamics as:
ed(t) = v>r sin ee(t) + di(t), (10)
ee(t) = u(t)+d2(t), (ii)
ro
>
CP
o o
< CD
CL
o ro
JD o
<u >
"D
<U
o ro
"D
<U
CO
ee(t) = e(t) - er(t), (8) i=
"D <U CO <0 .Q i
O CC
Q <
es(t) = e(t) • ed(t) - v(t) - vd(t) + Vr cos ee(t), ed(t) = Vr sin ee(t) - e(t) • es(t), (9)
ee(t) = u(t)+ud(t) - er (t). i
As in practice, the cross-track error is the main concern (Chen et al., 2019). To facilitate that, the lateral control subsystem can be formulated from (9) m
as: ¿o
"O
o +-'
ro E <u
ro <u
>
o
<D
(M
O >
CM o CM
0£ yy
0£ ZD O o
-J
<
o
X
o
LU
I—
>-
CC <
CD
S2 ■O
x
LU I—
o
o
y a
Reference k trajectory
Figure 1 - Considered UTV trajectory tracking problem
with d\(t) = 0(t) ■ es(t) and d2(t) = wd(t) + 0r(t), which represents collectively the disturbances caused by slippage, reference trajectory dynamics, and longitudinal channel coupling (reflected in the influence of the v(t) on the lateral controller).
By substituting (11) in the derivative of (10), the cross-track error dynamics can be presented in a compact ADRC form as:
ea(t) = bo ■ u(t) + fi(t),
(12)
where b0 = vr is the known part of the lateral model and fi(t) = vr(cos 9e(t)-1) ■ w(t) +vr cos 9e(t) ■ d2(t) + di(t) is denoted as the total disturbance of the lateral control subsystem. One can see that the lateral control is formulated as a regulation problem, where the error signal ed(t) has to be minimized by manipulating the control input w(t), all in the presence of perturbing effects like slippage disturbances, coupling dynamics with the longitudinal channel, and the reference trajectory dynamics.
To design the ADRC controller, model (12) is expressed as a following extended state-space model:
[edm ¡0 1 0\ fed(t)\ (0\ /ON
I ed(t) I = I 0 0 1 I • I ed(t) I + I bo I • u(t) + I 0 \fi(t)J \0 0 0) \Mt)J \0j \1,
fi(t), (13)
and the states can be estimated by an appropriate observer, for example, a linear ESO shown in (takomy et al., 2020):
'¿d(t)\ (0-e-d(t) I + I bo | • u(t) Ji(t)J \0
(h(t)\ 01 0
Ut) = 10 0 1
Kh(t)) 00 0
fiiA
+ I ll2
li3
(14)
where ¡n, ll2, and ll3 are the observer gains, which can be relatively simply tuned by the bandwidth parameterization from (Gao, 2003) by placing observer poles at the same location A = -wESOl :
(A + WESÜ!)3 = A3 + 111 • A2 + l\2 • A + lis,
(15)
where wESOl is the desired observer bandwidth.
By assuming ee(t) e {-n,n}, the saturated lateral control law can be proposed as:
{Wmax • sat (U\(t)/W max ) , when \ee(t)\ < §
Wmax, when § < ee(t) <n (16)
-Wmax, when - n < ee(t) < - §
where sat(a) = min(1, \a\) • sign(a) is a saturation function, wmax > 0 is saturation level of the lateral ADRC signal, denoted as:
ui(t) = 1 —pi • ed(t) - kdi • ed(t) - h(t)) , (17)
with the controller design parameters kpl and kdl. The saturation wmax is introduced to reduce slippage caused by the large centrifugal force, i.e. the high value of the UTV angular acceleration, as discussed in (Lu et al.,
LT) <M
7
o o
£± CP
ro
>
CP
o o
< CD
CL
o ro
JD o
<u >
"D
<U
o
ro +-'
"D
<U
CO
E
"D <U CO <0 .Q i
O CC
Q <
M—
O c OT <o <u
"D
O
ro E <u
CO >
"rö
<u
>
o
CM
o >
CM o CM
0£ LU
X LU I—
o o
2016). It should be also noted that control law (16) is designed to enable quick reduction of the course angle error in the range 9e(t) e {—n/2,n/2} and to avoid the control signal (17) saturation.
