Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Proceedings of the 2015 IEEE International Conference on Vehicular Electronics and Safety, Yokohama, Japan. Nov.

5-7,
2015.

Dynamic Programming for Model Predictive Control


of Adaptive Cruise Control Systems

Yu-Chen Lin*, Hsiang-Chieh Hsu Wen-Jen Chen


Department of Automatic Control Engineering Mechanical and Systems Research Laboratories
Feng Chia University Industrial Technology Research Institute
Taichung, Taiwan, R.O.C. Hsinchu, Taiwan, R.O.C.
yuchlin@fcu.edu.tw; M0303696@fcu.edu.tw wenjen@itri.org.tw

Abstract—This paper presents a model predictive control is able to effectively avoid collision with the vehicle in front by
approach for the design of vehicular adaptive cruise control always keeping the car at a safe distance because the systems
(ACC) systems by a finite horizon dynamic programming can apply the brakes more quickly than a drivers’ reaction time.
approach, which is aimed at providing automatic and steady The systems can also provide a much smoother throttle and
car-following capability and enhancing riding comfort. The brake modulation than most drivers could. In addition, the
formalism is based on the Bellman's optimality principle and maximum acceleration and minimum deceleration are both
receding horizon strategy to obtain the optimal feedback kept within pre-specified limitations in order to ensure driving
control gain as evaluated by a cost function. A quadratic cost comfort and safety. However, some of the above designed
function is developed that considers the contradictions between objectives are often contradictory. For instance, frequent
minimal tracking error and acceleration limits of the ACC acceleration and deceleration will reduce the comfort but is
vehicle. Hence, the characteristics of permissible following necessary to maintain an accurate and safe following distance
distance and acceleration command are expressed as linear between the leading vehicle and the following vehicle,
constraints, simultaneity. To solve the constrained finite- especially when the leading vehicles’ motion is not smooth.
horizon optimal control problem, a model based optimized After all, a well-regulated and robust cruise control
dynamic terminal controller is proposed to drive the system methodology is a necessary component of adaptive cruise
state into a terminal region as tracking error compensation. control systems.
Extensive simulations and comparisons for relevant traffic
scenarios of ACC systems cannot only perform to verify the In the past decade, much significant research in the
investigation of dynamics and control strategies of the ACC
proposed optimal predictive controller design but also preserve
the asymptotic stability. systems have been carried out and published. The central issue
of most published literature is to find a control strategy to
Keywords—model predictive control; dynamic derive the required vehicle speed so that it maintains a
programming; adaptive cruise control (ACC); receding horizon. predetermined distance between a host vehicle and a lead
vehicle. Some simplified approaches, such as the PID
I. INTRODUCTION controller [5], linear quadratic control (LQR) optimal synthesis
The statistics from NHTSA (National Highway Traffic approach [6], state feedback control [7], gain-scheduled linear
Safety Administration) show that 31% of traffic accidents are controller [8], and sliding mode control [9], were proposed to
due to rear-end collisions in the U.S [1]. Oftentimes, a rear-end guarantee following stability for the ACC controller design
collision happens when a front vehicle suddenly slows down or with time headway policy [10]. However, the drawback of the
stops, giving the following vehicles no warning and little time time headway policy is that it has poor robustness with traffic
to react. Therefore, radar-based adaptive cruise control (ACC) flow fluctuation. It means that the above control methods
is a longitudinal control system, widely used in recent models cannot be trustworthy, particularly with respect to the large
of vehicles. The velocity of the vehicle is regulated according difference between the desired speed and the actual speed of
to the behavior of the other moving vehicles in front by auto- the host vehicle. The system output caused high overshoot and
modulating the throttle and brake to achieve a safe following intermittent inconsistency when in the presence of variable set-
distance at all times. This feature is termed as vehicle speed points or disturbances. LQR optimal synthesis approach for
control mode (cruise control mode). In addition to the speed ACC controller design is not widely used, perhaps because it is
control mode, the ACC systems can also automatically adjust regarded as inadequately addressing the problems raised by
the set speed to maintain a specified inter-vehicle distance constraints, nonlinearities and uncertainty. Moreover, several
(SIVD) [2, 3] from a preceding vehicle. This additional feature studies [11-13] proposed a reference model-based nonlinear
is known as vehicle following mode. The ACC systems is a control approach for ACC controller design. This nonlinear
kind of advanced driver assistance systems [4] (ADAS) that is reference model acts as a feed-forward term to design the ACC
mainly designed to help the driver to improve driving safety, feedback controller and satisfy the system performance
fuel efficiency and riding comfort on the highway. The system requirements. However, the design of nonlinear controllers is
always complex, and explicitly calculating the value of the

