International Journal of Control, Automation and Systems 17(12) (2019) 3091-3101 ISSN:1598-6446 eISSN:2005-4092

http://dx.doi.org/10.1007/s12555-018-0828-9 http://www.springer.com/12555

Dual-mode Distributed Model Predictive Control for Platooning of Con-

nected Vehicles with Nonlinear Dynamics
Maode Yan* ■ , Wenrui Ma, Lei Zuo, and Panpan Yang

Abstract: This paper presents a dual-mode distributed model predictive control (DMPC) strategy for the platooning
of connected vehicles with nonlinear dynamics. A third-order nonlinear model is employed to describe the dynamic
characteristics of vehicles. In order to deal with the system nonlinearity and enhance the control precision, a DMPC
based optimization problem is formulated for the vehicle platoon control, in which the nonlinear dynamics and
the input boundaries are both considered as its constraints. Then, a dual-mode structure with the control scheme
from the optimization and a local state feedback controller is proposed to drive the vehicles to the desired platoon.
Comparing with other vehicle platoon algorithms, the proposed dual-mode DMPC strategy can significantly reduce
the computational burden and save communication resources. Furthermore, the iterative feasibility and the stabil-
ity of proposed control system are strictly analyzed. In final, numerical simulations are provided to validate the
effectiveness of the proposed approaches.

Keywords: Distributed model predictive control, dual-mode structure, nonlinear dynamics, vehicle platoon.

1. INTRODUCTION neural adaptive sliding-mode control method is proposed

to guarantee the string stability of vehicle platoon, where
Vehicle platoon has received substantial attention in the neural network is used to approximate the uncertain
recent years [1–4]. It has been regarded as an effective driving resistance. In [20], a distributed implementation
method to overcome the existing traffic problems [5–7], of model predictive control is formulated for multi-vehicle
due to its significant benefits in improving traffic effi- formation stabilization. A DMPC algorithm has been de-
ciency, enhancing road safety and reducing fuel consump- signed to keep the vehicle platoon having a rigid forma-
tions. The objective of vehicle platoon is driving the ve- tion and achieving γ -gain stability in [21]. Among these
hicles in the platoon to maintain a desired inter-vehicle existing methodologies, DMPC is one of classical meth-
distance and track a desired velocity. ods to solve the vehicle platoon problems. While applying
DMPC into the vehicle platoon, it can optimize the current
Vehicle platoon is first studied in the PATH project in
time slot with constraints and anticipate the events to take
1980’s [8]. Since then, various projects for vehicle pla-
the control actions accordingly. Hence, the DMPC based
toon are initiated around the world, such as GCDC in
vehicle platoon algorithms can provide better performance
Netherlands [9], SARTRE in Europe [10] and Energy-ITS
than the other ones and attract increasingly interests.
in Japan [11]. In addition, lots of methods, including slid-
ing mode control [12], consensus-based control [13], neu- Inspired by the advanced features of DMPC, there have
ral network control [14], model predictive control [15, 16] been many researches about vehicle platoon using DMPC
and so forth, have been applied to deal with vehicle pla- [22–26]. For example, the stability and string stability of
toon control problems. For instance, a consensus based the vehicle platoon control under DMPC algorithm are
control scheme is proposed to evaluate the performance of proved in [22]. DMPC is applied for the cooperation of
vehicle platoon under different communication topologies autonomous vehicle teams with linear high-fidelity vehi-
of initial states in [17]. In [18], an adaptive bidirectional cle models, and the collision avoidance is ensured in [23].
platoon control method is proposed to guarantee the con- In [24], the vehicle platoon with the linear double integra-
stant time headway inter-vehicle distance by using a cou- tor vehicles is studied, where both the unconstrained and
pled integrated sliding mode control algorithm. In [19], a constrained cases are considered. In [25], a linear DMPC

Manuscript received November 15, 2018; revised May 29, 2019; accepted June 19, 2019. Recommended by Associate Editor Andrea
Cristofaro under the direction of Editor Myo Taeg Lim. This work was supported by the National Natural Science Foundation of China (No.
61803040), the Science Research Plan in Shaanxi Province of China (Nos. 2018JQ6098, 2019GY-218) and the Fundamental Research Funds
for the Central University of China (Nos. 300102328403, 300102328303).

Maode Yan, Wenrui Ma, Lei Zuo, and Panpan Yang are with the School of Electronic and Control Engineering, Chang’an University, Xi’an,
P. R. China (e-mails: {mdyan, l_zuo, panpanyang}@chd.edu.cn, [email protected]).
* Corresponding author.