Now, by assuming fl(t) w f(t), ed(t) w ed(t), ed(t) w ed(t) and substituting (16) in (12), the following relation can take place:
ed(t) + foi ■ eed(t) + kpi ■ ed(t) w 0. (18)
^ From (18) one can see that the lateral control performances can be ad-g justed by appropriately selecting the parameters kpi and kdi. Applying the ^ already mentioned bandwidth parameterization, the controller parameters can be tuned by placing closed-loop controller poles at the common location A = —wCLl :
(A + WCL!)2 = A2 + kdi ■ A + kpi, (19)
<c where wCLl is the desired closed-loop lateral control system bandwidth.
Longitudinal controller design
Let us denote the UTV longitudinal velocity in the presence of slippage cd disturbance as vv(t) = v(t) + vd(t). Thus, the longitudinal velocity control g model can be formulated as:
v v(t) = Uv (t)+ v d(t), (20)
where uv(t) = V(t) represents the control signal which should be designed to make vv(t) track the reference target longitudinal velocity vr even in the presence of the disturbance vd(t). To apply the ADRC concept to the considered problem, system (20) can be rewritten in the extended state-space form as:
ft) (0 0)«+(0H>+(0) fv(2D
where fv(t) = vd(t) is the longitudinal channel total disturbance, which can be estimated by an appropriately constructed ESO:
ftS=(0i)f) + (0) uv « + (tl) ^ — ev («».»
<D
(M
O >
CM o CM
0£ yy
0£ ZD O o
-J
<
o
X
o
LU
I—
>-
CC <
CD
S2 ■O
x
LU I—
o
o
Control system verification
Since the hardware implementation of the control algorithm requires its discrete form, simulation experiments are carried out with a discrete-time version of the proposed UTV controllers. Therefore, the discretization of the lateral and longitudinal controllers is presented first, followed by simulation results.
Simulation validation
Since the lateral and the longitudinal ADRC control laws (17) and (23) are in linear forms, only the observer need to be discretized. To enhance the closed-loop system stability for a low sampling rate Ts, the current discrete models from (Miklosovic et al., 2006; Herbst & Madonski, 2023) are used. Therefore, for the lateral ESO (14), its discrete form is defined as:
^d (k)\ ed (k)
V/i(fc)/
+ (Bdi - Lci • Cdi • Bdi) • w(k - 1) + Lci • ed(k),
'¿d(fc - 1)N
= (4ii - Lci • Cdi • An) • I ¿d(k -1)
Ji(k - 1),
(27)
where:
Adi =
'1 Ts T2/2\ /bo •T-2/2N
0 1 Ts I ,Bdi = I bo •Ts | ,Cdi =(10 0)
,0 0 1 V 0
are the discrete observer matrices obtained by applying the zero-order hold (ZOH) discretization method from (Miklosovic et al., 2006). The discrete ESO gains vector Lcl is computed by placing the discrete observer poles in one location /3 = e-WES°Ts ,i.e.:
\zI - (Adi - Adr Lci- Cdi)| = (z - /3 )3
In the same manner, the discrete-time model of the longitudinal controller ESO (22) is obtained as:
nv(k)\ \hk))
= Adv - Lcv Cdv
Adv)(;V(k -1))
1710
+ B dv Lcv • Cdv • Bdv) • uv(k 1) + Lcv ' vv(k), (28) | ¡^
where:
^dv =d T{) ,Bdv = (b0 0,Cdv = (1 0)
\zI - (Adv - Adv • Lcv • Cdv)\ = (z - f)2.
The simulation experiments are conducted in MATLAB/Simulink using the model of the laboratory skid-steer UTV, with dimensions r = 0.3 m and B = 0.7 m. To demonstrate the effectiveness of the proposed ADRC algorithm, its performance is compared with the UTV control architecture realized by the standard industrial discrete PID controller with a noise filter in the lateral control subsystem and the PI controller in the longitudinal control subsystem. To make the test more realistic, the maximal UTV angular velocity is limited to wmax = 2 • n rad/s. The reference trajectory is defined using the kinematic model (6), with vr = 2 m/s and:
er(k • ts) = {
0.15 • k • Ts, when 22 s <k • Ts < 35 s. The simulations are realized through the following two cases.