202
Proceedings of the 2015 IEEE International Conference on Vehicular Electronics and Safety, Yokohama, Japan. Nov. 5-7,
2015.

hybrid performance function is difficult; in addition, it is velocity) between the ACC vehicle and the preceding vehicle
unsuitable for real-time implementation. are measured by using of radar equipment. The allowed SIVD
Upon different approaches, the model predictive controller between the two vehicles varies linearly with the preceding
(MPC) [14] is extensively employed to design the ACC vehicle’s speed Vp , as a result, the SIVD has remained
systems. The MPC is based on using a mathematical model to constant according to the headway time h :
predict the consequences of a controllable plant for solving
online a finite horizon optimal control problem. To be more SIVD  hV p (1)
precise, an open loop control sequence is optimized on a
receding horizon to minimize a nominal cost function subject In this work, the ACC vehicle model is based on [2],
to constraints on inputs and deterministically predicted states. which the corresponding equivalent discrete-time state-space
In [2], a controller using model-based headway control representation of the ACC vehicle model can be expressed by
algorithm with acceleration limitations for the transitional xk 1  Axk  Buk , x (0)  x0
maneuvers of ACC vehicles is presented. The advantage of the (2)
MPC framework is that it can be used to derive control law for yk  Cxk
all feasible initial conditions. For classical model predictive
control (MPC), the control action at each time step is derived where xk n and uk  1 denote, respectively, state and
by adapting a general quadratic program (QP) [2, 14]. However, control input vectors at time k ; the state vector xk , with some
solving the QP applying general purpose methods is usually
time consuming and simultaneity unsatisfied fuel efficiency slight abuse of notation, and
and ride comfort when the cruising process is rather restricted xk  ek R R
T     R  SIVD  R R
 
T ,

operating regions. Therefore, finding the optimal control gain
while simultaneously satisfying all conflicting objectives and     (3)
constraints is a primary purpose. 1 Ts 0   0  1 0 0 
A  0 1 T , B   0  , C  0 1 0
Unlike the previous publications, this paper presents an      
efficient way of using finite horizon dynamic programming [15] 0 0 1  Ts  Ts  0 0 1 
and taking a penalty function [14] as the soft constraints with      
pre-specified hard limitations for the design of discrete-time where ek denotes the driver permissible tracking error; R and
model-based predictive control (DMPC) of vehicular adaptive
cruise control (ACC) systems; it is aimed at providing steady  are denote the range rate and absolute acceleration of the
R
car-following capability and ride comfort. Moreover, the ACC vehicle, respectively. Ts and  are denote the discrete
purpose of setting up the SIVD with zero range rate and the sampling time and the corresponding time lag of the ACC
acceleration limits of the ACC vehicle can be merged into
system, respectively.
constrained optimal control problems as control constraints.
Not only does the proposed approach improves the ACC III. DMPC CONTROLLER DESIGN
performance to avoid frequent acceleration and deceleration
behavior, but the resulting feedback control law also guarantees Discrete-time model predictive control is based on
asymptotic stability for ACC controller design. We show that it optimizing the future plant control trajectory subject to
is appropriate to deal with the task to achieve an optimal trade- constraints. The result of the optimization process is the
off among ride comfort, acceleration and deceleration optimal control sequence u  u0 , u2 ,..., uN 1 for the
constraints. Effectiveness of the proposed control strategy is predictive horizon N . Based on the receding horizon strategy,
demonstrated by extensive numerical verifications of the only of the control sequence is applied to the ACC systems (2).
proposed ACC systems. The rest of the control sequence is employed for plant behavior
II. MODELING OF THE ACC SYSTEMS estimation by using the dynamic programming optimization.
The prototype of the ACC vehicle model consists of the To consider the car-following capability and ride comfort
ACC vehicle and the preceding vehicle, shown as in Fig. 1. issue in ACC systems, the vehicle acceleration and
The preceding vehicle steers independently, whereas, the ACC deceleration are directly concerned with the input force.
vehicle is to set up and keeps a SIVD with zero range rate Therefore, the control input is limited to both min-max
behind the preceding vehicle. constraints, and the formulation is:

