IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 8, NO.

7, JULY 2023 3925

Adaptive Bioinspired Preview Suspension Control

With Constrained Velocity Planning for
Autonomous Vehicles
Tenglong Huang , Jue Wang , and Huihui Pan , Senior Member, IEEE

Abstract—Autonomous vehicles equipped with numerous ad-

vanced sensors are capable of obtaining road preview information,
creating new opportunities for vehicle suspension systems. This
article proposes a novel preview suspension control method from
adaptive nonlinear control perspectives with less computational
burden and is more realistic, unlike optimization-based works or
existing linear state-space models-based results that neglected non-
linear terms. The X-shaped bio-inspired dynamics derived from Fig. 1. Autonomous vehicles with road preview information.
animal or insect skeleton structures are introduced to reduce en-
ergy consumption by utilizing beneficial geometrical nonlinearities.
Meanwhile, optimal velocity planning approach is investigated to sensors, as illustrated in Fig. 1, have powerful sensing capa-
balance vehicle passage time, vibration suppression, and longitu- bilities that enable them to obtain road excitation in advance.
dinal comfort by solving a multi-objective optimization problem Connected autonomous vehicles [5], [6], [7], [8] can also ob-
with the aid of road preview information. Moreover, acceleration tain sufficient environment information via cloud services or
constraint reduces the search space and computing requirements, vehicle-to-everything technology by virtue of wireless network
while ensuring planned velocity optimality. Simulation and ex-
periment results are provided to demonstrate the effectiveness communication [9], [10]. This provides new perspectives and
and advantages of the constructed energy-saving adaptive preview opportunities for handling vehicle suspension control problems
control framework with constrained velocity planning. using road preview information which is the focus of this devel-
Index Terms—Adaptive preview suspension, autonomous vehic-
opment. Although related research is emerging, it is still lacking.
les, bioinspired dynamic, constrained velocity planning, energy- There have been few effective attempts to design a preview
saving. suspension control scheme in the literature. The front vehicle is
taken as preview sensor to construct an optimal controller based
I. INTRODUCTION on the linear quadratic regulator (LQR) in [11]. The performance
improvement of output feedback H∞ /GH2 control strategy over
EHICLE chassis system consists of lateral, longitudinal,
V and vertical subsystems corresponding to steering, driv-
ing/braking, and suspension systems, respectively. The vehicle
linear quadratic gaussian preview is validated in [12]. Model
predictive control (MPC) provides another feasible solution to
this issue. An implicit MPC controller [13] is implemented to
suspension system [1], [2], [3] plays a crucial role in the per-
illustrate the benefits that come from MPC and preview data by
formance of ride comfort, thus, making it an active area and
means of experimental results on a high-mobility vehicle, which
has captured considerable attention from researchers. Particu-
is compared with the non-preview skyhook damping controller
larly, autonomous vehicles [4] equipped with various perception
and LQR method. Hybrid MPC with time-varying horizon is
utilized to handle the preview information in [14]. MPC in-
Manuscript received 3 April 2023; revised 29 April 2023; accepted 1 May
2023. Date of publication 8 May 2023; date of current version 21 August 2023. corporated with displacement trajectory optimization utilizing
This work was supported in part by the National Natural Science Foundation of measured road data is adopted in [15], improving ride comfort
China under Grants 62173108, U1964201, and 62022031, in part by the Post- with fewer optimization variables. To overcome memory and
Doctoral Science Foundation under Grants 2019T120270 and LBH-TZ2111, in
part by the Major Scientific and Technological Special Project of Heilongjiang computational burdens, the explicit regionless MPC in [16]
Province under Grant 2021ZX05A01, and in part by the Fundamental Research employing offline optimization is developed. An adaptive gain
Funds for the Central Universities under Grant HIT.OCEF.2022012. (Corre- scheduling preview controller is realized in [17] by adjusting the
sponding author: Huihui Pan.)
Tenglong Huang and Jue Wang are with the Research Institute of Intelligent gain automatically. Nevertheless, most of the recent emergence
Control and Systems, Harbin Institute of Technology, Harbin 150001, China of vehicle preview control methodology is optimization-based
(e-mail: [email protected]; [email protected]). or uses simplified linear state-space models. Accordingly, the
Huihui Pan is with the Research Institute of Intelligent Control and Sys-
tems, Harbin Institute of Technology, Harbin 150001, China, and also with the designed controllers present expensive computational burdens
Robot Innovation Center Company Ltd., Harbin Institute of Technology, Harbin or ignore the nonlinear characteristics.
150001, China (e-mail: [email protected]). The classical computationally efficient adaptive control
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/TIV.2023.3273620. method [18] achieves an impressive effect by controlling
Digital Object Identifier 10.1109/TIV.2023.3273620 the sprung mass in conjunction with zero dynamic analysis.
2379-8858 © 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
3926 IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 8, NO. 7, JULY 2023