1711
7
o o
v0 1) '"uv ^ 0 ) '~uv "> ' S
are the matrices determined by the ZOH discretization. The longitudinal ^
discrete ESO gains vector Lcv is computed by placing discrete observer a.
poles in the location f = e-UES°vTs, i.e.: -
< CD
CL
o ro
o
<u >
"D
CD
o
^0.05 • k • T,, when 0 s <k • T, < 8 s; ™
"D
<u
s, VVII^II \J J \ rv -L s <
-0.1 • k • Ts, when 8 s <k • Ts < 16 s; (29)
(29) ro
-0.05 • k • Ts, when 16 s <k • Ts < 22 s; E
s 35 o. "O
cd
co
ro
.Q
O Q1 Q
Simulation case #1 h
In this case, it is assumed that the UTV is not affected by the slippage disturbances (aR = 1 and aL = 1). Both of the considered control approaches (the proposed ADRC and the standard PI/PID) were tuned empirically with the goal to enable similar control performances in terms of minimization of the cross-track error ed. Therefore, in the lateral PID controller, four parameters: proportional, integral, and derivative gains, as well as the noise filter coefficient, are set as kPl = 4.5, kil = 1, kDl = 0.5 and N = 85, and the proportional and integral gains of the longitudinal PI controller are set as kP = 1.5, kI = 0.1. The ADRC control structure is tuned
co
ro E cd
CO >
"ro
CD
>
o
(O (O
CN
£
CN O CN
C¿ LjÜ
3
O
O ^
< o
X
O LU
H
^
<
< -J
CD
sz
>o
x
LU
O >
by choosing ^CLl = 1 rad/s and ^ESOl = 10 rad/s for the lateral controller, and ^clv = 14 rad/s and ^ESOv = 16 rad/s for the longitudinal controller. Both systems are discretized with the same sampling time Ts = 20 ms. The obtained results of the reference trajectory tracking performances, including cross-track errors, longitudinal velocities, and drive wheels velocities are depicted in Figure 3.
0.02
r 0 3
' -0.02 -0.04
0
15 20 t( s)
(a) Reference trajectory tracking
2.5 2
5 1.5
J, £ 1
0.5 0
0
15 20 t( s)
(b) Cross-track errors
(c) Longitudinal velocity control
15 20
t{ s)
f
1 j-PI/PID — ADRC
15 20 t(s)
(d) Right wheel velocity
(e) Left wheel velocity
Figure 3 - Simulation results for simulation case #1
In accordance with the set goal of this simulation case, one can see that both control approaches provide effective tracking of the reference trajectory, with similar values of the cross-track errors peaks that occur when dT(t) changes the value (see (29)). The drive wheels velocities are bounded
1712
5
10
25
30
35
5
10
25
30 35
0
5
10
25
30
35
0
5
10
25
30
35
and smooth for both systems, and it can be seen that the PI/PID algorithms generate slightly larger values.
Simulation case #2
In this case, the controllers have kept the settings from the previous one. The slippage disturbances with dynamics aR = 0.7 + 0.3sin(2 • k • Ts) and aL = 0.7 + 0.3sin(4 • k • Ts), which simulate vehicle motion over uneven dirt terrains, are included from t = 10 s.
lO CN
I
O
o
CO >
o o
a o
(a) Reference trajectory tracking
mm
■ PI/PID ■ADRC
10 15 20 25 30 35
too
(b) Cross-track errors
-pi
-ADRC
- Reference velocity
i 5 10 15 20 25 30 35 t{ s)
(c) Longitudinal velocity control
10 15 20 25 30 35
*00
10 15 20 25 30 35
*00
(d) Right wheel velocity
(e) Left wheel velocity
Figure 4 - Simulation results for simulation case #2
< o
CL
c
o
CO
0 o
JZ
0 >
TD 0
o
CO
"O 0 c c CO
E
c
"D
0 (O CO
.q
6
q: Q
< M—
o c z?
(O 0 u o
CO
E 0
(O >
CO
CO 0
1713
>
o
3
2
0
0
10
5
0
0
5
0
5
(m
O >
CM o CM
0£ yy
0£ ZD O o
-J
<
o
X
o
LU
I—
>-
Q1 <
CD
S2 ■O
x
LU I—
o
o
Also, the measurement errors of the UTV position (x,y), the angle orientation (e), and the longitudinal velocity (vv) are modeled as Gaussian noises with zero means and standard deviations [0.1 m, 0.1 m, 0.01 rad, 0.1 m/s], respectively. It corresponds with the errors of the GPS and inertial sensors in the real vehicle. The reference trajectory tracking performances and drive wheel velocities comparisons of PI/PID and ADRC are shown in Figure 4. It is noticeable that in the presence of the slippage dynamics, ADRC enables better tracking performances, with significantly lower values of the cross-track error peaks. In addition, it can be noted that the added measurement errors caused ruggedness in drive wheel velocities. However, from the zoomed part of Figure 4d and Figure 4e, it is evident that ADRC generates more smooth drive wheels velocities than the PI/PID algorithm, which is important for practical implementation.