ACC Vehicle Preceding Vehicle


umin  uk  umax (4)
e SIVD

where upper acceleration limit u min and lower acceleration


limit u max for ACC vehicle will be assumed equal to -0.25g and
R Vp 0.5g, respectively [2]. Notices that the lower acceleration limit
0 implies that the ACC vehicle do not have enough brakes to
Fig. 1. ACC systems configuration
avoid collision with the preceding vehicle. Moreover, the
relative distance should always be positive, i.e. R  0 .
The range R (relative distance) and range rate R (relative

203
Proceedings of the 2015 IEEE International Conference on Vehicular Electronics and Safety, Yokohama, Japan. Nov. 5-7,
2015.

To applying MPC on (2) and considering the input derive an optimal control law with the initial state x0 . The goal
constraints (4), a constrained finite horizon optimal control
problem is solved over a prediction horizon N during each of the design is to find the optimal control law uk from
sampling period, which is using the current state as the initial sequence of control inputs u , with depends on the N -step
state. The objective can be achieved by finding the sequence of prediction horizon, that minimize J ( xk , u) when it reaches
control inputs u  u0 , u2 ,..., uN 1 , then the solution of the
xN .
finite horizon optimal control problem exists and is given by
First, the Hamilton-Jacobi-Bellman (HJB) equation
u*  arg min J ( xk , u) (5) provides solutions to the optimal control problem of the
u0 ,u1,uN 1 systems (2). Following Pontryagin’s formalism [15], a
To find a sequence of control inputs, the following Hamiltonian function associated with the optimal control
performance index must be minimized within a future horizon problem is formulated as
which minimizes the generalized quadratic cost function
repeatedly over k :
N 1
H k ( x, u ,  ) 
1
2  
2
yk  xkref   ruk2  ( xk )
Q  (10)
T T
min J ( xk , u)    N    L( xk , uk ) (6)  k 1 xk 1  k ((uk ))
uk
k 0
where k  n is a co-state vector sequence 1, 2 ,..., N ; 
where   N  is a final state punishment function of the final
is a multiplier vector sequence 0 , 2 ,..., N 1 . (uk ) denotes
time N which is positive definite; L( xk , uk ) is a general time- the inequality constraint, and
varying function of the state and control input at each time k in  u  u 
0, N  1 . Using the MPC on (2), the performance index (uk ) :  k    max  (11)
considered here is quadratic, specified by  uk   umin 
Thus, instead of min-max constraint (4), we have
L( xk , uk ) 
1
2  y k  xk 
ref
2
 ruk2  ( xk )
Q
 (7) corresponding inequality constraint (uk )  0 . Necessary
conditions, for a constrained minimum are thus given by
and Co - state equation:
1 2 H ( xk )
  N   (CxN  x Nref ) (8) k  k  AT k 1  C T Q(Cxk  xkref )  ,
2 SN xk xk
Stationarity condition:
In the above equations, xk is the initial state obtained from the
  0, uk  umin , k  0,
systems at sampling period k ; xkref is the known preceding H k T kT ((uk )) 
 ruk  B k 1    0, umin  uk  umax , k  0,
vehicle reference state vector; yk is the predicted output vector; uk uk  0,
 uk  umax , k  0.
xN is the predicted terminal state within the prediction
Boundary condition:
horizon;  ( xk ) is the penalty function;  denotes
  N 
Euclidean norm. N is the prediction horizon  0  N  . N   C T S N (CxN  xNref )
xN
Moreover, suppose there exist S N  0, Q  0 and r  0 (12)
which that guarantee the strict convexity of the cost function Since the control uk is constrained to lie in an admissible
with respect to the controls. Design the penalty function [14] in
corresponding with state constraints as region; therefore, the Hamiltonian must be minimized over all
admissible uk for optimal values of the state and costate.
N 1
2
 ( xk )   ( xk  xkref ) (9)
k 0
P H k ( x* , u* ,  * )  H k ( x * , u,  * ), (13)
for all admissible uk , k  0,..., N  1.
where P is an 3  3 identity matrix. The penalty function
 ( xk ) , enforcing as the soft constraints, is penalized along By applying the optimality condition in (12) the optimal
with control effort. Thus, the performance index (6) indicates controller can be expressed as
that keep the tracking error small without using too much
uk   r 1B T k 1 (14)
control energy. Additionally, the tracking error y k  x kref , is
mainly used to track a desired reference trajectory xkref over a According to the sweep method and the boundary condition
(12), k can reasonable to assume that for all k  N as
specified time interval  0, N  1 . From (6), it is essential to