However, it must be pointed out that road excitation is isolated

by the unsprung mass in this approach and is thus no longer
applicable for suspension preview control. One viable option is
to take suspension space and its derivative as the system state
variables. Moreover, another critical issue worth considering
for suspensions is energy consumption, especially for active
suspensions. In this case, this article attempts to reduce the
energy consumption while decreasing the computational burden.
To achieve energy savings and provide a reference trajectory,
bioinspired dynamics are constructed elaborately and introduced
as a reference model. Inspired by the superior vibration isolation
achieved by the special structures of bird legs or animal skele-
tons [19], the study of limb-like anti-vibration structures has
been an interesting and hot topic. Scissor-like and X-type [20] Fig. 2. Quarter-vehicle model.
bionic structures have been designed and successfully applied to
spacecraft, Stewart platform, and underwater robots. It is worth the authors’ knowledge, this study is the first to propose
noting that asymmetrical X-shaped structures [21] are effective a nonlinear adaptive bionic preview suspension controller
in isolating vibrations while obtaining excellent energy savings that is based on a nonlinear state-space model. The con-
and, more importantly, the stability of the dynamic structure troller is more realistic and has low computational require-
has been proved by the Lyapunov method. Consequently, the ments, as no optimization is required.
bionic dynamics model with an asymmetric X-shaped structure 3) Moreover, the animal limb-inspired bioinspired dynamics
is designed and adopted in this article to optimize ride comfort are introduced as reference trajectories to take advantage
and energy consumption by exploiting beneficial non-linearities. of the beneficial nonlinearities, achieving significant en-
Human drivers intuitively slow down to suppress the dis- ergy savings. The proposed method is deployed to the
comfort caused by a bulge ahead, such as a speed bump, and physical platform, and the simulation and experimental
then accelerate after passing the bulge. Road preview infor- results demonstrate the advantages of this approach.
mation for autonomous vehicles offers the possibility of mim-
icking this behavior through longitudinal velocity planning. II. PROBLEM STATEMENT
Nevertheless, frequent acceleration and deceleration have detri- The quarter vehicle model of active suspension and con-
mental effects on vehicle performance. Thus, the velocity plan- strained multi-objective optimization problem are stated in this
ning is modeled as a longitudinal and vertical multi-objective section. Additionally, quantitative performance indicators are
optimization problem to strike a balance in this article. In introduced.
the literature [22], [23], several works have begun to exploit
road preview information for path planning, vertical trajectory A. Active Suspension Mathematical Model
planning and longitudinal velocity planning [24]. For example,
[14] proposes a novel comfort-orient speed planning algorithm, As shown in Fig. 2, the active suspension [25], [26] can be
however, it searches the entire speed candidates space, which described as a two-degrees-freedom spring-damper system. The
can easily lead to unreasonable planned speed or require a large weights of vehicle body and wheel are considered as sprung
adjustment interval. Relevant research is still limited and leaves mass m0 and unsprung mass m1 , respectively. The vertical
some room for improvement. displacements of m0 and m1 are represented as z0 and z1 with
Motivated by the challenges and discussions above, this article road surface height z2 . Fk and Fc denote the forces produced
proposes an adaptive bioinspired preview suspension control by the nonlinear spring and damper of suspension. The spring
scheme with constrained velocity planning for autonomous ve- and damper forces for wheel tire are described as Fu and Fr .
hicles, which offers the following contributions: And u is the control input produced by electric actuator. Based
1) A constrained velocity planning algorithm is proposed to on Newton’s law and the mechanical characteristics of spring
balance the vertical ride comfort, longitudinal comfort, and damper, the vertical dynamics of vehicle body motion and
and passage time. The presented algorithm allows for arbi- unsprung mass can be derived as follow
trary velocity regulation interval and avoids unreasonable m0 z̈0 = − Fk (z0 , z1 ) − Fc (ż0 , ż1 ) + u(t) (1)
or undesired velocity jumps by taking constraints into
account. The optimality of the planned velocity trajectory m1 z̈1 = Fk (z0 , z1 ) + Fc (ż0 , ż1 )
is ensured, and velocity candidates that do not satisfy the − Fu (z1 , z2 ) − Fr (ż1 , ż2 ) − u(t) (2)
constraints are removed from search space, resulting in a
reduced computational demand. where
2) Based on the planned velocity and road information, this Fc (ż0 , ż1 ) = kc1 (ż0 − ż1 ) + kc2 (ż0 − ż1 )2
article presents a preview controller employing generated
road excitation from an adaptive control perspective. To Fk (z0 , z1 ) = km (z0 − z1 ) + kl (z0 − z1 )2 + kn (z0 − z1 )3

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
HUANG et al.: ADAPTIVE BIOINSPIRED PREVIEW SUSPENSION CONTROL WITH CONSTRAINED VELOCITY PLANNING 3927

Fu (z1 , z2 ) = ku (z1 − z2 ) , Fr (ż1 , ż2 ) = kr (ż1 − ż2 )

and the nonlinear and linear stiffness of the suspension spring are
kn , kl and km . The suspension damper coefficients are kc1 and
kc2 . Correspondingly, ku and kr are the coefficients of equivalent
spring and damper for wheel tire.
Benefiting from non-linear stiffness and passive damping
properties, quasi-zero stiffness bioinspired dynamics resulted
from an asymmetrical X-shape structure can effectively isolate
vibrations while achieving excellent energy efficiency. There-
fore, reference bioinspired dynamics with ideal damping are
constructed elaborately in this article, as illustrated on the right- Fig. 3. Overview of the proposed method.
hand side of Fig. 2, to provide a reference trajectory for the plant.
The implementation of the reference bioinspired dynamics is
detailed in the next section.

B. Constrained Optimization
With the advent of emerging technologies, the acquisition of
irregular road excitation information for autonomous vehicles
has become possible. Feasible candidate solutions include de-
tection via optical sensor or cloud-assisted network communi-
cation, as illustrated in Fig. 1. Inspired by the fact that human
drivers slow down to reduce vibration when approaching a bump
and accelerate to pass quickly when leaving it, an intuitive idea
is to further optimize performance through velocity planning
when global road information is assumed to be available. In Fig. 4. Schematic of the constrained velocity planning.
detail, the optimization-based velocity planning problem can be
formulated as striking a balance of the following aspects: III. MAIN RESULTS
r Reasonable planned velocity (the vehicle acceleration con-
This section first provides details on how to optimize longitu-
straint should be satisfied to avoid generating an unreason-
dinal and vertical performance simultaneously by searching for
able or unsatisfactory planned velocity).
r Vehicle body vibrations suppression (which can be quan- a reasonable velocity trajectory that satisfies constraints using
road preview information as in Section III-A. Then, based on the
tified by the vertical acceleration of sprung mass).
r Fast passage of vehicle (which can be reflected intuitively planned velocity and corresponding road excitation, a preview
adaptive suspension nonlinear controller that does not require
by the vehicle passage time).
r Excellent longitudinal comfort (which can be evaluated by optimization is designed in Section III-C by combining bionic
dynamics, as in Section III-B, derived from the animal limb
the vehicle longitudinal acceleration).
structure.
The schematic overview of the proposed method is illustrated
C. Performance Index in Fig. 3. Firstly, a constrained planned velocity trajectory can
be obtained by solving a multi-objective optimization prob-
To quantity and compare effects of various control frame-
lem to balance the multi-dimensional performance cost. The
works, the root mean square (RMS) of vehicle body acceler-
corresponding time-domain road excitation can be generated
ation [27] and positive mechanical power P + of actuator are
employing the road information obtained from communication
introduced as performance indexes for ride comfort and energy
or optical sensors and the planned velocity. Then, the control
consumption, respectively, as follow
input is calculated by means of the proposed adaptive preview
 
 L  L
controller.
1 1
RM Sz̈0 = z̈02 dt, RM SP + = (P + (t))2 dt
L 0 L 0 A. Constrained Velocity Planning
(3)
Road preview information allows autonomous vehicles to
adjust the planned velocity according to road conditions ahead
where L denotes the time duration, and the power P + can be
in order to find the optimal trade-off between longitudinal and
calculated as
vertical performance, which is the primary issue in this sub-

u(t) (ż0 − ż1 ) , for u(t) (ż0 − ż1 ) > 0 section. First, as depicted in Fig. 4, the road ahead is partitioned
+
P (t) = depending on the locations of road bumps, with the both sides of
0, else bumps and the areas entering the end point or leaving the initial