FPGA-in-the-loop simulation validation
In order to analyze the challenges of the practical implementation, the FIL (FPGA-in-the-loop) simulation with the proposed ADRC-based UTV control is carried out. Such a validation method reduces the gap between pure simulation design (which may be too idealized) and implementation on the real vehicle (which may be time-consuming). To this end, a hardware description language code of the discrete lateral and longitudinal controllers is generated in the fixed-point word-length (WL) format, for the signals and coefficients representations, and implemented in FPGA board Basys 3 with Xilinx chip XC7A35T. By connecting JTAG port of the FPGA board and the personal computer USB port, the FIL simulations are executed by exchanging signals between the UTV model, defined in MATLAB/Simulink, and the controllers, implemented in the FPGA chip. The graphical structure of the FIL simulation methodology is demonstrated in Figure 5.
Figure 5 - FIL validation methodology
The FIL simulations are performed with the setup as in simulation case #2. The reference trajectory and longitudinal velocity tracking performances, for the ADRC algorithm implemented in the fixed-point format with WLs of 24, 22, and 20 bits, are compared with the pure MATLAB/Simulink simulation results (shown in simulation case #2) and gathered in Figure 6.
10
20
30
40
50
60
(a) Reference trajectory tracking
(b) Longitudinal velocity control Figure 6 - FIL simulation reference trajectory tracking results
In order to make the impact of the fixed-point format quantization error (QE) more visible, the differences in cross-track errors, between the pure MATLAB/Simulink simulation and the appropriate FIL simulations, denoted
lO CN
7
O O
CO >
o o
< CL
c
o
CO
0 o
JZ
0 >
TD 0
o
CO
"D 0 c c CO
E c
"O
0 (O CO
.q
6 q:
Q
<
M—
O c z?
(O 0 u o
CO £ 0
to >
CO
CO 0
>
o
0
(M
O >
CM o CM
0£ yy
0£ ZD O o
-J
<
o
X
o
LU
I—
>-
CC <
as eQE, are shown in Figure 7. The obtained drive wheels velocities are shown in Figure 8. In addition, the appropriate FPGA resource occupancy values, for different values of WLs, are shown in Table 1.
0.06 0.04 , 0.02 0
-0.02 -0.04 -0.06 -0.
0
10
15
20
25
30
35
Figure 7 - The difference in the cross-track error between the Simulink simulation and the FIL simulations with the appropriate WL of the fixed-point format
CD
S2 ■O
x
LU I—
o
o >
Table 1 - Resource occupancy of FPGA chip XC7A35T for the fixed-point ADRC controllers implementation with different WLs
FPGA resource WL = 24 WL = 22 WL = 20 Available
LUTs 5433 (26.2%) 4821 (23.2%) 4441(21.3%) 20800 (100%)
FF blocks 545 (1.3%) 525 (1.2%) 499(1.2%) 41600 (100%)
DSP blocks 82 (91%) 81 (90%) 78 (87%) 90 (100%)
As it can be seen, the FIL simulation results are close to the simulation results, which confirms the high performances of the FPGA-based implementation of the lateral and longitudinal ADRC controllers. As expected, the slight performance degradation occurs with the decrease of WL, i.e. with lower precision of the signals and coefficients in the implemented controllers. Although increasing WL requires more FPGA resource occupancy, from Table 1 one can see that it is not a significant increase that enables the realization of a high-precision fixed-point ADRC implementation on low-cost FPGA chip, such as XC7A35T.
5
10
a
16 ■ 14 ■ 12 ■
2, 8 ■ g 64 2 0 ^
^ * * ' * * * 'ft
i ; I .• '! i I I I
I ,v I 1
J_/_I_Li_1_L_Ц_/_L
' I /
V V v v
i i
I i V \
2.5 3 3.5 4 4.5 5
" adrcs
adrcf
10 15 20
t(s)
(a) Right wheel velocity
25
30
35
.■WwrtvW*-'!«*.
6 ll'
2.5
■ adrcs
adrcf
3.5 4 4.5 5
0
10
15
20
25
30
35
t(B)
(b) Left wheel velocity Figure 8 - FIL simulation drive wheel velocities
Conclusion
To improve the UTV trajectory tracking performances, the systematic use of an ADRC-based control structure was suggested in this research. The slippage disturbances and the internal UTV model uncertainties were treated collectively as lumped (total) disturbances in the lateral and longitudinal control channels. The efficiency of the proposed ADRC approach was illustrated in comparison with the classical PI/PID control, through the typical UTV trajectory tracking simulation scenarios.