©ICROS, KIEE and Springer 2019

3092 Maode Yan, Wenrui Ma, Lei Zuo, and Panpan Yang

scheme has been applied for cooperative multi-vehicle In addition, the stability of the vehicle platoon con-
systems with communication delays. A DMPC algorithm trol system is demonstrated through Lyapunov stabil-
is presented for heterogeneous vehicle platoon with unidi- ity theory.
rectional topologies in [26]. Although the existing DMPC
based methods show great performance in vehicle platoon The remainders of this paper are organized as follows:
control, the above results mainly consider the vehicle dy- In Section 2, preliminaries and problem formulation are
namics as linear systems, which are inadequate to describe presented. Section 3 introduces the dual-mode DMPC
the dynamic characteristics of vehicles. Hence, it is neces- based vehicle platoon algorithm. Section 4 analyzes the
sary to apply the nonlinear vehicle dynamics into vehicle feasibility and stability of the proposed control system.
platoon control. The numerical simulations are provided in Section 5. Sec-
There are some researches taking the nonlinear vehicle tion 6 concludes this paper and offers future work.
models into consideration. In [27], a robust H∞ output- Notation: R stands for the set of real numbers. Rn
feedback controller is designed to achieve yaw stabiliza- stands for the n-dimension real space. Given a matrix M,
tion for the vehicle dynamics with differential steering. In M > 0 (M ≥ 0) means the matrix is positive definite (pos-
[28], the nonlinear model according to the vehicle lateral itive semi-definite). The relation M1 ≥ M2 for symmetric
dynamics and a vision system, which is represented by an matrices means that M1 −M2 ≥ 0. For a given column vec-
uncertain T-S fuzzy model, is applied in the lane depar- tor v, ∥v∥ represents the Euclidean norm. The P-weighted
∆ √
ture method. Vehicle suspension systems with network- norm is defined as ∥v∥ p = vT Pv, where P is a given ma-
induced delays are studied and an event-triggered H∞ trix with appropriate dimension. Given matrix Q, λ (Q)
output-feedback control algorithm is proposed to reduce and λ̄ (Q) represent the minimum and maximum of the
the communication burden in [29]. In this paper, the third- absolute values of the eigenvalues for Q.
order nonlinear vehicle dynamics with input boundaries is
considered in a DMPC based vehicle platoon algorithm. 2. PRELIMINARIES AND PROBLEM
Moreover, there are few studies considering the compu- FORMULATION
tational load of DMPC and the communication resources
required in vehicle platoon control. Hence, how to effec- 2.1. Vehicle dynamics
tively apply the DMPC into the vehicle platoon is still an Consider a vehicle platoon consisted of Na + 1 vehicles,
open problem. Motivated by this fact, a dual-mode DMPC where the leader is denoted as 0, and the followers are
strategy for platooning of connected vehicles with non- indexed from 1 to Na .
linear dynamics is presented to reduce the computational For each follower i, its dynamics is described as [30,
burden and save the communication resources. 31]:
The main contributions of this paper are three-fold.
• The third-order nonlinear vehicle dynamics is em- ẋi (t) = fi (xi (t), ui (t)), (1)
ployed to fully describe vehicles. The model can
make the proposed control scheme more practical where xi (t) = [pi (t), vi (t), ai (t)]T ∈ R3 is the state of fol-
comparing to the existing results with linear models lower i; ui (t) ∈ R is the control input. The function
or lower order ones. The input boundaries subject to fi (xi (t), ui (t)) ∈ R3 is shown by
physical limitations of actuators are taken into con-  
vi (t)
sideration to further improve the availability.
fi (xi (t), ui (t)) =  ai (t) , (2)
• A dual-mode DMPC based platoon algorithm is pro- gi (vi (t), ai (t)) + hi ui (t)
posed for the vehicle platoon control. In this dual-
mode strategy, a distributed model predictive con- where
troller and a local state feedback controller are devel- ( )
oped to achieve better performance, reduce the com- 1 σ ĀiCdi 2 dmi
gi (vi (t), ai (t)) = − ai (t) + vi (t) +
putational burden and save the communication re- ςi 2mi mi
sources. Comparing with the other vehicle platoon al- σ ĀiCdi vi (t)ai (t)
gorithms using DMPC, the efficiency of the proposed − ;
mi
approach is greatly improved and it can be used in
more applications. hi = ςi1mi ; pi (t), vi (t) and ai (t) denote the position, veloc-
• The feasibility and stability of the proposed dual- ity and acceleration of the follower i respectively; ςi is the
mode DMPC strategy are rigorously analyzed. By it- time constant of the motor; mi is the mass of the ith ve-
eratively ensuring the control input constraint and the hicle; σ is the mass density of air; Āi is the cross section
terminal constraint in each optimization problem, the area of the ith vehicle; Cdi is coefficient of drag; dmi is the
feasibility of the proposed control scheme is proven. mechanical resistance.
 Dual-mode Distributed Model Predictive Control for Platooning of Connected Vehicles with Nonlinear Dynamics 3093

2.2. Communication topology Denoting the error as x̃i = xi − xi,des , we obtain

The communication topology among vehicles is de-
scribed as a directed graph G = {V, E}, where V = x̃˙i (t) = Ai x̃i (t) + Bi (ui (t) − u0 ) = Ai x̃i (t) + Bi ūi (t),
{0, 1, 2, ..., Na } is the set of nodes, and E ⊂ V × V is the (6)
collection of all directed edges between two connected
nodes. Let A = [ai j ] ∈ R(Na +1)×(Na +1) be the adjacency ma- where x̃i = [pi + id0 − p0 , vi − v0 , ai − a0 ]T ; ūi (t) = ui (t) −
trix, which is used to describe the directional communi- u0 .
cation weight between two vehicles. If there is a directed Assumption 3: For the linearized system in (6), there
edge from node j to i, then ai j > 0; otherwise ai j = 0, i, j ∈ exists a state feedback control law ūi (t) = Ki x̃i (t), such
∆
{0, 1, 2, .., Na }. Denote Ni = { j|ai j > 0, j ∈ {1, 2, .., Na }} that the closed-loop system is stable, i.e., Aci = Ai + Bi Ki is
as the set of neighbor followers of ith follower. Similarly, stable. It can be obtained that ui (t) = Ki x̃i (t) + u0 accord-
define Ci = { j|a ji > 0, j ∈ {1, 2, .., Na }}, which means that ing to (6).
the ith follower can send its information to the jth follower Lemma 1: Suppose that Assumption 3 holds for each
in the set. follower i. There exist a constant εi > 0 and a matrix Pi > 0
Assumption 1: There should be a spanning tree in the ∆
for follower i, such that: the set Ωi (εi ) = {x̃i : Vi (x̃i ) ≤ εi2 }
graph G, where the leader is the root node. is a control invariant set with the control law: ui (t) = u0 +
Ki x̃i (t), i.e., any initial state x̃i (0) ∈ Ωi (εi ) implies x̃i (t) ∈
2.3. Problem formulation ∆
Ωi (εi ), and ui (t) = u0 + Ki x̃i (t) ∈ Ui , where Vi (x̃i (t)) =
As it is known, the objective of vehicle platoon is to
∥x̃i (t)∥Pi . Furthermore, for any x̃i (t) ∈ Ωi (εi ), we have
2
drive the vehicles into the desired platoon while following
the given references. That is V̇i (x̃i (t)) ≤ − ∥x̃i (t)∥2Q∗i and Q∗i = Qi + KiT Ri Ki , where Ki is
a state feedback control gain according to Assumption 3.

 lim ∥vi (t) − v0 (t)∥ = 0, Qi , Ri are positive definite matrices.
t→∞
i = 1, ..., Na , (3)
 lim ∥pi−1 (t) − pi (t) − d0 ∥ = 0, Proof: For any given symmetric matrix Qoi , there exists
t→∞
Pi > 0 satisfying the following Lyapunov equation
where d0 is the desired distance between two consecutive
vehicles. (Aci )T Pi + Pi Aci = −(Q∗i + Qoi ). (7)
In order to improve the control performance, we intend
to develop a novel DMPC based tracking algorithm such We obtain the closed-loop nonlinear error system of (1) is
that vehicles can asymptotically converge to the desired shown as
platoon from arbitrary initial positions.  
v0
Assumption 2: Assume the leader has a constant ve- x̃˙i (t) = fi (xi (t), Ki x̃i (t) + u0 ) −  a0 
locity. Then, according to (2), its input can be presented ȧ0
as
= fi∗ (x̃i (t), Ki x̃i (t) + u0 ).
σ A0Cd0 v20
u0 = + dm0 . (4) ∆
2 Denote ϕ (x̃i (t)) = fi∗ − Aci x̃i (t) and substitute it into
V̇i (x̃i (t)), it follows that
3. DUAL-MODE DMPC ALGORITHM
V̇i (x̃i (t)) =x̃iT (t)[(Aci )T Pi + Pi Aci ]x̃i (t)
Since the dual-mode structure requires a local state
+ 2ϕ T (x̃i (t))Pi x̃i (t). (8)
feedback law in the terminal set [32, 33], we linearize the
vehicle’s nonlinear dynamics at its desired state as follows Invoking (7) and (8), it yields that
ẋi (t) = Ai xi (t) + Bi ui (t), (5)
V̇i (x̃i (t)) = − x̃iT (t)(Q∗i + Qoi )x̃i (t) + 2ϕ T (x̃i (t))Pi x̃i (t)
[ ]
where − 12 o − 21 ∥ϕ (x̃i (t))∥Pi
  ≤ −λ (Pi Qi Pi ) + 2
0 1 0 ∥x̃i (t)∥Pi
  ×Vi (x̃i (t)) − x̃iT (t)Q∗i x̃i (t).
Ai =  0 0 1 ,
σ Ā C v σ ĀiCdi v0
0 − ςii mdii 0 − ς1i − mi
  Since
0
Bi =  0  . ∥ϕ (x̃i (t))∥Pi
lim = 0,
1
ςi mi ∥x̃i (t)∥P →0
i
∥x̃i (t)∥Pi
3094 Maode Yan, Wenrui Ma, Lei Zuo, and Panpan Yang

