Post on 06-Dec-2021
Road Condition Based Adaptive Model Predictive Control for Autonomous Vehicles
Xin Wang School of Construction Machinery
Changโan University Xiโan, Shaanxi 710064, China
Department of Automotive Engineering
Clemson University Greenville, SC 29607, USA
wangxin@chd.edu.cn
Longxiang Guo Department of Automotive
Engineering Clemson University
Greenville, SC 29607, USA longxig@clemson.edu
Yunyi Jia* Department of Automotive
Engineering Clemson University
Greenville, SC 29607, USA yunyij@clemson.edu
ABSTRACT
Road conditions are of critical importance for motion
control problems of the autonomous vehicle. In the existing
studies of Model Predictive Control (MPC), road condition is
generally modeled with the system dynamics, sometimes
simplified as common disturbances, or even ignored based on
some assumptions. For most of such MPC formulations, the
cost function is usually designed as fixed function and has no
relations with the time-varying road conditions. In order to
comprehensively deal with the uncertain road conditions and
improve the overall control performance, a new model
predictive control strategy based on a mechanism of adaptive
cost function is proposed in this paper. The relation between the
cost function and road conditions is established based on a set
of priority policies which reflect the different cost requirements
under different road grades and friction coefficients. The
adaptive MPC strategy is applied to solve the longitudinal
control problem of autonomous vehicles. Simulation studies are
conducted on the MPC method with both the fixed cost
function and the adaptive cost function. The results show that
the proposed adaptive MPC approach can achieve a better
overall control performance under different road conditions.
Keywords
Model Predictive Control, Longitudinal control, Adaptive cost
function, Road condition.
1 INTRODUCTION
With the rapid development of autonomous driving
technology, many basic self-driving functions have been
implemented and applied to autonomous vehicles successfully.
*Corresponding Author, yunyij@clemson.edu
The latest autonomous vehicles have the functions of adaptive
cruise control, lane keeping, autonomous parking and crash
avoidance, etc., due to the applications of various control
methods and other information techniques. Many people
believe that autonomous vehicles will replace traditional
vehicles gradually and become a major transportation tool
someday in the future [1]. Nowadays, people are no longer
satisfied with merely the safety and the basic functions of the
autonomous vehicle. Higher demands are placed on the
driving/riding comfort and control efficiency. Hence, more
sophisticated control strategies and methods are expected to
improve the control performance with lower control costs.
In recent years, Model Predictive Control (MPC) method
has gained much attention in the field of autonomous vehicles,
due to its unique advantages in solving control problems that
are hard to model accurately and/or have constraints [2]. In
most of the autonomous driving problems, system constraints
are inevitable. These constraints include mechanical input
limitations, the acceleration ability and the safe distance
between the vehicle and other objects, etc. Naturally, MPC
becomes an ideal choice to solve such problems.
In motion control problems of the autonomous vehicle,
road condition is an ineligible factor which influences the
control performance significantly. For moving vehicles, road
conditions mainly refer to the road grade and the friction
coefficient. The former is primarily decided by the
physiographic condition and the design of the road, while the
latter is subject to many factors, including the road pavement
material, the age of the road, and even the weather [3]. Since
road conditions are time-varying and uncertain, comprehensive
modeling is difficult to realize. Therefore, it is meaningful to
deal with the road condition by other ways.
In the existing MPC studies of autonomous vehicles, road
conditions are considered more or less. In most cases, the road
Proceedings of the ASME 2018Dynamic Systems and Control Conference
DSCC2018September 30-October 3, 2018, Atlanta, Georgia, USA
DSCC2018-9095
1 Copyright ยฉ 2018 ASME
Attendee R
ead-Only C
opy
condition, especially the friction coefficient, is modeled with
the system dynamics, as parameters or disturbances. Borrelli et
al. in [4] applied nonlinear MPC to the autonomous steering
system. In their study, the bicycle model and the road-tire
model of the vehicle are combined to establish the system
dynamics. The road friction coefficient is considered as a
parameter of the system dynamics. Falcone et al. in [5] and [6]
also studied the active steering problem based on the same
system dynamics as [4], but using a successive online
linearization approach. In [7], Turri et al. studied the lane
keeping and obstacle avoidance problems based on linear MPC.
The extended bicycle mode is utilized and the friction
coefficient is modeled with the longitudinal tire force. Carvalho
et al. in [8] presented their design of a nonlinear MPC
controller, which can track the lane centerline while avoiding
collisions with obstacles. In their study, the friction coefficient
is also incorporated in the longitudinal tire force model. In [9],
Kamal et al. introduced an energy-saving MPC strategy and
applied it to the adaptive cruise control problem. A concept of
equivalent acceleration is proposed considering the effect of the
road grade. In some other MPC studies, road conditions are
treated as the system constraints. Yi et al. in [10] studied the
MPC-based trajectory planning for critical driving maneuvers.