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
3928 IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 8, NO. 7, JULY 2023

point as speed regulation zones. Then, adjustment locations are Step 3: Among the remaining velocity candidate points
selected with the spatial granularity Δx in adjustment areas. after Step 1, there are still velocity change branches that do not
Each adjustment location contains a speed candidate set Sv = satisfy the constraints (4), which are removed in this step, and
{vmin , vmin + Δv, . . ., vmax }, which is generated by velocity only reasonable velocity change branches are retained. For illus-
increment Δv from the minimum speed vmin to the maximum tration purposes, one possible planned velocity is represented by
speed vmax . The search space, as displayed in Fig. 4, consists of red arrows in Fig. 4. The corresponding unreasonable velocity
all velocity candidates of adjustment positions. change branches, drawn as a dashed line, are deleted. Note that
It is worth mentioning that a new efficient preview planning this operation is applied for all remaining velocity points when
method is proposed in [14] by means of the speed candidate performing a velocity search maneuver, only a small portion is
space above. Despite that, some key aspects still need to be illustrated in Fig. 4 for clarity.
improved: 1) The planned velocity is searched from the en- Step 4: With Steps 1-3 described previously, a constrained
tire candidate space, which potentially results in unrealistic or multi-objective optimization problem that takes longitudinal
unreasonable velocity changes between two adjacent velocity and vertical performance into account is solved in this step
adjustment points. 2) One way to avoid this issue is to enlarge to generate the optimal velocity trajectory. The overall cost
spatial granularity Δx, which means that a large regulation function Ξ(vi , vj ) from velocity candidate vi to next velocity
interval, namely, the speed regulation is not flexible enough. candidate vj in the adjacent position can be written as
3) An entire space search for all speed candidates would cause
Ξ (vi , vj ) = κ1 Γ1 (vi , vj , lij ) + κ2 Γ2 (vi , vj , lij )
a high computational burden.
For this reason, an acceleration constrained planning algo- + κ3 Γ3 (vi , vj , lij ) (7)
rithm is proposed and detailed in this subsection, where the
where Γ1 (vi , vj , lij ) = Cij (Bi , v, i ) represents the vertical
assumptions remain the same as [14]. For clarity, an uneven v 2 −v 2
road with three bumps is plotted as Fig. 4, where each speed ride comfort, Γ2 (vi , vj , lij ) = i2lij j denotes the longitudinal
regulation zone width is set to w = 2Δx. acceleration is utilized to quantify longitudinal comfort, and
Step 1: Velocity candidate points that do not satisfy the 2lij
Γ3 (vi , vj , lij ) = vi +v j
is the vehicle passage time. κ1 , κ2
acceleration or deceleration limits are removed, as points con-
and κ3 are the associated weights for the different performance
tained in the removed areas in Fig. 4. That is, the corresponding
indicators. Then the optimal desired velocity trajectory can
velocity candidate is removed from the search space, if
⎧ 2 2 be obtained, employing dynamic programming, by minimizing
⎨ vi −vi+1 > d the following cost function
2lij max , for vi ≥ vi+1
2
−v 2 (4)
⎩ v i+1 i
> amax , for vi+1 > vi Θ (vm ) = min {f (vm , vn ) + Θ (vn )} (8)
2lij vm ,vn ∈Sv

where positive constants amax and dmax denote the preset max- where f (vm , vn ) = Ξ(vm , vn ) if vm , vn ∈ Sv are the velocity
imum acceleration and deceleration, respectively. vi and vi+1 candidates in adjacent columns; Otherwise, f (vm , vn ) = ∞.
denote two candidate velocities at adjacent positions and lij is The proposed constrained velocity planning algorithm has the
the distance between them. One possible velocity search case is following key features:
annotated in Fig. 4. r Acceleration and deceleration limits (amax , dmax ) can be
Step 2: Vertical costs of passing each adjustment location pre-specified arbitrarily and thus unreasonable velocity can
at feasible velocities are calculated to evaluate the vertical per- be avoided.
formance at various velocities. The vertical cost Ci (Bi , v, i ) r The spatial granularity Δx between adjacent locations and
of passing i-th bump Bi of width i with velocity v ∈ Sv can the width w of speed regulation zone can be set flexibly to
be approximated as meet user requirements.
 i r Compared to unconstrained velocity planning, the search
v v 
Ci (Bi , v, i ) = 1 Υ21 (t) + 2 Υ22 (t) + 3 Υ23 (t) dt space of the proposed algorithm and the corresponding
i 0 computational burden are reduced by the considering con-
(5) straints.
where 1 , 2 and 3 denote the weighting parameters of the cor-
r Meanwhile, the optimality of the generated velocity trajec-
responding vertical performance components Υ1 = z̈0 , Υ2 = tories is ensured.
z0 − z1 , Υ3 = z1 − z2 , respectively. Then, the vertical cost Cij The proposed constrained velocity planning is essentially an
of the j -th column velocity candidate points in the left or right optimization problem, by minimizing overall cost consisting of
of i-th bump Bi can be expressed accordingly as the vehicle passage time Γ3 , the longitudinal comfort cost Γ2 ,
and the vertical ride comfort cost Γ1 , in the feasible velocity
i + Δx
Cij (Bi , v, i ) = j−1 Ci (Bi , v, i ), j = 1, . . ., candidate set which satisfies the constraint (4). It can be con-
Δx cluded from (7) that the overall cost Ξ(vi , vj ) is a weighted sum
(6)
of the vehicle passage time Γ3 , the longitudinal comfort cost Γ2 ,
where positive constant  ∈ (0, 1) is used to indicate that the and the vertical ride comfort cost Γ1 . As the weighting factor
vehicle is approaching or leaving a bump. As an illustration, for κi , i = 1, 2, 3 of the corresponding cost indicator is larger, the
B3 in Fig. 4, C31 = C3 , C32 = C3 , C32 = 2 C3 . proportion of the corresponding performance in the total weight

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
HUANG et al.: ADAPTIVE BIOINSPIRED PREVIEW SUSPENSION CONTROL WITH CONSTRAINED VELOCITY PLANNING 3929

d1 + d2 represent rotational and horizontal motion. The dis-

placement of the sprung mass Mv relative to the base is defined
as dr = dv − z1 . It is clear that

Si sin δi + d2r
ηi = arctan − δi (9)
Si cos δi − di
 2
dr
Fig. 5. Illustration of the limb-like bio-inspired dynamics. (a) A crane with its di = Si cos δi − Si2 − Si sin δi + (10)
asymmetric bones and articulations. (b) Asymmetric X-shaped mechanisms ex- 2
ist in the leg. (c) Static bio-inspired structure. (d) Comparison after deformation,
the deformed parts are shown in yellow. where i = 1, 2, S1 = Sl , and S2 = Sr .
With the aid of Hamilton principle, the bionic dynamics can
be described as
increases, which implies a greater focus on the corresponding
aspect. In practice, the weighting factors can be adjusted accord- M d¨r + p1 (dr ) + k2 dr + ς1 d˙r + ς2 jn p2 (dr ) d˙r = −M z̈1
ing to the actual requirements to balance the demand for various (11)
performances.
where ς1 , ς2 indicate the air drag and friction factor. And
Remark 1: By design of the presented optimal constrained  2
h ) d(dr )
planning algorithm, spatial granularity Δx and velocity in- p1 (dr ) = k1 dh d(d
d(dr ) d(dv ) , p 2 (d r ) = dη
d(dr ) can be further
crement Δv can be set arbitrarily, while meeting acceleration determined by combining geometric relationship (9), (10) men-
limits. This ensures velocity feasibility and operation flexibility, tioned above, as
which also helps to reduce the computational requirements by 
removing the unreasonable candidates. k1
p1 (dr ) = Sl cos δ1 + Sr cos δ2 − Sl2 − ξ 2 (dr )
Remark 2: Based on the planned longitudinal velocity trajec- 2
tory and road information, the transformation approach in [14]  
can be used to generate time-domain road excitation. Note that − Sr2 − ξ 2 (dr )
the road information sampling time T is independent of the
ξ (dr ) ξ (dr )
controller, which can be set arbitrarily to match the processing ×  +
requirements. The time-domain excitation processing is detailed Sl2 2
− ξ (dr ) Sr − ξ 2 (dr )
2