there exists a constant ε̄i > 0, such that any x̃i (t) ∈ Ωi (ε̄i ) Dual-mode structure:
satisfies For each follower i, when the follower states are outside
the terminal set Ωi (εi ), the DMPC input signal ui = û∗i is
∥ϕ (x̃i (t))∥Pi 1
≤ λ (Pi−0.5 Qoi Pi−0.5 ). (9) applied by optimizing Problem 1; when the follower states
∥x̃i (t)∥Pi 2 enter the terminal set Ωi (εi ), the stabilizing state feedback
law ui (t) = Ki x̃i (t) + u0 is then applied.
Therefore, if we choose εi ∈ (0, ε̄i ), it follows that
V̇i (x̃i (t)) ≤ −x̃iT (t)Q∗i x̃i (t). Hence the set Ωi (εi ) is an in- Dual-mode DMPC based vehicle platoon algorithm:
variant set with the control law: ui (t) = u0 + Ki x̃i (t). □ Requirements:
Sampling period δ ; control horizon T ; initial states
Assumption 4: For all the followers, it is assumed that of all vehicles xi (0), i = 0, ..., Na ; assumed state trajec-
each follower has the same constant prediction horizon tory xi0 (t), assumed state trajectories of neighbors x0j (t),
T ∈ (0, ∞) and sampling period δ ∈ (0, T ], and the syn- t ∈ [0, T ], j ∈ Ni ; k = 0.
chronization is ensured. Denote the same time instants as Initialization:
tk , k = 0, 1, ..., where t0 = 0 and tk+1 = tk + δ . xi (t0 ) = xi (0), x̂ia (s;t0 ) = xi0 (s), x̂aj (s;t0 ) = x0j (s), s ∈
On this basis, the vehicle platoon using DMPC can be [0, T ].
described in the following problem associated with each Iteration:
follower i. For each follower i:
Problem 1: While (xi (tk ) − xi,des (tk )) ∈ / Ωi (εi ) do
1) Receive its neighbors’ assumed state trajectories
min Ji (x̂i (s;tk ), ûi (s;tk ), x̂ia (s;tk ), x̂−i
a
(s;tk )), (10) x̂aj (s;tk ), j ∈ Ni , s ∈ [tk ,tk + T ]. According to x̂ia (s;tk ),
ûi (s;tk )
x̂aj (s;tk ), solve Problem 1 to yield optimal control se-
subject to: quence û∗i (s;tk ) and state sequence x̂i∗ (s;tk ), s ∈ [tk ,tk + T ].
2) If (x̂i∗ (to ;tk ) − xi,des (to )) ∈ Ωi (εi ), to ∈ [tk ,tk + δ ]
x̂˙i (s;tk ) = fi (x̂i (s;tk ), ûi (s;tk )), s ∈ [tk ,tk + T ], holds, then generate the control input ui (s;tk ), s ∈ [tk ,tk +
x̂˙ia (s;tk ) = fi (x̂ia (s;tk ), ûai (s;tk )), s ∈ [tk ,tk + T ], δ ] as:
x̂˙aj (s;tk ) = f j (x̂aj (s;tk ), ûaj (s;tk )), s ∈ [tk ,tk + T ], {
û∗i (s;tk ), for s ∈ [tk ,to ],
ûi (s;tk ) ∈ Ui , s ∈ [tk ,tk + T ], ui (s;tk ) = (12)
Ki x̄˜i (s;tk )+u0 , for s ∈ [to ,tk +δ ],
∥x̂i (tk + T ;tk ) − xi,des (tk + T )∥2Pi ≤ εi2 , (11)
where x̄˜i (s;tk ) = x̄i (s;tk ) − xi,des (s). And design the as-
where sumed control input ûai (s;tk+1 ), s ∈ [tk+1 ,tk+1 + T ] as

Ji (x̂i (s;tk ), ûi (s;tk ), x̂ia (s;tk ), x̂−i

a
(s;tk )) ûai (s;tk+1 ) = Ki x̄˜i (s;tk ) + u0 , (13)
∫ tk +T
= ∥x̂i (s;tk ) − xi,des (s)∥2Qi + ∥ûi (s;tk ) − u0 ∥2Ri where x̄i (s;tk ), s ∈ [to ,tk+1 + T ], is determined by the sys-
tk
tem x̄˙i (s;tk ) = fi (x̄i (s;tk ), Ki x̄˜i (s;tk ) + u0 ), with initial state
+ ∥x̂i (s;tk ) − x̂ia (s;tk )∥Fi
x̄i (to ;tk ) = x̂i∗ (to ;tk ).
+ ∑ ai j x̂i (s;tk ) − x̂aj (s;tk ) − di, j G ds 3) If (x̂i∗ (s;tk ) − xi,des (s)) ∈ / Ωi (εi ), s ∈ [tk ,tk + δ ], gen-
i
j∈Ni erate the control input as ui (s;tk ) = û∗i (s;tk ), s ∈ [tk ,tk + δ ]
+ ∥x̂i (tk + T ;tk ) − xi,des (tk + T )∥2Pi ; and then design the assumed control input ûai (s;tk+1 ), s ∈
[tk+1 ,tk+1 + T ] as
x̂i (tk ;tk ) = xi (tk ); x̂ia (tk ;tk ) = x̂i∗ (tk ;tk−1 ); x̂aj (tk ;tk ) =
x̂∗j (tk ;tk−1 ); εi is the constant determined in Lemma 1; (11) ûai (s;tk+1 )
is the terminal constraint. The assumed control trajecto- {
û∗i (s;tk ), if s ∈ [tk+1 ,tk +T ],
ries for the follower i and its neighbors j are generated as = (14)
follows: Ki x̃i (s;tk+1 )+u0 , if s ∈ [tk +T,tk+1 +T ].
a