LO CM
I
o o
£± cp
CO
>
cp o o
< CD
CL
o ro
JD o
<u >
"D
<U
O <0
"D
<U
CO
E
"D <U CO <0 .Q
O CC
Q <
U1
<u
"D
<0 E <u
(0 >
"ro
<u
1717
>
o
c ro
5
adrcfil;wl=22 - adrcfil;wl=20
0
0
5
7
adrchl;wl=22 - adrcf1l;wl=20
5
4 e
Additionally, the FIL simulation was used to experimentally validate the fixed-point FPGA implementation of the proposed lateral and longitudinal ADRC controllers. The conducted applied research has validated the efficacy of the proposed systematic procedure of applying the ADRC methodology for the specific UTV trajectory tracking problem, which includes its design, tuning, FPGA implementation, and performance verification.
References
2 7 ol
>
2 0 2
yy 0£
U
8 Al-Jarrah, A., Salah, M. & Almomani, F. 2019. Controlling a Skid-Steered
^ Tracked Mobile Robot with Slippage Using Various Control Schemes. In: 2019 ^ 20th International Conference on Research and Education in Mechatronics (REM). i Wels, Austria, pp.1-7, May 23-24. Available at: uu https://doi.org/10.1109/REM.2019.8744123.
>- Burke, M. 2012. Path-following control of a velocity constrained tracked vehicle
<c incorporating adaptive slip estimation. In: 2012 IEEE International Conference on ^ Robotics and Automation. Saint Paul, MN, USA, pp.97-102, May 14-18. Available at: https://doi.org/10.1109/ICRA.2012.6224684.
Chen, S., Xue, W., Lin, Z. & Huang, Y. 2019. On Active Disturbance Rejection tn Control for Path Following of Automated Guided Vehicle with Uncertain Velocities. In: 2019 American Control Conference (ACC). Philadelphia, PA, USA, pp.24462451, July 10-12. Available at: https://doi.org/10.23919/ACC.2019.8815348.
Dai, Y., Zhu, X., Zhou, H., Mao, Z. & Wu, W. 2018. Trajectory Tracking Control i for Seafloor Tracked Vehicle By Adaptive Neural-Fuzzy Inference System Algo-i- rithm. International Journal of Computers Communications & Control, 13(4), pp. 0 465-476 [online]. Available at: o https://www.univagora.ro/jour/index.php/ijccc/article/view/3267 [Accessed: 20 " March 2024].
De Luca, A., Oriolo, G. & Samson, C. 1998. Feedback control of a nonholo-nomic car-like robot. In: Laumond, J.P. (Ed.) Robot Motion Planning and Control. Lecture Notes in Control and Information Sciences. 229, pp.171-253. Berlin, Heidelberg: Springer. Available at: https://doi.org/10.1007/BFb0036073.
Gao, Z. 2003. Scaling and bandwidth-parameterization based controllertuning. In: Proceedings of the 2003 American Control Conference. Denver, CO, USA, pp.4989-4996, June 04-06. Available at: https://doi.org/10.1109/ACC.2003.1242516.
Gao, Z. 2006. Active disturbance rejection control: a paradigm shift in feedback control system design. In: 2006 American Control Conference. Minneapolis, MN, USA, p.7, June 14-16. Available at: https://doi.org/10.1109/ACC.2006.1656579.
Gonzalez, R., Fiacchini, M., Alamo, T., Guzman, J.L. & Rodriguez, F. 2010.
1719
Adaptive Control for a Mobile Robot Under Slip Conditions Using an LMI-Based Approach. European Journal of Control, 16(2), pp. 144-155. Available at: https://doi.org/10.3166/ejc.16.144-155.
Herbst, G. & Madonski, R. 2023. Tuning and implementation variants of £ discrete-time ADRC. Control Theory and Technology, 21, pp. 72-88. Available at: https://doi.org/10.1007/s11768-023-00127-0.
Hiramatsu, T., Morita, S., Pencelli, M., Niccolini, M., Ragaglia, M. & Argiolas, o
O
<
A. 2019. Path-Tracking Controller for Tracked Mobile Robot on Rough Terrain. International Journal of Electrical and Computer Engineering, 13(2), pp. 59-64. Available at: https://doi.org/10.5281/zenodo.2571922.