204
Proceedings of the 2015 IEEE International Conference on Vehicular Electronics and Safety, Yokohama, Japan. Nov. 5-7,
2015.

k  Gk xk  vk (15) N 1
V  xN   V  x j    V  xk  (21)
T T ref k j
where Gk  C Sk C and vk  C Sk x k
. Using (15) into the
co-state equation (12), we now can solve the Sk to find the where V ( xk )  V  xk 1   V  xk  is the first difference.
Gk and vk ,then Since V  xk  is assumed a continuously differentiable
T T
Gk  A Gk 1  Gk 1 B( B Gk 1 B  R ) B Gk 1  A 1 T function; therefore, V  xk  can be expanded by using Taylor
(16)
series about the operating point of xk renders
 C T QC
V  xk 1   V  xk   VkT  xk 1  xk  (22)
and
vk   AT  AT Gk 1B ( B T Gk 1B  r ) 1 B T  vk 1 Therefore, we can ignore term higher than first order to
(17) receive
 ( xk ) V  xk   VkT  xk 1  xk  (23)
 C T Qxkref 
xk
From (21) and (23), the (21) can rewritten as
To compute Gk and vk for each xk using (6), we need to N 1
V  x N   V  x j    VkT  xk 1  xk  (24)
work backward from k  N . When we continued to decrement k j
k and apply the Bellman’s principle of optimality [15], the
optimal control law for each k  N  1,...,0 can be expressed According the (19), Similarly, we rewrite (6) as
N 1
1
as
uk   K kG xk  K kv vk 1 , v N  C T S N x Nref (18)
 2

J ( x j , u)    yk  xkref   ruk2  ( xk )    N  (25)
k j 2
Q

Adding (24) on both sides of (25) such that


where K kG and Kkv are the suboptimal feedback control gain N 1
1
and feedforward control gain, respectively. By solving the  
J ( x j , u)  V x j  
k j 2 
yk  xkref
2

Q

ruk2   ( xk )
(26)
modified equations (16), (17) and (11), then optimal control
uk from the set of admissible controls to minimize VkT  xk 1  xk       N   V  xN 