in [14]. 2
1 1
p2 (dr ) =  + 
2 Sl2 − ξ 2 (dr ) 2 Sr2 − ξ 2 (dr )
B. Bioinspired Dynamics
Animal limb structures [28] with the notable capability to where ξ(dr )  Sl sin δ1 + d2r .
suppress vibration, such as the shin bones and articulation of When the bioinspired dynamics are applied to autonomous
a crane in Fig. 5(a), provide inspiration for vehicle suspension vehicles, the structure and parameters can be chosen rationally
system. Exploiting beneficial nonlinearities by introducing the according to the bionic dynamic properties as in [29], while the
limb-inspired bionic structure contributes to achieving favorable unsprung acceleration signal z̈1 is obtained by the sensor, then
performance at a lower cost. A common asymmetric structure, the reference is achievable.
as displayed in Fig. 5(b), is modelled and introduced in this Remark 3: The road excitation information z2 , ż2 can be
section to emulate this characteristic by exploiting beneficial incorporated into the adaptive control framework by taking
nonlinearities. Noting that beneficial non-linearities present in z0 − z1 and ż0 − ż1 as the state variables. However, this poses
the X-shaped mechanism are not only rigorously analyzed math- an additional problem: how to determine the corresponding
ematically [29] but are deployed and validated on experimental reference trajectory. 0 is taken as the reference in the most
platforms as in [28], [29]. By elaborately designing the X-shaped existing results, which is unreasonable obviously, for example
structure as in [29] and introducing it into the controller design to a continuous sinusoidal pavement. The sinusoidal reference is
exploit the beneficial non-linearities can consequently achieve also used, which is still doubtful, e.g. a flat road. The bionic
effective vibration isolation performance with less energy con- dynamics in this section provides a viable reference that isolates
sumption. The geometric relationship of the asymmetric static vibration while reducing energy consumption.
bionic mechanism is drawn in Fig. 5(c). Sl and Sr denote the Remark 4: The introduced bionic dynamics contribute to
rods lengths on the left and right side with initial angle δ1 and reducing energy consumption by employing beneficial non-
δ2 , respectively. k1 and k2 represent linear spring stiffness in linearities. Nonetheless, the bionic model constructed in this ar-
horizontal and vertical directions. dv is the vertical displacement ticle is more suitable for urban road scenarios. The performance
of the sprung mass M , and road input or excitation is denoted of the bionic dynamics is closely related to the bionic structure
as z1 . and its parameters, such as the asymmetric rod length. When
To analyze the dynamic characteristics, the deformation com- excessively severe road excitation occurs, the direct introduction
parisons are plotted in Fig. 5(d) where η  η1 + η2 , dh  of the bionic structure designed in this article may not achieve

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
3930 IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 8, NO. 7, JULY 2023

desired results. For special road scenarios, such as off-road ter- The estimations of uncertain vector Λ and γ are denoted as Λ̂
rain and mountainous regions, further adjustments are required and γ̂. Correspondingly, the resulting estimate errors are written
depending on the road excitation and bionic dynamics response as γ̃  γ̂ − γ and Λ̃  Λ̂ − Λ. Finally, the controller u is given
characteristics [29]. by

C. Adaptive Preview Control Design u = γ̂(−ρ2 e2 − e1 − Λ̂T ψ − Z − R + ṙdr + α̇1 ) (19)

Let vehicle suspension deflection sd  z0 − z1 and its dy- where ρ2 ∈ R+ and the rate of convergence of e2 can be adjusted
namic rd  ż0 − ż1 be the state variables, in the light of (1) and by adapting parameter ρ2 . Meanwhile, the adaptive laws are
(2), the suspension model can be rearranged as developed as
ṡd = rd (12) ˙
Λ̂ = K−1 e2 ψ, γ̂˙ = Projγ (20)
1
ṙd = Λ ψ + u + Z + R
T
(13)
γ where K =diag(κ1 , κ2 , κ3 ) is an invertible diagonal matrix with
T
the design parameters κi , i = 1, 2, 3. The projection operation
where Λ  [− m km
0
, −m kl
0
, −m
kn
0
, − kmc10 , − kmc20 ] and γ Projγ can be stated as
m0 m1
m0 +m1 are unknown because it contains uncertain ⎧
sprung mass m0 ∈ (mmin , mmax ), and thus, γ ∈ ⎨0, if γ̂ = mmax +m1 and γ̂˙ 0 > 0
⎪ mmax m1

( mmmin +m1 , mmax +m1 ).

min m1 mmax m1
ψ  [sd , s2d , s3d , rd , rd2 ]T Projγ = 0, if γ̂ = mmmin ˙
+m1 and γ̂ 0 < 0
and min m1

2 3 ⎪
⎩γ̂˙ , otherwise.
Z − m1 (z0 − z1 ) − m1 (z0 − z1 ) − m1 (z0 − z1 ) −
km kl kn
0
2
m1 (ż0 − ż1 ) − m1 (ż0 − ż1 ) + m1 z1 + m1 ż1
kc1 kc2 ku kr
are known
vector. R = − m1 z2 − m1 ż2 denotes the road excitation
ku kr with γ̂˙ 0 = ϑe2 (ρ2 e2 + e1 + Λ̂T ψ + Z + R − ṙdr − α̇1 ), ϑ ∈
information, which can be obtained by the optical sensor or R+ .
wireless network communication. Subsequently, an adaptive By design of the virtual control input (16), the controller
controller is synthesized to track the desired bioinspired (19), and the adaptive laws (20), the reference bionic dynamics
dynamics (11). tracking errors e1 , e2 converge asymptotically to the origin.
Step 1: Develop virtual controller α1 in this step so that sd The asymptotic convergence is proved with the following
is capable of tracking the reference trajectory sdr . Lyapunov function candidate
The tracking errors are defined as 1 2 1 2 1 2 1 T
V = e + e + γ̃ + Λ̃ KΛ̃ (21)
e1 = sd − sdr , e2 = rd − rdr − α1 (14) 2 1 2 2 2ϑγ 2

where sdr and rdr are the reference trajectory for sd , rd which The derivative of (21) with respect to time can be represented
can be calculated from bioinspired dynamics (11). α1 denotes as
the virtual control input, based on (12)–(14), we have 1 ˙ ˙
. V̇ = e1 ė1 + e2 ė2 + γ̃ γ̂ + Λ̃T KΛ̂
e1 = e2 + α 1 (15) ϑγ