{
a
û∗i (s;tk−1 ), if s ∈ [tk ,tk−1 +T ], End if
ûi (s;tk ) = 4) Generate the assumed state trajectory x̂ia (s;tk+1 ),
Ki x̃i (s;tk )+u0 , if s ∈ [tk−1 +T,tk +T ],
a
{ ∗ s ∈ [tk+1 ,tk+1 + T ] by ûai (s;tk+1 ), with x̂ia (tk+1 ;tk+1 ) =
û j (s;tk−1 ), if s ∈ [tk ,tk−1 +T ], x̂i∗ (tk+1 ;tk ). Receive the leading information x0 (s).
ûaj (s;tk ) =
K j x̃aj (s;tk )+u0 , if s ∈ [tk−1 +T,tk +T ], 5) Apply the control input ui (s;tk ), s ∈ [tk ,tk + δ ].
6) k = k + 1.
where x̃ia (s;tk ) = x̂ia (s;tk ) − xi,des (s) and x̃aj (s;tk ) = End while
x̂aj (s;tk ) − x j,des (s). 7) Apply the control input ui (tk ) = Ki x̃i (tk ) + u0 .
 Dual-mode Distributed Model Predictive Control for Platooning of Connected Vehicles with Nonlinear Dynamics 3095

4. FEASIBILITY AND STABILITY ANALYSES The feasible state trajectory candidate satisfies the termi-
nal constraint.
In this section, the feasibility and stability of the dual- In conclusion, the feasibility of tk+1 is proved. Namely,
mode DMPC based vehicle platoon algorithm will be an- Problem 1 is feasible for all k ≥ 0. □
alyzed.
When the follower states enter the terminal set, the lo-
4.1. Feasibility analysis cal feedback control input ui (t) = u0 + Ki x̃i (t) is applied
When the follower states are out of the terminal set, according to Lemma 1.
Problem 1 needs to be feasible at the initial time instant According to the above analysis, dual-mode DMPC
t0 , i.e., there exists a control trajectory driving the state strategy for vehicle platoon control is feasible.
trajectories to satisfy all the constraints.
Assumption 5: For each follower i, at time t0 = 0, with 4.2. Stability analysis
the initial state xi (0), there exists a prediction horizon T ∈ This subsection presents the stability conditions of the
(0, ∞) such that Problem 1 has a solution. proposed dual-mode DMPC strategy. The stability anal-
ysis is divided into two steps. In the first step, when
Theorem 1: Suppose that Assumptions 3-5 hold for
the follower states are outside the terminal set, the sum
each follower i. Then at any tk > 0, Problem 1 always
of optimal control objective functions of all the followers
has a feasible solution.
will prove to be an appropriate Lyapunov function under
Proof: In order to prove this theorem, we need to prove certain conditions. In the second step, when the follower
that if Problem 1 is feasible at time tk , then at time tk+1 , it states enter the terminal set, the local Lyapunov function
can find a feasible control trajectory for this problem. in Lemma 1 will be used.
As mentioned before, let û∗i (s;tk ), s ∈ [tk ,tk + T ] be the When the follower states are outside the terminal set,
feasible solution of Problem 1 at time tk . At time tk+1 , the apply a feasibility control trajectory candidate to the ob-
feasibility control trajectory candidate ûi (s;tk+1 ) is con- jective function. Then we have
structed as follows
⌢ ⌢
Ji (xi (s;tk+1 ), ui (s;tk+1 ), x̂ia (s;tk+1 ), x̂−i
a
(s;tk+1 ))
ûi (s;tk+1 )
{ ≥ Ji (x̂i∗ (s;tk+1 ), û∗i (s;tk+1 ), x̂ia (s;tk+1 ), x̂−i
a
(s;tk+1 )),
û∗i (s;tk ), if s ∈ [tk+1 ,tk + T ], (19)
= (15)
Ki x̂i (s;tk+1 )+u0 , if s ∈ [tk +T,tk+1 +T ],
˜
⌢ ⌢
where ui (s;tk+1 ) = ûai (s;tk+1 ) and xi (s;tk+1 ) = x̂ia (s;tk+1 ),
where x̂˜i (s;tk+1 ) = x̂i (s;tk+1 ) − xi,des (s). And the corre- for s ∈ [tk+1 ,tk+1 + T ].
sponding feasible state trajectory candidate will be gen- According to (19), we obtain
erated as:
Ji (x̂i∗ (s;tk+1 ), û∗i (s;tk+1 ), x̂ia (s;tk+1 ), x̂−i
a
(s;tk+1 ))
x̂˙i (s;tk+1 ) = fi (x̂i (s;tk+1 ), ûi (s;tk+1 )), − Ji (x̂i∗ (s;tk ), û∗i (s;tk ), x̂ia (s;tk ), x̂−i
a
(s;tk ))
s ∈ [tk+1 ,tk+1 + T ], (16) ≤ Ji (xi (s;tk+1 ), ui (s;tk+1 ), x̂ia (s;tk+1 ), x̂−i
⌢ ⌢ a
(s;tk+1 ))

where the initial state is given as x̂i (tk+1 ;tk+1 ) = xi (tk+1 ). − Ji (x̂i∗ (s;tk ), û∗i (s;tk ), x̂ia (s;tk ), x̂−i
a
(s;tk )). (20)
According to Lemma 1, we have x̂˜i (tk + T ;tk+1 ) ∈ Ωi (εi ),
We define that:
so it results in ûi (s;tk+1 ) = Ki x̂˜i (s;tk+1 ) + u0 ∈ Ui , s ∈ [tk +
T,tk+1 + T ]. Combining with û∗i (s;tk ) ∈ Ui , s ∈ [tk+1 ,tk + ∆(Ji )
T ], the feasible control trajectory candidate satisfies the ∆ ⌢ ⌢
control constraint = Ji (xi (s;tk+1 ), ui (s;tk+1 ), x̂ia (s;tk+1 ), x̂−i
a
(s;tk+1 ))
− Ji (x̂i∗ (s;tk ), û∗i (s;tk ), x̂ia (s;tk ), x̂−i
a
(s;tk )). (21)
ûi (s;tk+1 ) ∈ Ui , s ∈ [tk+1 ,tk+1 + T ]. (17)
7
At time tk + T , it can be obtained that x̂i (tk + T ;tk+1 ) = Splitting the terms in ∆(Ji ), we obtain ∆(Ji ) = ∑ ∆i .
i=1
x̂i∗ (tk + T ;tk ). According to the terminal constraint, we
∫ 2
have ⌢
tk +T

∆1 = xi (s;tk+1 ) − xi,des (s)
tk+1 Qi
∥x̂i (tk + T ;tk+1 ) − xi,des (tk + T )∥2Pi ≤ εi2 . (18) 2
⌢
+ ui (s;tk+1 ) − u0
∆ Ri
Since the set Ωi (εi ) = {x̃i : Vi (x̃i ) ≤ εi2 } is a control in- ∗
− ∥x̂i (s;tk ) − xi,des (s)∥2Qi
variant set with the control law ui (t) = u0 + Ki x̃i (t) ∈ Ui ,
it follows that ∥x̂i (tk+1 + T ;tk+1 ) − xi,des (tk+1 + T )∥2Pi ≤ εi2 . − ∥û∗i (s;tk ) − u0 ∥2Ri ds
3096 Maode Yan, Wenrui Ma, Lei Zuo, and Panpan Yang