Hong, S., Choi, J.S., Kim, H.W., Won, M.C., Shin, S.C., Rhee, J.S. & Park, H.u. £ 2009. A path tracking control algorithm for underwater mining vehicles. Journal of Mechanical Science and Technology, 23(8), pp. 2030-2037.
Hu, J., Tao, J., Zhao, W. & Han, Y. 2019. Modeling and simulation of steering control strategy for dual-motor coupling drive tracked vehicle. Journal of the Brazilian Society of Mechanical Sciences and Engineering, 41, art.number:190. Available at: https://doi.org/10.1007/s40430-019-1692-0.
Huang, H., Zhai, L. & Wang, Z. 2018a. A Power Coupling System for Electric Tracked Vehicles during High-Speed Steering with Optimization-Based Torque Distribution Control. Energies, 11(6), art.number:1538. Available at: >
https://doi.org/10.3390/en11061538.
Huang, P., Zhang, Z., Luo, X., Zhang, J. & Huang, P. 2018b. Path Track- ^ ing Control of a Differential-Drive Tracked Robot Based on Look-ahead Distance. IFAC-PapersOnLine, 51(17), pp. 112-117. Available at: https://doi.org/10.1016/jjfacol.2018.08.072. |
Janarthanan, B., Padmanabhan, C. &Sujatha, C. 2012. Longitudinal dynamics of a tracked vehicle: Simulation and experiment. Journal of Terramechanics, 49(2), pp. 63-72. Available at: https://doi.org/10.1016/jjterra.2011.11.001. !§
Lakomy, K., Patelski, R. & Pazderski, D. 2020. ESO Architectures in the Tra- 6 jectory Tracking ADR Controller for a Mechanical System: A Comparison. In: q Bartoszewicz, A., Kabzinski, J. & Kacprzyk, J. (Eds.) Advanced, Contemporary ^ Control. 1196, pp.1323-1335. Cham: Springer. Available at:
https://doi.org/10.1007/978-3-030-50936-1_110. m
Li, Y., Yu, J., Guo, X. & Sun, J. 2020. Path Tracking Method Of Unmanned Agricultural Vehicle Based On Compound Fuzzy Control. In: IEEE 9th Joint International Information Technology and Artificial Intelligence Conference (ITAIC). E
s
>
S
"O
<u
CO
Chongqing, China, pp.1301-1305, December 11-13. Available at: https://doi.org/10.1109/ITAIC49862.2020.9338981.
Lu, H., Xiong, G. & Guo, K. 2016. Motion Predicting of Autonomous Tracked Vehicles with Online Slip Model Identification. Mathematical Problems in Engineering, 2016(1), art.number:6375652. Available at: https://doi.org/10.1155/2016/6375652.
<u
>
o
Miklosovic, R., Radke, A. & Gao, Z. 2006. Discrete implementation and generalization of the extended state observer. In: 2006 American Control Conference. Minneapolis, MN, USA, p.6, June 14-16. Available at: https://doi.org/10.1109/ACC.2006.1656547.
> Mitsuhashi, T., Chida, Y. &Tanemura, M. 2019. Autonomous Travel of Lettuce
Harvester using Model Predictive Control. IFAC-PapersOnLine, 52(30), pp. 155— 160. Available at: https://doi.org/10.1016/j.ifacol.2019.12.514.
Nonami, K., Kartidjo, M., Yoon, K.J. & Budiyono, A. 2013. Autonomous Control □E Systems and Vehicles, Intelligent Unmanned Systems. Tokyo: Springer. Available § at: https://doi.org/10.1007/978-4-431-54276-6.
° Pentzer, J., Brennan, S. & Reichard, K. 2014. Model-based Prediction of Skid-
steer Robot Kinematics Using Online Estimation of Track Instantaneous Centers of Rotation. Journal of Field Robotics, 31(3), pp. 455-476. Available at: o https://doi.org/10.1002/rob.21509.
Sabiha, A.D., Kamel, M.A., Said, E. & Hussein, W.M. 2022. ROS-based ct trajectory tracking control for autonomous tracked vehicle using optimized back-^ stepping and sliding mode control. Robotics and Autonomous Systems, 152, art.number:104058. Available at: https://doi.org/10.1016/j.robot.2022.104058.
Tang, Z., Liu, H., Zhao, Z., Lu, J., Guan, H. & Chen, H. 2021. Trajectory tracking of unmanned tracked vehicle based on model-free algorithm for off-road driv-<5 ing conditions. In: 2021 IEEE International Conference on Unmanned Systems o (ICUS). Beijing, China, pp.870-877, October 15-17. Available at: g https://doi.org/10.1109/ICUS52573.2021.9641176.