Aiming to track the desired trajectory sdr generated from the Combining (18) and the designed controller (19) yields
bionic dynamics (11), the virtual control input α1 is designed as 
1
V̇ = − ρ1 e21 + e2 e1 + ΛT ψ + u + Z + R − ṙdr − α̇1
α1 = −ρ1 e1 (16) γ
where the parameter ρ1 ∈ R+ is manually set to achieve satis- 1 ˙ ˙
+ γ̃ γ̂ + Λ̃T KΛ̂
factory performance. Obviously, with the virtual control input ϑγ
(16), the differential inequality V̇1 ≤ 0 of Lyapunov function 1
V1 = 12 ρ1 e21 holds when e2 = 0. This means that sdr can be = − ρ1 e21 − ρ2 e22 + γ̃e2 ϑ(−ρ2 e2 − e1 − Λ̂T ψ − Z
ϑγ
tracked asymptotically, if e2 = 0, with the virtual controller (16).
1 ˙ ˙
To ensure e2 can converge to 0, design controller u in Step 2. − R + ṙdr + α̇1 ) + γ̃ γ̂ + Λ̃T KΛ̂ − e2 Λ̃T ψ
Step 2: To further ensure that the asymptotic convergence ϑγ
of tracking error e2 with uncertain parameter m0 is achieved, By invoking the adaptive laws (20), we obtain
the controller u and adaptive laws are given in this step.
Combining (15) with (16), we get 1
V̇ = − ρ1 e21 − ρ2 e22 + γ̃e2 ϑ(−ρ2 e2 − e1 − Λ̂T ψ
ϑγ
e1 ė1 = −ρ1 e21 +e1 e2 (17)
1
Invoking (13) and bionic dynamic (11), the error dynamic of − Z − R + ṙdr + α̇1 ) + γ̃Projγ
ϑγ
e2 can be expressed as
+ Λ̃T KK−1 e2 ψ − e2 Λ̃T ψ
1
ė2 = ΛT ψ + u + Z + R − ṙdr − α̇1 (18)
γ ≤ − ρ1 e21 − ρ2 e22 ≤ 0

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
HUANG et al.: ADAPTIVE BIOINSPIRED PREVIEW SUSPENSION CONTROL WITH CONSTRAINED VELOCITY PLANNING 3931

the first time in this article. Meanwhile, unlike conventional non-

linear adaptive controllers [27] based on the unsprung subsystem
that cannot be applied to preview suspension, the presented
controller provides a novel way to handle nonlinear preview
suspension control.

IV. COMPARATIVE STUDY

A. Preview Controller Performance
Based on the Hyundai Elantra suspension, the effectiveness
of the proposed adaptive preview control method is tested and
analyzed by means of a B-class random road obtained via the
filtered white noise method. The bionic dynamics provides a ref-
erence trajectory for the preview suspension, where the suspen-
sion (1)–(2) parameters: m0 = 240 kg, m1 = 23.61 kg, km =
15394 N/m, kl = −73696 N/m2 , kn = 3170400 N/m3 , kc1 =
Fig. 6. Sprung mass acceleration z̈0 . (a) Time-domain response of z̈0 . 1385.4 Ns/m, kc2 = 524.28 Ns/m2 , ku = 181818.88 N/m, kr =
(b) Frequency component of z̈0 by fast fourier transform. 138 Ns/m. And 12.5% sprung mass uncertainty is assumed,
namely, mmin = 0.875m0 , mmax = 1.125m0 . The uncertain
parameter sprung mass m0 is initially chosen as mmin , corre-
T
spondingly, Λ(0)[− mkmin m
, − mkmin
l
, − mkmin
n
, − mkmin
c1
, − mkmin
c2
]
and γ(0) mmmin min m1
+m1 . The settings of bioinspired reference
model (11) are chosen as: Mv = 960 kg, δ1 = π6 rad, Sl = 0.1
m, Sr = 0.2 m, ς1 = 50 Ns/m, ς2 = 0.15 Ns/m, k2 = 350 N/m,
k1 = 500 N/m.
In order to analyze and evaluate control effectiveness, the
suspension performances in the following cases are provided.
r Passive: Passive suspension.
r SAP: A classical standard adaptive control method which
is designed based on the suspension sprung mass dynamics
(22)–(23), as stated in Remark 5. The controller structure
is detailed in [18], [21] with parameters settings k1 =
1 × 104 , k2 = 20 in the comparative results. The reference
values of the sprung mass states z0 and ż0 are set to 0.
Fig. 7. Control input and quantitative comparison. (a) Control input. (b) RMS r PAP: The proposed preview adaptive controller is given
of the actuator power P + . (c) RMS of the acceleration z̈0 .
in (19) with ρ1 = 1 × 104 , ρ2 = 20, κ1 = κ2 = κ3 = 1,
ϑ = 1. And the reference trajectories sdr , rdr are obtained
by the bioinspired dynamics (11).
Following the stability analysis in [18], [21], we can obtain Sprung mass acceleration z̈0 comparative results are illus-
V̇ → 0 when t → ∞ by employing LaSalle invariance principle, trated in Fig. 6 to evaluate the vehicle ride comfort where the
correspondingly, e1 and e2 converge to 0 asymptotically. vehicle velocity is chosen as 36 km/h. As shown in Fig. 6(a),
Remark 5: Classic adaptive control methods take the dis- the active suspension controller SAP enables effective vibration
placement z0 and velocity ż0 of the sprung mass m0 as state isolation caused by uneven road excitation. Moreover, the ver-
variables [18], namely, x1 = z0 , x2 = ż0 . The controller u(t) is tical vibration is further suppressed by the PAP employing road
developed based on the dynamics equations of m0 as follows preview information, which means that the ride comfort per-
formance is further enhanced. In particular, the high-frequency
ẋ1 = x2 (22) vibrations are significantly reduced as seen in Fig. 6(b).
m0 ẋ2 = − Fk (z0 , z1 ) − Fc (ż0 , ż1 ) + u(t) (23) The control inputs of SAP and the proposed PAP are re-
ported in Fig. 7(a). Benefiting from the bionic dynamics, PAP
Clearly, (23) do not contain the road excitation which is can achieve better ride comfort with fewer control input re-
isolated by the unsprung mass m1 . This means that it cannot quirements than SAP by exploiting beneficial non-linearities,
be applied for preview suspension control. which contribute to reducing energy consumption. RM SP +
Remark 6: To the best of the authors’ knowledge, the pro- and RM Sz̈0 as performance indexes of energy consumption
posed nonlinear adaptive preview controller has a simple struc- and ride comfort defined in (3) are calculated and plotted in
ture with low computational requirements, which does not rely Fig. 7(b) and (c) for quantitative evaluation. Compared with SAP,
on a simplified linear model or optimization, and is proposed for the RM SP + of PAP is significantly reduced (over 60%) while

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
3932 IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 8, NO. 7, JULY 2023

Fig. 8. Uneven road.