=0, (22) and

and ∆5
∫ ∫
tk+1 +T 2 ⌢
tk+1 +T
∆2 =
⌢
xi (s;tk+1 ) − xi,des (s)
= ai j
tk +T
∑ xi (s;tk+1 ) − x̂aj (s;tk+1 ) − di, j ds
Gi
tk +T Qi j∈Ni
2 ∫ [
tk+1 +T
⌢ ⌢
+ ui (s;tk+1 ) − u0 ds
Ri
≤ ai j ∑ xi (s;tk+1 ) − xi,des (s)
Gi
∫ tk+1 +T j∈Ni tk +T
2 ]
⌢ a
= xi (s;tk+1 ) − xi,des (s)
tk +T Qi + x̂ j (s;tk+1 ) − x j,des (s) G ds
2 [
i
]
⌢ ∫ tk+1 +T
+ Ki (xi (s;tk+1 ) − xi,des (s)) + u0 − u0 ds λ̄ (G0.5 ) λ̄ (G 0.5
)
∫ tk+1 +T 2
Ri
≤ ai j ∑ i
εi + i
ε j ds
⌢ j∈Ni tk +T λ (Pi0.5 ) λ (Pj0.5 )
= xi (s;tk+1 ) − xi,des (s) [ ]
tk +T
2
Qi
λ̄ (G0.5
i ) λ̄ (G0.5
i )
⌢ = ai j ∑ δ εi + εj . (28)
+ Ki (xi (s;tk+1 ) − xi,des (s)) ds j∈Ni λ (Pi0.5 ) λ (Pj0.5 )
Ri
∫ tk+1 +T 2
⌢ Since for all s ∈ [tk ,tk+1 ], (xi∗ (s;tk ) − xi,des (s)) ∈
/ Ωi (εi ), we
= xi (s;tk+1 ) − xi,des (s) ∗ ds. (23)
tk +T Qi get
∫
We have (xi (s;tk+1 ) − xi,des (s)) ∈ Ωi (εi ), s ∈ [tk + T,tk+1 + tk+1
⌢

∆6 = − ∥x̂i∗ (s;tk ) − xi,des (s)∥Q

2
ds
T ]. From Lemma 1, ∆2 satisfies tk
i

∫
∫ λ (Qi ) tk+1 ∗
tk+1 +T ⌢
dVi (xi (s;tk+1 )) ≤− ∥x̂i (s;tk ) − xi,des (s)∥2Pi ds
∆2 ≤ − ds λ̄ (Pi ) tk
ds

tk +T
2 λ (Qi ) 2
⌢ ≤− δ εi . (29)
≤ xi (tk + T ;tk+1 ) − xi,des (tk + T ) λ̄ (Pi )
Pi
2
⌢ And the remaining term is
− xi (tk+1 + T ;tk+1 ) − xi,des (tk+1 + T )
Pi ∫ tk+1
= ∥x̂i∗ (tk + T ;tk ) − xi,des (tk + T )∥2Pi ∆7 = − ∥û∗i (s;tk ) − u0 ∥2Ri ds
2 tk
∫ tk+1
⌢ ∗
− xi (tk+1 + T ;tk+1 ) − xi,des (tk+1 + T ) , (24) x̂i (s;tk ) − x̂aj (s;tk ) − di, j ds
Pi −
tk
ai j ∑ G
i
j∈Ni
and we have ∫ tk+1
2 − ∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥Fi ds
⌢ tk
∆3 = xi (tk+1 + T ;tk+1 ) − xi,des (tk+1 + T ) ∫ tk +T
Pi
− ∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥Fi ds. (30)
− ∥x̂i∗ (tk + T ;tk ) − xi,des (tk + T )∥2Pi . (25) tk+1

Theorem 2: For all the followers, suppose that: As-

Then, it follows that
sumptions 1-5 hold, Fi > a ji ∑ G j , i = 1, ..., Na and
j∈Ci
∆2 + ∆3 ≤ 0. (26)
[ ]
λ (Qi ) 2 λ̄ (G0.5 ) λ̄ (G0.5
)
Based on the triangle inequality, we have εi ≥ ai j ∑ i
εi + i
εj , (31)
λ̄ (Pi ) j∈Ni λ (Pi ) λ (Pj0.5 )
0.5
∫
tk +T
⌢
∆4 =ai j
tk+1
∑ xi (s;tk+1 ) − x̂aj (s;tk+1 ) − di, j
Gi the sum of all the followers’ objective functions is strictly
j∈Ni
∗ monotonically decreasing. Then the states of each fol-
− ∑ x̂i (s;tk ) − x̂aj (s;tk ) − di, j ds
G i lower will enter the terminal set Ωi (εi ) finally.
j∈Ni
∫ ∗
tk +T Proof: According to the above analyses, we obtain
=ai j ∑ x̂i (s;tk ) − x̂∗j (s;tk ) − di, j
G i
tk+1
∗
j∈Ni
∆(Ji )
− ∑ x̂i (s;tk ) − x̂aj (s;tk ) − di, j ds ∫ tk+1
∥û∗i (s;tk ) − u0 ∥2Ri ds
G
j∈Ni
i
≤−
∫ ∗
tk
tk +T ∫ ∗
≤ai j ∑ x̂ j (s;tk ) − x̂aj (s;tk ) ds, (27) tk+1
x̂i (s;tk ) − x̂aj (s;tk ) − di, j ds
tk+1 j∈Ni
G i −
tk
ai j ∑ G
i
j∈Ni
 Dual-mode Distributed Model Predictive Control for Platooning of Connected Vehicles with Nonlinear Dynamics 3097

∫ tk+1
− ∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥Fi ds ≤ ∆(J)
tk
∫ < 0.
tk +T ∗
+ ai j ∑ x̂ j (s;tk ) − x̂aj (s;tk ) ds
G i It can be concluded that the states of each follower i with
tk+1 j∈Ni
∫ tk +T the initial states outside Ωi (εi ) will enter the terminal set
− ∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥Fi ds. (32) Ωi (εi ) finally. □
tk+1