Tao, J., Liu, H., Li, Y., Guan, H., Liu, J. & Chen, H. 2021. Design of Trajec-uu tory Tracking Controller of Unmanned Tracked Vehicles Based on Torque Control. o In: 2021 IEEE International Conference on Unmanned Systems (ICUS). Beijing, " China, pp.85-92, October 15-17. Available at:
https://doi.org/10.1109/ICUS52573.2021.9641159.
Wong, J.Y. 2022. Theory of Ground Vehicles, 5th Edition. Hoboken, NJ, USA: John Wiley & Sons. ISBN: 978-1-119-71970-0.
Zhang, X., Zhang, X., Xue, W. & Xin, B. 2021. An overview on recent progress of extended state observers for uncertain systems: Methods, theory, and applications. Advanced Control for Applications: Engineering and Industrial Systems, 3(2), e89. Available at: https://doi.org/10.1002/adc2.89.
Zou, T., Angeles, J. & Hassani, F. 2018. Dynamic modeling and trajectory tracking control of unmanned tracked vehicles. Robotics and Autonomous Systems, 110, pp. 102-111. Available at: https://doi.org/10.1016/j.robot.2018.09.008.
o
1721
Diseño sistemático de la trayectoria de un vehículo de orugas no tripulado basado en ADRC seguimiento con validación i FPGA-en-el-bucle
Momir R. Stankovica, autor de correspondencia, g;
Rafal Madonskib, Stojadin M. Manojlovica
a Universidad de Defensa de Belgrado, Academia Militar, £
Belgrado, República de Serbia g-
b Universidad Tecnológica de Silesia, Facultad de Control Automático, —
Electrónica e Informática, Gliwice, República de Polonia
CAMPO: control automático, ingeniería de control,
robótica móvil (3
TIPO DE ARTÍCULO: artículo científico original
Resumen:
A
a
Introducción/objetivo: El problema del control de seguimiento de o trayectoria en un vehículo de orugas no tripulado (UTV) representa una tarea desafiante, debido a la dinámica de deslizamiento desconocida e inmensurable que inevitablemente existe durante el movimiento. Por lo tanto, la aplicación de esquemas de control industrial estándar suele ser limitada.
JD O
e v d e k c a
a
Métodos: En este artículo, se propone un esquema de control activo de rechazo de perturbaciones (ADRC) para los canales de control longitudinal (control de velocidad longitudinal del vehículo) y canales de control lateral (control del ángulo de trayectoria del vehículo) del UTV para manejar colectivamente todas las in- E certidumbres del modelado de la planta y las perturbaciones de
deslizamiento actuantes. s
as
Resultados: Se presenta un procedimiento paso a paso para ¿
aplicar el algoritmo ADRC para el caso específico de seguimien- g
to de trayectoria de UTV. Incluye diseño sistemático, discretiza- < ción, así como análisis y validación del rendimiento utilizando simulaciones FPGA-en-el-bucle (FIL).
Conclusión: El método de validación basado en FIL propuesto reduce la brecha entre el diseño de simulación pura (que puede estar demasiado idealizado) y la implementación en el vehículo real (que puede llevar mucho tiempo). Los resultados experimentales obtenidos muestran las ventajas de la estructura de control propuesta sobre los controladores PI/PID industriales.
e t
s
>
S
a t e
v o k n a t
S
Palabras claves: vehículos de orugas no tripulados, seguimiento de trayectoria, control de rechazo activo de perturbaciones, control de velocidad, controlador PID, FPGA-en-el-bucle, validación
ГМ
de hardware.
^ Проектирование движения беспилотной гусеничной
о машины по заданной траектории на основе контроллера
- ADRC и моделирования ш
^ Момир Р. Станкович3, корресподент,
g Рафал Мадонски6, Стоядин М. Манойловича
^ a Универстет обороны в Белграде, Военная академя,
< г Белград, Республика Сербия
б Силезский технологический университет, Факультет q автоматического управления, электроники и вычислительной
ш техники, г Гливице, Республика Польша
£ РУБРИКА ГРНТИ: 50.43.00 Системы автоматического
¡5- управления, регулирования и
контроля
ВИД СТАТЬИ: оригинальная научная статья
от Резюме:
о Введение/цель: Разработка системы автономного сле-
g жения за заданной траекторией беспилотной гусеничной
машины является сложной задачей из-за неизвестной и ш неизмеримой динамики скольжения. Поэтому применение
о стандартных алгоритмов промышленного управления за-
частую ограничено.