Fig. 10. Planned velocity trajectories subject to C2 generated with various

weighting parameter settings.

acceleration and deceleration that exceed the desired constraints.

The parameters are chosen as κ1 = 10, κ2 = 10, κ3 = 10 to
strike a balance, as depicted in Fig. 9(b). Using the UC method,
the vehicle enters B1 with a sharp deceleration to suppress
Fig. 9. Planned velocity trajectories subject to different constraints. vibration and leaves B1 with a high acceleration to reduce Γ3 .
The unreasonable speed changes are avoided effectively when
RM Sz̈0 is also reduced (over 25%). This means that effective constraints are considered. However, the UC is particularly sus-
energy saving is achieved and ride comfort is further improved ceptible to violating the constraints. With the parameter settings
by the presented PAP. (PS) κ1 = 0.01, κ2 = 10, κ3 = 10 as in Fig. 9(c), the planned
velocities obtained by C1, C2, C3 increase to vmax rapidly while
satisfying the constraints. Note that this allows to set the spatial
B. Preview Control With Constrained Velocity Planning granularity Δx arbitrarily through the proposed constrained
Comparative results are provided in this subsection to prove velocity planning, as the constraints satisfaction means that the
the advantages and effectiveness of the constrained velocity unreasonable velocity changes at a small Δx are avoided.
planning method detailed above, based on the uneven road (UR) An additional advantage of the proposed constrained velocity
comprising bumps (B1 , B2 , B3 ) and random roads. As depicted planning is that the search space is refined as elaborated in Fig. 4.
in Fig. 8, the typical uneven road conditions, such as speed The velocity candidate points and branches that do not satisfy
bumps, manhole covers, and slopes, are modeled by choosing the constraints are removed, which contributes to reduce the
different bumps settings. computational demand. Each planning algorithm is executed
As described in Fig. 4, the initial velocity and target velocity 50 times with computer configurations in [30] and the average
in start location 0 m and end location 2000 m are set to 4 execution times of for UC, C1, C2, and C3 are 22.4 ms 22 ms,
m/s and 5 m/s in this study. The maximum velocity vmax and 21.4 ms, and 17.9 ms. In contrast to UC, the solution times of
minimum velocity vmin are 4 m/s and 24 m/s. The vertical C1, C2, and C3 are reduced by more than 2%, 4%, and 19%,
costs Ci (Bi , v, i ), i = 1, 2, 3 are approximated with 1 = 20, respectively.
2 = 1, 3 = 1, and  = 0.5 in Step 2 . Meanwhile, to generate The comparative planned trajectories illustrated in Fig. 10 are
the velocity candidate set in each adjustment location for bumps, used to analyze the role of the each cost weighting parameter.
initial point, and target point, the velocity increment Δv, spatial From the planned trajectories with PS1 (κ1 = 10, κ2 = 10,
granularity Δx, and speed regulation zone width w are chosen κ3 = 10), PS2 ( κ1 = 1, κ2 = 10, κ3 = 10), PS3 (κ1 = 0.1,
as 1 m/s, 25 m, and 100 m, respectively. κ2 = 10, κ3 = 10) and PS4 (κ1 = 0.01, κ2 = 10, κ3 = 10)
The results in the following cases are presented in Fig. 9 to in Fig. 10(a), it can be seen that the vehicle tends to a higher
analyze the effects of the acceleration constraints: UC (uncon- speed as the vertical cost weighting parameter κ1 decreases. As
strained velocity planning), C1 (amax = 5 m/s2 , dmax = −5 displayed in Fig. 10(b) with PS5 (κ1 = 0.1, κ2 = 0.1, κ3 =
m/s2 ), C2 (amax = 3 m/s2 , dmax = −3 m/s2 ), C3 (amax = 1 0.1), PS6 (κ1 = 0.1, κ2 = 0.1, κ3 = 0.01), and PS7 (κ1 =
m/s2 , dmax = −1 m/s2 ). The planned velocity trajectories gen- 0.1, κ2 = 0.1, κ3 = 0.001), when κ3 decreases, the vehicle
erated by selecting cost weight parameters κ1 = 10, κ2 = 1, attempts to maintain a low speed and avoid speed variation.
κ3 = 1 are shown in Fig. 9(a). It can be observed that the UC The smaller parameter κ3 means that the increased importance
generates unrealistic velocity changes to reduce Γ3 and Γ1 when of ride comfort and longitudinal performance. In particular,
leaving B1 and entering B2 . The changes in speed result in as can be seen from PS7, the vehicle maintains the minimum

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
HUANG et al.: ADAPTIVE BIOINSPIRED PREVIEW SUSPENSION CONTROL WITH CONSTRAINED VELOCITY PLANNING 3933

Fig. 11. Multiple objectives optimization.

Fig. 13. Quantitative comparison for UR with constrained velocity planning.

(a) Control input for UR. (b) RMS of the actuator power P + for UR. (c) RMS
of the acceleration z̈0 for UR.

Fig. 12. Sprung mass acceleration z̈0 for UR with constrained velocity plan-
ning. (a) Time-domain response of z̈0 for UR. (b) Frequency component of z̈0
for UR.

velocity and accelerates at the final stage to reach the target

velocity. Compared to the PS5, the vehicle with PS8 (κ1 = 0.1,
κ2 = 0.01, κ3 = 0.1) tends to accelerate and decelerate rapidly
to reduce vibration and Γ3 . Possible trajectories that consider Γ1 ,
Γ2 and Γ3 simultaneously are plotted as PS9 (κ1 = 10, κ2 = 2,
κ3 = 6) and PS10 (κ1 = 15, κ2 = 10, κ3 = 11). PS9 puts more
Fig. 14. Suspension experimental setup. D/A and A/D denote Digital-to-
emphasis on ride comfort, and PS10 is more balanced. Analog conversion and Analog-to-Digital conversion, respectively.
Further, to reflect the performance trade-offs, the cost distri-
bution of representative velocity trajectories is shown in Fig. 11.
For clarity, the costs are normalized, with 1 denoting the maxi- comfort. Another major advantage the PAP can offer is the small
mum cost of each performance index in all selected trajectories. computational burden. With the above controller parameters
The remaining costs are transformed as the proportion relative and the computer configurations in [30], the execution time of
to the maximum cost. Combined with Fig. 10, the planned PAP is about 11.9 s. The adaptive control method demands low
velocity generation depends on the relative size of the weighting computational resources because no optimization is required.
parameters, and the larger weighting parameters indicates more Moreover, the benefit and enhancement in terms of ride comfort
emphasis on that aspect. Considering the multi-objective per- achieved by PAP are further illustrated through the frequency
formance, the velocity with PS10 is chosen as the final planned components of z̈0 displayed in Fig. 13(b).
trajectory. Then, the spatial road excitation is converted into In addition, from Fig. 13(a), PAP requires less control ampli-
temporal excitation using transformation method in [14] with tude to suppress vibration. An interesting phenomenon observed
the aid of the final velocity trajectory. Comparative simulations in Fig. 13(a) is that the control input of SAP is distinct from PAP.
are carried out based on the passive, SAP, and PAP with the same A key reason for this is that the reference values for SAP are set
controller parameters mentioned above. to 0, whereas the reference trajectory for PAP is provided by
As observed in Fig. 12(a), vehicle vibrations caused by uneven the elaborately designed bionic dynamics (11). The RMS of P +
roads can be effectively isolated by the SAP and PAP. Making use and z̈0 are calculated quantitatively and given in Fig. 13(b) and
of the preview road information, the PAP further improves ride (c). The energy consumption of the PAP is reduced by more than