H k ( x * , u ,  * ) can be selected. Thus, the optimal one may conclude the existence of the above equality (20), by
performance cost on can be determined before applying the using the (20), it is easy to show that
control law. V  xk   V  xk 1   V  xk 
2 (27)
IV. STABILITY ANALYSIS   1  yk  xkref   ruk2   ( xk )
2 Q
Stability analysis should be noted that the DMPC 2 2
controller is locally stabilizing the discrete-time ACC systems   1 2  yk  xkref    1 2 min  Q   yk  xkref 
Q
in (2). We shall utilize the Lyapunov function theory to
analyze the asymptotic stability criteria for our proposed the This implies that the difference of V  xk  along the systems
ACC systems.
trajectory is negative. Thus, the systems (2) controlled by (18)
Theorem I. Consider the discrete-time closed-loop systems (2) is asymptotically stability. This completes the proof. □
with the Lyapunov functional
V ( xk )  min J ( xk , u) (19) V. SIMULATION RESULT
uk
The objective of this section is to evaluate the performance
Suppose there exist Q  0, r  0 for each k such that of the proposed DMPC method for ACC systems. Therefore,
satisfying the following: two controller design algorithms are simulated and compared,
N 1 i.e.: a discrete-time model-based predictive control (DMPC),
1
 
k j 2
2

Q
 
yk  xkref  ruk2  ( xk )  VkT  xk 1  xk   0,
(20)
an unconstrained linear quadratic regulator (LQR), and a
constrained linear quadratic regulator (C-LQR) controller. For
V  xN   N  the simulation, the headway time h , sampling time Ts and
the time lag corresponding  of the ACC vehicle model (2)
where VkT ( x ) is the gradient vector for each xk , then are choice to be 1 sec , 0.1 sec and 0.5 sec , respectively. The
systems (2) subjected to the optimal control law described by weighting of the performance index (6) are defined as
(18) for all j  0,..., N  1 would be asymptotically Q  diag (1, 1, 1), r  1 . The four traffic scenarios are carried
Proof: Assume that V  xk   0 exist and is continuously out with a discrete-time ACC vehicle model, whose simulation
results are shown below.
differentiable. Then

205
Proceedings of the 2015 IEEE International Conference on Vehicular Electronics and Safety, Yokohama, Japan. Nov. 5-7,
2015.

Case A. Stalled preceding vehicle scenario


First scenario considers that the preceding vehicle
maintains a stopped state, i.e. V p  0 . The ACC vehicle is
initially traveling at a velocity of 30 m/s , and the preceding
vehicle is in an idle stop and completely stationary state.
Assuming that the stopping distance is defined as specific to
the 110 m event. The ACC vehicle must be decelerate and
successfully stop behind the preceding vehicle. Since the
preceding vehicle’s velocity is zero; therefore, the feasible
SIVD need to be defined by ACC vehicle is 0m.
Figure 2 illustrates the simulation results for the proposed
constrained DMPC, conventional LQR, and constrained C-
LQR algorithms for the ACC systems. Figure 2(b) shows that
the initial speed of the preceding vehicle is 0m/s, and the
dashed line represents the preceding vehicle status. In Fig. 2, a
sign “*” indicates that a collision has occurred. Although these
Fig. 3. ACC systems operates in constant velocity preceding vehicle scenario.
control methods are all able to successfully let ACC vehicle (a) position, (b) velocity, (c) acceleration, and (d) control input.
would stop in a foreseeable period of time. However, in the
case of limited deceleration with C-LQR, the ACC vehicle Since the relative distance between ACC vehicle and
doesn't have enough braking time to avoid a collision with the preceding vehicle is too far, the ACC systems firstly needs to
preceding vehicle. Therefore, the ACC vehicle must instantly increase the speed to track the preceding vehicle. Next, the
begin to decelerate, rather than acceleration. In addition, since ACC vehicle needs to decelerate to slowly and appropriately
the velocity limits have not considered into the LQR, the ACC follow up a preceding vehicle when it enter an appropriate
vehicle also collides with the preceding vehicle. Furthermore, distance. As in the case of the C-LQR method, the ACC
we can certainly observe from Fig. 2(d) that unconstrained vehicle across SIVD, i.e. the crash occurred, because the ACC
LQR control laws are more than maximum control input umax ; vehicle doesn't have enough deceleration to avoid a collision
thus, it will lead to a discomfort driving and fuel consumption. with the preceding vehicle.