Na When the follower states enter the terminal set, the sta-
Then, ∆(J) = ∑ ∆(Ji ) is obtained bility of vehicle platoon control is proved by the local Lya-
i=1
punov function Vi (x̃i (t)) = ∥x̃i (t)∥2Pi . According to Lemma
∆(J) 1, we have
Na [∫
V̇i (x̃i (t)) ≤ −x̃iT (t)Q∗i x̃i (t).
tk+1
≤ −∑ ∥û∗i (s;tk ) − u0 ∥2Ri ds (36)
i=1 tk
∫ tk+1 ∗ By combining the two parts including outside the ter-
+ ai j ∑ x̂i (s;tk ) − x̂aj (s;tk ) − di, j ds minal set and in the terminal set, we prove the stability of
G i
tk j∈Ni
∫ ] the dual-mode DMPC strategy in vehicle platoon control.
tk+1
+ ∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥Fi ds
tk 5. SIMULATION RESULTS
[ ∫
Na tk +T ∗
+ ∑ ai j ∑ x̂ j (s;tk ) − x̂aj (s;tk ) ds In order to verify the developed results, the numerical
G
i
i=1 tk+1 j∈Ni
∫ ] simulations for the proposed dual-mode DMPC are pro-
tk +T
∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥Fi ds , vided by considering vehicle platoon with 6 vehicles (in-
− (33)
tk+1 cluding one leader and five followers). Index the vehicles
as 0,...,5, where 0 denotes the leader and 1,...,5 are the
where
followers. Vehicles are interconnected as the predecessor-
[ ∫
Na tk +T ∗ leader following communication topology, i.e., each fol-
∑ ai j ∑ x̂ j (s;tk ) − x̂aj (s;tk ) ds
tk+1
G i lower can communicate with its predecessor and leader
i=1 j∈Ni
∫ tk +T
] except the follower 1, which only communicates with the
− ∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥Fi ds leader.
tk+1 Related system parameters are chosen as follows: mi =
∫ tk +T Na 2 kg; ςi = 0.5; σ ĀiCdi = 2; dmi = 2. And the control input
= a ji ∑ ∑ ∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥G j is bounded as ui (t) ∈ [−100, 100].
tk+1 i=1 j∈Ci
In the DMPC, the parameters of its objective function
Na
− ∑ ∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥Fi ds. (34) are set as Qi = 5I; Ri = 0.1; Gi = 0.1I; Fi = I and ai j =
i=1 0.05. The local state feedback control gain is given by
Ki = [−6, −1, 1]. According to the Lemma 1, the matrix
According to Fi > a ji ∑ G j , we have Pi is determined as
j∈Ci
 
∆(J) 12.78 −4.35 −5.34
Na [∫ Pi =  −4.35 4.74 −2.60  .
tk+1
< −∑ ∥û∗i (s;tk ) − u0 ∥2Ri ds −5.34 −2.60 10.54
i=1 tk
∫ tk+1 ∗ According to Theorem 2, the prediction horizon is given
+ ai j ∑ x̂i (s;tk ) − x̂aj (s;tk ) − di, j ds as T = 1 s, and the time interval is δ = 0.2 s. The terminal
G i
tk
∫
j∈Ni
] set is determined as εi = 0.3. The desired spacing is set as
tk+1
+ ∥x̂i∗ (s;tk ) − x̂ia (s;tk )∥Fi ds d0 = 5 m. The initial states of the leader are set as p0 (0) =
tk 25 m, v0 (0) = 5 m/s, and a0 (0) = 0 m/s2 . The initial states
< 0. (35) of the followers are set as: x1 (0) = [22, 6, 1]T ; x2 (0) =
[17, 7, 0]T ; x3 (0) = [12, 3, 2]T ; x4 (0) = [3, 6, 1]T ; x5 (0) =
Finally it can be obtained that [0, 2, 1]T . The control input of the leader is u0 = 27, which
Na [
is obtained according to (4).
∑ Ji (x̂i∗ (s;tk+1 ), û∗i (s;tk+1 ), x̂ia (s;tk+1 ), x̂−i
a
(s;tk+1 )) Based on these parameters, the simulation results are as
i=1 follows:
]
The states (including position, velocity and accelera-
− Ji (x̂i∗ (s;tk ), û∗i (s;tk ), x̂ia (s;tk ), x̂−i
a
(s;tk )) tion) and the related errors between real and desired values
3098 Maode Yan, Wenrui Ma, Lei Zuo, and Panpan Yang

60 100
F1
Positions (m)

80 F2
40 F3
60 F4
F5
20 40
L0 F3
F1 F4
F2 F5 20
0
0 1 2 3 4 5 6

U
0
Time (s)
-20
Tracking errors (m)

1 -40

0 -60
-1
F3 -80
F1
-2 F4
F2 -100
F5
-3 0 1 2 3 4 5 6
0 1 2 3 4 5 6 Time (s)
Time (s)
Fig. 4. Control inputs u.
Fig. 1. Trajectories and tracking errors of vehicles.

8 for followers are illustrated in Figs. 1-3, respectively. The

Velocities (m/s)

control inputs for all followers are shown in Fig. 4.

6
From Fig. 1, it can be seen that followers track the
4
leader and the tracking errors converge to 0 at about 5.8 s.
L0 F3
2 F1 F4 Meanwhile, there is no collision between any consecutive
F2 F5
0
vehicles during this process. The vehicles can form the
0 1 2 3 4 5 6 desired formation. Fig. 2 shows that velocities of follow-
Time (s)
ers will converge to the leader’s velocity (5 m/s), which
Velocity errors (m/s)

4 means that all followers track the velocity of leader. The

2 convergence time of the velocity errors is about 3.2 s. The
acceleration curves of all vehicles converge to 0 at about
0
F3
2.8 s in Fig. 3. The vehicles in platoon can run with a con-
F1
-2
F2
F4 stant velocity finally.
F5
-4 From Figs. 1-3, it can be obtained that all vehicles sat-
0 1 2 3 4 5 6
isfy the control objective in (3). Vehicle platoon is sta-
Time (s)
ble and reaches the desired formation under the dual-mode
Fig. 2. Velocities and velocity errors of vehicles. DMPC strategy.
Fig. 4 illustrates that the control inputs are bounded and
converge to u0 . The figure reveals that the control input
20 constraints are satisfied and the feasibility of the proposed
Accelerations (m/s2)

strategy is proved.
10
In order to show the advantage of the real nonlinear ve-
0
hicle dynamics used in the proposed algorithm, we con-
L0 F3
-10 F1 F4 duct the simulation of the dual-mode DMPC algorithm in
F2 F5
-20
which a linear model is used, but the nonlinear model is
0 1 2 3 4 5 6 used for vehicle simulation. For simplicity and clarity, we
Time (s)
Acceleration errors (m/s2)

take the follower 1 as an example to illustrate the differ-

20
ences between the performances of DMPC based on linear
10 model and nonlinear model. The position trajectories and
0 tracking errors of follower 1 are illustrated in Fig. 5. The
-10
F1
F3
F4
control inputs of follower 1 are shown in Fig. 6.
F2
F5 From Fig. 5, it can be seen that the tracking errors can
-20
0 1 2 3 4 5 6 not completely converge to 0 when the linear model is
Time (s) used. Fig. 6 shows the curve of control input based on the
linear model has small oscillation when it converges to the
Fig. 3. Accelerations and acceleration errors of vehicles. desired value. Apparently, the DMPC based on nonlinear
 Dual-mode Distributed Model Predictive Control for Platooning of Connected Vehicles with Nonlinear Dynamics 3099

40
Table 1. The run time of the dual-mode DMPC program
and the general DMPC program. 20

Class 1 2 3 4 5
0
Dual-mode 25.083 s 24.497 s 24.054 s 24.462 s 24.704 s
DMPC
-20 linear model
General 34.423 s 33.682 s 32.803 s 32.936 s 33.800 s nonlinear model
DMPC

U
-40

model has better performance than the linear model. -60

In addition, we compare the run time of the proposed
dual-mode DMPC program with the general DMPC pro- -80
gram under same conditions. The results are shown in Ta-
ble 1. From the table, we can find that the run time of the -100
0 1 2 3 4 5 6
proposed dual-mode DMPC program is shorter than the Time (s)
general DMPC program, which validates the advantages
of the proposed approaches. Fig. 6. Control inputs u of follower 1.