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
3934 IEEE TRANSACTIONS ON INTELLIGENT VEHICLES, VOL. 8, NO. 7, JULY 2023

TABLE I
BIONIC DYNAMICS PARAMETERS

Fig. 16. Control input and quantitative comparison obtained by experiment.

(a) Control input. (b) RMS of P + . (c) RMS of z̈0 .

48% in comparison with SAP, while the RMS of P + is reduced

by about 89% by introducing bionic dynamics as in Fig. 16(b).

Fig. 15. Sprung mass acceleration z̈0 obtained by experiment. (a) Time-
domain response of z̈0 . (b) Frequency component of z̈0 . V. CONCLUSION
This article presents a novel adaptive preview suspension
57% compared to the SAP. Also, the RM Sz̈0 is reduced by more control method that is not dependent on a simplified linear state
than 51%. space model and does not require optimization for the first time.
Meanwhile, drawing inspiration from the animal limb structure,
C. Experiment Results bioinspired dynamics are used to provide reference trajecto-
ries. Accordingly, the controller has a simple structure, low
The proposed method is experimentally validated using the computational burden, and energy efficiency. Employing road
platform as in Fig. 14. The real-time motion information of the preview information, a constrained velocity planning approach
active suspension is obtained through sensors and data acquisi- is proposed to balance vertical ride comfort, vehicle passage
tion devices. The road excitation z2 generated in the previous time, and longitudinal comfort. Not only is the restriction on
subsection employing the presented constrained velocity plan- the spatial granularity Δx eliminated, but unreasonable velocity
1
ning is reduced to 10 z2 as the road input for this subsection. To changes are avoided. Comparative simulations and experiments
avoid exceeding the limitations of the experimental setup, the demonstrate the effectiveness and advantages of the whole
height of B3 is reduced to 3 cm. The suspension parameters can framework including constrained velocity planning and adaptive
be obtained from [31]. The SAP controller parameters are set preview controller.
as k1 = 80, k2 = 20 with PAP parameters ρ1 = 100, ρ2 = 20, Future work focuses on introducing path planning to bypass
κ1 = κ2 = κ3 = 1, ϑ = 1. The bionic dynamics parameters some severe road bulges or depressions by using preview road
setting is shown in Table I. information in the proposed velocity planning and preview
The time-domain response of z̈0 experimentally obtained in control framework to further balance multi-objective costs and
Fig. 15(a) show that the vibration is suppressed, especially improve ride comfort. Meanwhile, at the control level, actuator
at the bumps. Meanwhile, from the frequency components in failures, dead zones, time delays, and hysteresis can be further
Fig. 15(b), it can be seen that the ride comforts of both SAP and considered. Faster finite time and fixed time convergence are
PAP are enhanced compared to the passive suspension. More- expected, while transient performance and state constraints can
over, the sprung acceleration z̈0 is further reduced by considering be satisfied by employing invertible transformations or barrier
the road preview information in the controller design. functions.
Fig. 16(c) illustrates quantitatively the advantage of the adap-
tive preview suspension control algorithm in terms of ride com-
fort. Compared to passive suspension, the RMS of z̈0 is reduced REFERENCES
by over 18% and 48% for SAP and PAP, respectively. Note that
[1] H. Taghavifar and S. Rakheja, “Multi-objective optimal robust seat sus-
control inputs of PAP in Fig. 16(a) are generated from a bionic pension control of off-road vehicles in the presence of disturbance and
perspective, which differ significantly from SAP because of the parametric uncertainty using metaheuristics,” IEEE Trans. Intell. Veh.,
bio-inspired dynamics. The PAP reduces the RMS of z̈0 by about vol. 5, no. 3, pp. 372–384, Sep. 2020.

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.
HUANG et al.: ADAPTIVE BIOINSPIRED PREVIEW SUSPENSION CONTROL WITH CONSTRAINED VELOCITY PLANNING 3935