Case C. Preceding vehicle’s acceleration and deceleration


scenario
An acceleration and deceleration of preceding vehicle in
scenarios is considered. It is also defined that the ACC vehicle
is initially traveling at a velocity of 30 m/s , the preceding
vehicle accelerates at a distance of 50 m/s from 10 m/s to
30m/s and then decreases to 10 m/s . In this case, the initial
T
state defined as x0   50 m 30 m/s 0 . Similarly the
simulation results of the acceleration and deceleration driving
scenario are presented in Fig. 5.

Fig. 2. ACC systems operates in stalled preceding vehicle scenario. (a)


position, (b) velocity, (c) acceleration, and (d) control input.

Case B. Preceding vehicle’s constant velocity scenario


In this scenario, it is defined that the ACC vehicle runs at
the initial speed of 30 m/s , and the preceding vehicle runs at a
constant speed of 10 m/s . Assuming that the initially relative
distance is defined as specific to the 110 m event. Figure 3
shows the simulation results of the ACC vehicle when the
ACC systems operates in constant velocity preceding vehicle
scenario.
Fig. 4. ACC vehicle operates in preceding vehicle acceleration and
deceleration scenario. (a) position, (b) velocity, (c) acceleration, and (d)
control input.

206
Proceedings of the 2015 IEEE International Conference on Vehicular Electronics and Safety, Yokohama, Japan. Nov. 5-7,
2015.

Figure 5(c) shows that the DMPC control laws initially Lyapunov stability theory, a guaranteed cost control law with
commands a deceleration to reduce the range rate as the ACC linear state feedback for ACC systems is developed to ensure
vehicle approaches the SIVD. However, the preceding vehicle systems stability when state and control constraints are
is still accelerating when the ACC vehicle approaches the simultaneously considered. Simulation results are presented to
SIVD; thus, the SIVD need to increase. Otherwise, the SIVD illustrate the distinguished performance, collision avoidance
need to decrease when the preceding vehicle is decelerating. and acceleration constrained of the ACC vehicle, the ACC
As shown the simulation, we observe that the proposed DMPC vehicle have been explicitly incorporated into the formulation
responds to the preceding vehicle’s status by controlling the of the control algorithm to successfully perform the penalty
ACC vehicle to accelerate to track the increasing SIVD. for all feasible traffic scenarios that require penalty.