о
Методы: В данной статье представлена схема активного управления подавлением помех (ADRC) для продольного (управление продольной скоростью транспортного средства) и бокового (управление углом наклона транспортного средства) каналов управления UTV, позволяющая в совокупности справляться со всеми неопределенностями, связанными с моделированием установки и действующими возмущениями при скольжении.
Результаты: В статье представлена пошаговая процедура применения алгоритма ADRC для конкретного случая отслеживания траектории беспилотной гусеничной машины.
Процедура включает: систематическое проектирование,
дискретизацию, а также анализ производительности и ва-лидацию с использованием моделирования FPGA-in-the-loop (FIL).
Ключевые слова: беспилотные гусеничные машины, отслеживание траектории, активное управление подавлением помех, регулирование скорости, PID-контроллер, моделирование FIL, валидация аппаратного обеспечения.
1723
7
о о
Выводы: Предлагаемый метод проверки на основе встроенного моделирования FIL сокращает разрыв между имитационным проектированием (которое может быть ^ слишком идеализированным) и реализацией на реаль- о ном транспортном средстве (что может занять много времени). Полученные экспериментальные результаты демонстрируют преимущества предложенной системы < управления по сравнению со стандартными промышлен- cl ными контроллерами в различных условиях эксплуатации беспилотной гусеничной машины.
о го
о ф
> "О
ф
Про]ектова^е крета^а беспосадног гусеничног возила по задащ путали на основу ADRC регулатора и FIL о
симулаци]а
a ф
Момир Р. Станкови^3, аутор за преписку,
Рафал Мадонски6, Сто]адин M. Мащлови^15 |
a Универзитет одбране у Београду, Во]на академи]а, Београд, Република Срби]а
"о ф
б (Я
б Шлески технолошки универзитет, Факултет за аутоматско управ^а^е, електронику и рачунарство, 6
Гливице, Република Поиска о
<
ОБЛАСТ: аутоматско управ^а^е, рачунарске науке, роботика КАТЕГОРША (ТИП) ЧЛАНКА: оригинални научни рад
Сажетак:
го
Увод/цил>: Проjектова^е система аутономног праЪеъа Е задате траjекториjе беспосадног гусеничног возила пред- ю ставка сложен задатак због постоjа^а непознате и не- & меръиве динамике проклизаваъа. Стога jе примена стан- ^ дардних индустр^ских управъачких алгоритама често ограничена.
(Я ф
"О
ф
>
о
с
го '
от
Методе: Предложена je примена регулатора на основу управлаша са активним потискивашем порeмelаjа (ADRC), посебно проjeктованих за лонгитудинални и ла-терални канал управлаша возила. Примена наведеног ал° горитма омогу1ила je високе перформансе управлаша у условима постоjаша нестационарности модела обjакта о управлаша и утицаjа порeмelаjа проклизаваша.
g Резултати: Представлена je детална процедура примеси не ADRC алгоритма за праПеше задате путане беспосад-о ног гусеничног возила, ща je обухватила проjeктовашe, ^ дискретизац^у, симулациону анализу перформанси и ек-< спериманталну верификац^у на основу симулац^а са FP-GA хардвером у петли управлаша (FIL симулац^е).
х
ш Заклучак: Предложена методолог^а валидац^е проjeкто-
>
ваног система управлаша на основу FIL симулац^а омо-губила je смашеше проjектантског времена измену чисто рачунарских симулац^а (щ'е су наjчeшhe сувише идеализо-ване) и експерименталних верификац^а на реалном систему. Доб^ени резултати су показали предности предложе-ног решена у односу на стандардне индустр^ске регула-<| торе у различитим условима употребе беспосадног гусе-
о ничног возила.
SZ
У Кшучне речи: беспосадно гусенично возило, праПеше пу-
х таше, управъаше са активним потискивашем порeмelаjа
(ADRC), управъаше брзином, PID регулатор, FIL симулаци-jа, хардверска валидац^а.
о >
EDITORIAL NOTE: The second author of this article, Rafal Madonski, is a current member of the Editorial Board of the Military Technical Courier. Therefore, the Editorial Team has ensured that the double blind reviewing process was even more transparent and more rigorous. The Team made additional effort to maintain the integrity of the review and to minimize any bias by having another associate editor handle the review procedure independently of the editor - author in a completely transparent process. The Editorial Team has taken special care that the referee did not recognize the author's identity, thus avoiding the conflict of interest.