6. CONCLUSION AND FUTURE WORK

• The vehicle dynamics in (1) is deterministic without
The platoon control for connected vehicles is studied external disturbance. Due to the ubiquitous existence
using the dual-mode DMPC strategy in this paper. The of external disturbance, the robust stabilization of ve-
nonlinear dynamics with input boundaries is applied into hicle platoon system should be guaranteed. Robust
the DMPC scheme, which makes the proposed control DMPC will be studied for vehicle platoon control and
scheme more practical than the linear DMPC. Moreover, the relevant proofs will be reconsidered.
comparing with the general DMPC, the dual-mode struc- • To further improve the performance of vehicle pla-
ture has two stages in the control process, where a dis- toon control, multiple constraints such as state con-
tributed model predictive controller and a local state feed- straints and jerk constraints should be considered for
back controller are applied respectively to save compu- vehicle platoon control. These constraints can make
tation and communication resources. Then the feasibil- the control scheme more practical and keep the vehi-
ity and stability of the proposed control system are both cle platoon procedure smooth. Moreover, the feasibil-
strictly analyzed, in which the sufficient conditions for ity of constrained DMPC will attract more attention.
system stability are derived by Lyapunov stability theory.
In final, numerical simulations are provided to verify the
feasibility and effectiveness of the proposed approaches. REFERENCES
In addition, the following problems will be investigated
in our further work. [1] J. Zhang, F.-Y. Wang, K. Wang, W.-H. Lin, X. Xu, and
C. Chen, “Data-driven intelligent transportation systems:
A survey,” IEEE Transactions on Intelligent Transporta-
50 tion Systems, vol. 12, no. 4, pp. 1624-1639, 2011.
Positions (m)

40 [2] X. Liu and Y. Zou, “A kind of synergic control for a col-

lection of vehicles,” International Journal of Control, Au-
30 tomation and Systems, vol. 10, no. 3, pp. 459-469, 2012.
linear model
nonlinear model
20 [3] M. Jerret, R. McConnell, J. Wolch, R. Chang, C. Lam,
0 1 2 3 4 5 6
G. Dunton, F. Gilliland, F. Lurmann, T. Lslam, and K.
Time (s)
Berhane, “Traffic-related air pollution and obesity forma-
Tracking errors (m)

1 tion in children: a longitudinal, multilevel analysis,” Envi-

0 ronmental Health, vol. 13, no. 1, pp. 49, 2014.

-1 [4] S. Solyom and E. Coelingh, “Performance limitations in

-2
vehicle platoon control,” IEEE Intelligent Transportation
linear model
nonlinear model Systems Magazine, vol. 5, no. 4, pp. 112-120, 2013.
-3
0 1 2 3 4 5 6
[5] M. Amoozadeh, H. Deng, C. N. Chuah, H. M. Zhang, and
Time (s)
D. Ghosal, “Platoon management with cooperative adap-
tive cruise control enabled by VANET,” Vehicular Commu-
Fig. 5. Trajectories and tracking errors of follower 1. nications, vol. 2, no. 2, pp. 110-123, 2015.
3100 Maode Yan, Wenrui Ma, Lei Zuo, and Panpan Yang