Case D. Preceding vehicle’s emergency braking scenario ACKNOWLEDGMENT


In this scenario, the preceding vehicle decelerates at the This research was sponsored by Ministry of Economics,
acceleration of 4 m/s 2 within 5 sec . The initially traveling at Taiwan, R.O.C., under the contract No.: 104-EC-17-A-23-
a velocity of ACC vehicle is defined as 30 m/s , and the 0803.
initially relative distance between two vehicles is defined as REFERENCES
50 m . The simulation and comparison results of the
[1] NHTSA: “Vehicle backover avoidance technology study,” Report to
acceleration and deceleration driving scenario are presented in Congress, Nov., 2006.
Fig. 6. [2] V. Bageshwar, W. Garrard, and R. Rajamani, “Model predictive control
of transitional maneuvers for adaptive cruise control vehicles,” IEEE
Trans. Veh. Technol., vol. 53, no. 5, pp. 365-374, 2004.
[3] Z. Ali, and D. M. Pathan, “Parametric study of nonlinear adaptive cruise
control vehicle model by vehicle mass,” Emerging Trends and Appli. in
Information Communication Tech. Communications in Computer and
Information Science - Springer-Verlag, vol. 281, pp. 81-91, 2012.
[4] K. Bengler, K. Dietmayer, B. Färber, M. Maurer, C. Stiller, and H.
Winner, “Three decades of driver assistance systems,” IEEE Intelligent
Transportation Syst. Magazine, vol. 6, no. 4, pp. 6-22, 2014.
[5] V. Milanes, J. Villagra, J. Godoy, and C. Gonzalez, “Comparing fuzzy
and intelligent PI controllers in stop-and-go manoeuvres,” IEEE Trans.
Control Syst. Tech., vol. 20, no. 3, 2012.
[6] Z. Liang, Z. Ren, and X. Shao, “Decoupling trajectory tracking for
gliding reentry vehicles,” IEEE/CAA J. Automayica Sinica, vol. 2, no. 1,
pp. 115-120, 2015.
[7] W. U. N. Fernando, and S. Kumarawadu, “Discrete-time neuroadaptive
control using dynamic state feedback with application to vehicle motion
control for intelligent vehicle highway systems,” IET Control Theory
and Appli., vol. 4, no. 8, pp. 1465-1477, 2010.
Fig. 5. ACC systems operates in emergency braking scenario. (a) position, (b) [8] G. R. Widmann, M. K. Daniels, L. Hamilton, etc., “Comparison of lidar-
velocity, (c) acceleration, and (d) control input. based and radarbased adaptive cruise control systems,” In: SAE 2000
world congress, Detroit, 2000.
Figure 6 shows that due to the constraints deriving from [9] B. Ganji, A. Z. Kouzani, S. Y. Khoo, and S. Z. Mojtaba, “Adaptive
driver ride comfort, the C-LQR is unable to provide enough curise control of a HEV using sliding mode control,” Expert Systs. with
braking force. In addition, since the acceleration limits have Appli., vol. 41, pp. 607-615, 2014.
not considered into the LQR, the ACC vehicle leads to [10] J. Zhou, and H. Peng, “Range policy of adaptive cruise control vehicles
for improved flow stability and string stability,” IEEE Trans. Intelligent
discomfort and fuel consumption when driving. For our Transportation Syst., vol. 6, no. 2, pp. 229-237, 2005.
proposed DMPC, the ACC vehicle can beforehand decelerate [11] J. J. Martinez, and C. Canudas-de-Wit, “A safe longitudinal control for
to avoid a collision when the preceding vehicle brakes more adaptive cruise control and stop-and-go scenarios,” IEEE Trans. Control
urgently. Syst. Tech., vol. 15, no 2, pp. 246-258, 2007.
[12] P. Shakouri, and A. Ordys, “Nonlinear model predictive control
VI. CONCLUSION approach in design of Adaptive Cruise Control with automated
switching to cruise control,” Control Engineering Practice, vol. 26, pp
This paper proposes a novel model-based predictive 160-177, 2014.
control approach for design of the discrete-time ACC systems [13] P. Shakouri, A. Ordys, and R. A. Mohamad, “Adaptive cruise control
with uncertainties based on finite horizon dynamic with stop&go function using the state-dependent nonlinear model
programming approach. Theoretical development and predictive control approach,” ISA Transactions, vol. 51, pp. 622-631,
2012.
numerical verification show that the proposed discrete-time
[14] D. Q. Mayne.,“Constrainted model predictive control: stability and
model-based predictive control strategy is feasible for optimality,” Automatica, vol. 36, pp. 789-814, 2000.
implemented to generate appropriate control inputs to avoid a [15] F. L. Lewis, D. L. Vrabie, and V. L. Syrmos, Optimal Control 3rd ed,
collision with preceding vehicle and to get the comfortable Wiley, 2012.
driving, and increase the cruising stability. Based on the

207

You might also like