[2] J. J. Rath, M. Defoort, C. Sentouh, H. R. Karimi, and K. C. Veluvolu, [23] S. Teng et al., “Motion planning for autonomous driving: The state of
“Output-constrained robust sliding mode based nonlinear active suspen- the art and future perspectives,” IEEE Trans. Intell. Veh., vol. 8, no. 6,
sion control,” IEEE Trans. Ind. Electron., vol. 67, no. 12, pp. 10652–10662, pp. 3692–3711, Jun. 2023.
Dec. 2020. [24] J. F. Medina-Lee, J. Godoy, A. Artuñedo, and J. Villagra, “Speed pro-
[3] H. Li, X. Jing, H.-K. Lam, and P. Shi, “Fuzzy sampled-data control for file generation strategy for efficient merging of automated vehicles on
uncertain vehicle suspension systems,” IEEE Trans. Cybern., vol. 44, no. 7, roundabouts with realistic traffic,” IEEE Trans. Intell. Veh., vol. 8, no. 3,
pp. 1111–1126, Jul. 2014. pp. 2448–2462, Mar. 2023.
[4] H. Pan, X. Chang, and W. Sun, “Multitask knowledge distillation guides [25] J. Zhang, W. Sun, and H. Jing, “Nonlinear robust control of antilock
end-to-end lane detection,” IEEE Trans. Ind. Inform., vol. 19, no. 9, braking systems assisted by active suspensions for automobile,” IEEE
pp. 9703–9712, Sep. 2023. Trans. Control Syst. Technol., vol. 27, no. 3, pp. 1352–1359, May 2019.
[5] L. Chen et al., “Milestones in autonomous driving and intelligent vehicles: [26] H. Pan, D. Zhang, W. Sun, and X. Yu, “Event-triggered adaptive asymptotic
Survey of surveys,” IEEE Trans. Intell. Veh., vol. 8, no. 2, pp. 1046–1056, tracking control of uncertain MIMO nonlinear systems with actuator
Feb. 2023. faults,” IEEE Trans. Cybern., vol. 52, no. 9, pp. 8655–8667, Sep. 2022.
[6] T. Huang, J. Wang, H. Pan, and W. Sun, “Finite-time fault-tolerant in- [27] H. Pan and W. Sun, “Nonlinear output feedback finite-time control for
tegrated motion control for autonomous vehicles with prescribed perfor- vehicle active suspension systems,” IEEE Trans. Ind. Inform., vol. 15,
mance,” IEEE Trans. Transport. Electrific., early access, Dec. 26, 2022, no. 4, pp. 2073–2082, Apr. 2019.
doi: 10.1109/TTE.2022.3232521. [28] X. Jing, “The X-structure/mechanism approach to beneficial nonlinear
[7] X. Wang and W. Sun, “Trajectory tracking of autonomous vehi- design in engineering,” Appl. Math. Mech., vol. 43, pp. 979–1000, 2022.
cle: A differential flatness approach with disturbance-observer-based [29] Z. Wu, X. Jing, J. Bian, F. Li, and R. Allen, “Vibration isolation by ex-
control,” IEEE Trans. Intell. Veh., vol. 8, no. 2, pp. 1368–1379, ploring bio-inspired structural nonlinearity,” Bioinspiration Biomimetics,
Feb. 2023. vol. 10, no. 5, 2015, Art. no. 056015.
[8] H. Pan, Y. Hong, W. Sun, and Y. Jia, “Deep dual-resolution net- [30] T. Huang, H. Pan, W. Sun, and H. Gao, “Sine resistance network-based mo-
works for real-time and accurate semantic segmentation of traffic tion planning approach for autonomous electric vehicles in dynamic envi-
scenes,” IEEE Trans. Intell. Transp. Syst., vol. 24, no. 3, pp. 3448–3460, ronments,” IEEE Trans. Transport. Electrific., vol. 8, no. 2, pp. 2862–2873,
Mar. 2023. Jun. 2022.
[9] F.-Y. Wang, “Parallel control and management for intelligent transporta- [31] M. Zhang, X. Jing, and G. Wang, “Bioinspired nonlinear dynamics-based
tion systems: Concepts, architectures, and applications,” IEEE Trans. adaptive neural network control for vehicle suspension systems with
Intell. Transp. Syst., vol. 11, no. 3, pp. 630–638, Sep. 2010. uncertain/unknown dynamics and input delay,” IEEE Trans. Ind. Electron.,
[10] D. Cao et al., “Future directions of intelligent vehicles: Potentials, possi- vol. 68, no. 12, pp. 12646–12656, Dec. 2021.
bilities, and perspectives,” IEEE Trans. Intell. Veh., vol. 7, no. 1, pp. 7–10,
Mar. 2022.
[11] M. Rahman and G. Rideout, “Using the lead vehicle as preview sensor in
convoy vehicle active suspension control,” Veh. Syst. Dyn., vol. 50, no. 12, Tenglong Huang received the B.E. degree in au-
pp. 1923–1948, 2012. tomation from the Henan University of Technology,
[12] A. Akbari and B. Lohmann, “Output feedback H∞/GH2 preview control Zhengzhou, China, in 2019. He is currently working
of active vehicle suspensions: A comparison study of LQG preview,” Veh. toward the Ph.D. degree with the Research Institute
Syst. Dyn., vol. 48, no. 12, pp. 1475–1494, 2010. of Intelligent Control and Systems, Harbin Institute
[13] M. D. Donahue and J. K. Hedrick, “Implementation of an active suspen- of Technology, Harbin, China. His research interests
sion, preview controller for improved ride comfort,” M.S. thesis, Master include motion planning, vehicle dynamics control,
of Sci. (Plan II). Boston Univ., Univ. of California at Berkeley, Berkeley, adaptive control, fault-tolerant control, and intelligent
CA, USA, 2001. vehicles.
[14] J. Wu, H. Zhou, Z. Liu, and M. Gu, “Ride comfort optimization via speed
planning and preview semi-active suspension control for autonomous
vehicles on uneven roads,” IEEE Trans. Veh. Technol., vol. 69, no. 8,
pp. 8343–8355, Aug. 2020. Jue Wang received the B.S. degree in automation
[15] C. Gohrle, A. Schindler, A. Wagner, and O. Sawodny, “Design
and the M.S. degree in pattern recognition and in-
and vehicle implementation of preview active suspension controllers,”
telligent system from Huaqiao University, Xiamen,
IEEE Trans. Control Syst. Technol., vol. 22, no. 3, pp. 1135–1142,
China, in 2016 and 2019, respectively. She is cur-
May 2014. rently working toward the Ph.D. degree in control
[16] J. Theunissen et al., “Regionless explicit model predictive control of active
science and engineering with the Research Institute
suspension systems with preview,” IEEE Trans. Ind. Electron., vol. 67,
of Intelligent Control and Systems, Harbin Institute
no. 6, pp. 4877–4888, Jun. 2020.
of Technology, Harbin, China. Her research interests
[17] A. Soliman and D. Crolla, “Limited bandwidth active suspension employ- include adaptive control, fault-tolerant control, and
ing wheel base preview,” SAE Tech. Paper, Tech. Rep. 2001-01-1063,
intelligent vehicles.
2001.
[18] W. Sun, H. Gao, and O. Kaynak, “Adaptive backstepping control for
active suspension systems with hard constraints,” IEEE/ASME Trans.
Mechatron., vol. 18, no. 3, pp. 1072–1079, Jun. 2013.
[19] H. Pan, X. Jing, W. Sun, and Z. Li, “Analysis and design of a bioin- Huihui Pan (Senior Member, IEEE) received the
spired vibration sensor system in noisy environment,” IEEE/ASME Trans. Ph.D. degree in control science and engineering from
Mechatron., vol. 23, no. 2, pp. 845–855, Apr. 2018. the Harbin Institute of Technology, Harbin, China, in
[20] X. Jing, Y. Chai, X. Chao, and J. Bian, “In-situ adjustable nonlinear pas- 2017, and the Ph.D. degree in mechanical engineering
sive stiffness using X-shaped mechanisms,” Mech. Syst. Signal Process., from Hong Kong Polytechnic University, Hong Kong,
vol. 170, 2022, Art. no. 108267. in 2018. Since December 2017, he has been with the
[21] H. Pan, X. Jing, W. Sun, and H. Gao, “A bioinspired dynamics-based Research Institute of Intelligent Control and Systems,
adaptive tracking control for nonlinear suspension systems,” IEEE Trans. Harbin Institute of Technology. His research interests
Control Syst. Technol., vol. 26, no. 3, pp. 903–914, May 2018. include nonlinear control, vehicle dynamic control,
[22] F.-Y. Wang et al., “Verification and validation of intelligent vehicles: and intelligent vehicles. Dr. Pan is an Associate Editor
Objectives and efforts from China,” IEEE Trans. Intell. Veh., vol. 7, no. 2, for IEEE TRANSACTIONS ON INTELLIGENT VEHICLES
pp. 164–169, Jun. 2022. and Mechatronics.

Authorized licensed use limited to: GUANGZHOU UNIVERSITY. Downloaded on November 21,2024 at 07:08:11 UTC from IEEE Xplore. Restrictions apply.