[6] H. Wen, J. Sun, and X. Zhang, “Study on traffic congestion [18] J.-W. Kwon and D. Chwa, “Adaptive bidirectional pla-
patterns of large city in China taking Beijing as an exam- toon control using a coupled sliding mode control method,”
ple,” Procedia-Social and Behavioral Sciences, vol. 138, IEEE Transactions on Intelligent Transportation Systems,
pp. 482-491, 2014. vol. 15, no. 5, pp. 2040-2048, 2014.
[7] Y. Zhang, T. Liu, Q. Bai, W. Shao, and Q. Wang, “New [19] M. Yan, J. Song, L. Zuo, and P. Yang, “Neural adap-
systems-based method to conduct analysis of road traffic tive sliding-mode control of a vehicle platoon using output
accidents,” Transportation research part F: traffic psychol- feedback,” Energies, vol. 10, no. 11, pp. 1906, 2017.
ogy and behaviour, vol. 54, pp. 96-109, 2018. [20] W. B. Dunbar and R. M. Murray, “Distributed receding
[8] S. E. Shladover, C. A. Desoer, J. K. Hedrick, M. horizon control for multi-vehicle formation stabilization,”
Thomizuke, J. Walrand, W.-B. Zhang, D. H. McMahon, Automatica, vol. 42, no. 4, pp. 549-558, 2006.
H. Peng, S. Sheikholeslam, and N. McKeown, “Automated [21] H. Li, Y. Shi, and W. Yan, “Distributed receding horizon
vehicle control developments in the PATH program,” IEEE control of constrained nonlinear vehicle formations with
Transactions on Vehicular Technology, vol. 40, no. 1, pp. guaranteed γ -gain stability,” Automatica, vol. 68, pp. 148-
114-130, 1991. 154, 2016.
[9] R. Kianfar, B. Augusto, A. Ebadighajari, U. Hakeem, J. [22] W. B. Dunbar and D. S. Caveney, “Distributed receding
Nilsson, A. Raza, R. S. Tabar, N. V. Irukulapati, C. En- horizon control of vehicle platoons: Stability and string
glund, P. Falcone, S. Papanastasiou, L. Svensson, and H. stability,” IEEE Transactions on Automatic Control, vol.
Wymeersch, “Design and experimental validation of a co- 57, no. 3, pp. 620-633, 2012.
operative driving system in the grand cooperative driving [23] T. Keviczky, F. Borrelli, K. Fregene, D. Godbole, and
challenge,” IEEE Transactions on Intelligent Transporta- G. Balas, “Decentralized receding horizon control and
tion Systems, vol. 13, no. 3, pp. 994-1007, 2012. coordination of autonomous vehicle formations,” IEEE
[10] T. Robinson, E. Chan, and E. Coelingh, “Operating pla- Transactions on Control Systems Technology, vol. 16, no.
toons on public motorways: An introduction to the sartre 1, pp. 19-33, 2008.
platooning programme,” Proceedings of the 17th World [24] A. Maxin, C. F. Caruntu, and C. Lazar, “Distributed model
Congress on Intelligent Transport Systems, vol. 1, pp. 1- predictive control algorithm for vehicle platooning,” Pro-
12, 2010. ceedings of the 20th International Conference on System
Theory, Control and Computing, pp. 657-662, 2016.
[11] S. Tsugawa, S. Kato, and K. Aoki, “An automated truck
platoon for energy saving,” Proceedings of the IEEE Inter- [25] H. A. Izadi, B. W. Gordon, and Y. Zhang, “Decentral-
national Conference on Intelligent Robots and Systems, pp. ized receding horizon control for cooperative multiple ve-
4109-4114, 2011. hicles subject to communication delay,” Journal of Guid-
ance, Control, and Dynamics, vol. 32, no. 6, pp. 1959-
[12] C. S. Chin and C. Wheeler, “Sliding-mode control of an
1965, 2009.
electromagnetic actuated conveyance system using con-
tactless sensing,” IEEE Transactions on Industrial Elec- [26] Y. Zhang, S. E. Li, K. Li, F. Borrelli, and J. K. Hedrick,
tronics, vol. 60, no. 11, pp. 5315-5324, 2013. “Distributed model predictive control for heterogeneous
vehicle platoons under unidirectional topologies,” IEEE
[13] M. Yan, Y. Tang, P. Yang, and L. Zuo, “Consensus based Transactions on Control Systems Technology, vol. 25, no.
platoon algorithm for velocity-measurement-absent vehi- 3, pp. 899-910, 2017.
cles with actuator saturation,” Journal of Advanced Trans-
portation, vol. 2017, pp. 1-8, 2017. [27] R. Wang, H. Jing, C. Hu, M. Chadli, and F. Yan, “Robust
H∞ output-feedback yaw control for in-wheel motor driven
[14] Q. Zhu, B. Dai, Z. Huang, Z. Sun, and D. Liu, “An adaptive electric vehicles with differential steering,” Neurocomput-
longitudinal control method for autonomous follow driving ing, vol. 173, pp. 676-684, 2016.
based on neural dynamic programming and internal model
[28] H. Dahmani, M. Chadli, A. Rabhi, and A. El Hajjaji, “Road
structure,” International Journal of Advanced Robotic Sys-
curvature estimation for vehicle lane departure detection
tem, vol 14, no. 6, pp. 1-13, 2017.
using a robust Takagi-Sugeno fuzzy observer,” Vehicle Sys-
[15] R. Kianfar, P. Falcone, and J. Fredriksson, “A control tem Dynamics, vol. 51, no. 5, pp. 581-599, 2013.
matching model predictive control approach to string sta- [29] G. Wang, M. Chadli, H. Chen, and Z. Zhou, “Event-
ble vehicle platooning,” Control Engineering Practice, vol. triggered control for active vehicle suspension systems
45, pp. 163-173, 2015. with network-induced delays,” Journal of the Franklin In-
[16] B. Li, H. Y. Yang, Z. Q. Chen, and Z. X. Liu, “Distributed stitute, vol. 356, no. 1, pp. 147-172, 2019.
containment control of multi-agent systems with general [30] M. A. S. Kamal, M. Mukai, J. Murata, and T. Kawabe,
linear dynamics and time-delays,” International Journal of “Model predictive control of vehicles on urban roads for
Control, Automation and Systems, pp. 1-9, 2018. improved fuel economy,” IEEE Transactions on Control
[17] Y. Li, K. Li, T. Zheng, X. Hu, H. Feng, and Y. Li, “Eval- Systems Technology, vol. 21, no. 3, pp. 831-841, 2013.
uating the performance of vehicular platoon control under [31] V. Turri, B. Besselink, and K. H. Johansson, “Cooperative
different network topologies of initial states,” Physica A: look-ahead control for fuel-efficient and safe heavy-duty
Statistical Mechanics and its Applications, vol. 450, pp. vehicle platooning,” IEEE Transactions on Control Sys-
359-368, 2016. tems Technology, vol. 25, no. 1, pp. 12-28, 2017.
 Dual-mode Distributed Model Predictive Control for Platooning of Connected Vehicles with Nonlinear Dynamics 3101

[32] D. Q. Mayne, J. B. Rawlings, C. V. Rao, and P. O. Scokaert, Lei Zuo received his B.Sc. and Ph.D.
“Constrained model predictive control: Stability and opti- degrees form Northwestern Polytechnical
mality,” Automatica, vol. 36, no. 6, pp. 789-814, 2000. University (NWPU), in 2011 and 2017, re-
[33] W. B. Dunbar, “Distributed receding horizon control of dy- spectively. Since June 2017, he has been
namically coupled nonlinear systems,” IEEE Transactions with the school of Electronic and Control
on Automatic Control, vol. 52, no. 7, pp. 1249-1263, 2007. Engineering, Chang’an University, Xi’an,
where he is currently an Assistant Pro-
[34] S. Bououden, M. Chadli, and H. R. Karimi, “A robust pre- fessor. His current research interests in-
dictive control design for nonlinear active suspension sys- clude multi-agent systems, coverage con-
tems,” Asian Journal of Control, vol. 18, no. 1, pp. 122- trol, model predictive control and vehicle platoon control.
132, 2016.
[35] S. Bououden, M. Chadli, L. Zhang, and T. Yang, “Con-
Panpan Yang received his Ph.D. from
strained model predictive control for time-varying delay
Northwestern Polytechnical University,
systems: Application to an active car suspension,” Inter-
Xi’an, China, in 2016. He is currently a
national Journal of Control, Automation and Systems, vol.
lecturer in the School of Electronic and
14, no. 1, pp. 51-58, 2016.
Control Engineering, Chang’an Univer-
sity, Xi’an, China. His main research
interests include control and application
Maode Yan received his B.S., M.S. and of mobile robots, modeling, control and
Ph.D. degrees in School of Marine Science analysis of flocking system, cooperative
and Technology, Northwestern Polytech- control of autonomous vehicles.
nical University, Xi’an, China, in 1996,
1999, and 2001, respectively. He is cur- Publisher’s Note Springer Nature remains neutral with regard
rently a Professor with the School of Elec- to jurisdictional claims in published maps and institutional affil-
tronic and Control Engineering, Chang’an iations.
University, Xi’an, China. His research in-
terests are networked control systems, ve-
hicle platoon modeling and control, robots and their formation
control, embedded systems and Applications.

Wenrui Ma received his B.S. degree from

Chang’an University in 2017. He is cur-
rently a graduated student in Electronic
and Control Engineering, Chang’an Uni-
versity, Xi’an, China. His current research
interests are model predictive control, for-
mation control and intelligent system.