They introduced a quadratic MPC control method considering
friction limits in evasion maneuvers. In [11], Jalalmaab et al.
proposed an MPC-based collision avoidance scheme for
autonomous vehicles. The maximum road friction coefficient is
considered in the acceleration constraints, and a particular
optimization method is developed to estimate the max road
friction coefficient. In [12], Beal et al. presented a vehicle
stabilization approach by utilizing a model predictive envelope
controller to bound the vehicle motion within the stable region
of the state space. The estimated road condition is used to
define the state boundaries. Among the above studies, the factor
of road grade is not mentioned in [4-8,10-12], while in [9], the
road friction coefficient is not taken into account.
Due to the difficulty of being detected, the information of
road conditions is always obtained through model-based
estimation techniques in many applications. Chen et al. in [13]
designed a recursive least square estimator to achieve the road
friction coefficient online. Kidambi, et al. in [14], developed a
mass and grade estimation method using longitudinal
acceleration. Similar studies can also be found in [15-19].
Whereas, in some of the MPC studies, due to the complexity of
the system, road conditions are not taken into account in the
control process, as in [20-21].
In spite of these existing studies, research and applications
regarding adaptive cost function strategies have rarely been
addressed. In most of the existing MPC studies, road conditions
are usually simplified based on some assumptions, and not all
the factors are modeled with system dynamics. Meanwhile, in
these MPC studies, traditional cost functions with fixed penalty
weights are used, and there is no special design for the cost
function. Therefore, the main contribution of this paper is the
development of a new MPC strategy with the adaptive cost
function in the receding optimization process based on the
time-varying road conditions. This approach keeps all the
advantages of MPC method and at the same time considers the
different requirements of the optimization objectives under
different road conditions. By using the proposed approach, the
control performance can be accomplished with a much lower
control cost.
The remainder of this paper is organized as follows:
Section II describes the control problem to be studied, and
formulates the system dynamics. The MPC controller is
designed and solved in Section III. In Section IV, the road
condition adaptive strategy is devised. Simulation study is
presented and discussed in Section V. Finally, conclusions and
future work of this paper are addressed in Section VI.
2 PROBLEM FORMULATION
The purpose of this paper is to investigate a new road
condition adaptive MPC strategy. To achieve this goal, the
longitudinal control problem is chosen as a preliminary study,
because it is less complicated than other motion control
problems. For the same reason, a mass point model is utilized
to describe the system dynamics.
In the longitudinal control problem, the actual velocity ๐ฃ
of the vehicle must well track the given reference velocity ๐ฃ๐ .
An MPC controller will be designed to implement this control
with consideration of the time-varying road conditions.
Figure 1. Force diagram of a moving vehicle on the upgrade road
Both the road grade and the friction coefficient are
considered in the control process. We use ๐ to represent the
road grade, as shown in Figure 1, which can be detected via the
attitude sensor equipped in the vehicle. The friction coefficient,
which has strong stochastic characteristics, is hard to detect by
existing sensors. Therefore, the slip ratio is employed to replace
the friction coefficient approximately in this study, which is
defined by
๐ = 1 โ ๐ฃ๐ ๐ {๐ฃ โ 0 for braking๐ โ 0 for driving (1)
where s is the slip ratio; ๐ is the angular speed of the vehicle
wheel; ๐ is the tire radius.
As shown in Figure 1, the autonomous vehicle is regarded
as a mass point in the longitudinal control problem. We can
establish its dynamic model by using Newtonโs Second Law
directly [22, 23], and thus have
๐๏ฟฝฬ๏ฟฝ = ๐๐ข โ 12 ๐ถ๐๐๐๐ด๐๐ฃ2 โ ๐ถ๐๐ฝ(๐ )๐๐cos(๐) โ ๐๐sin(๐) (2)
where ๐ฃ is the velocity of the vehicle; ๐ข is the acceleration
control input per unit mass to accelerate or decelerate the car
which is derived from the control strategy; ๐ is the vehicle
mass; ๐๐ is the air density; ๐ถ๐ is the drag coefficient; ๐ด๐ is
2 Copyright ยฉ 2018 ASME
the frontal area of the vehicle; ๐ถ๐ is the rolling resistance
coefficient; ๐ฝ is an influence factor which is related to the road
friction coefficient/slip ratio ๐ .
In general, the system dynamics of a vehicle is always
non-linear. For the reason that a non-linear MPC problem is
harder to solve than a linear MPC problem, we can simplify the
non-linear model to a linear one. By linearization and
discretization, a linear CARIMA model can be obtained.
CARIMA model is the most common used transfer function
model in MPC [24]. It is a discrete and one-step-ahead
prediction model, and can be expressed by
๐(๐ง)๐ฆ๐ = ๐(๐ง)๐ข๐ + ๐ค๐๐ฅ (3)
where ๐ข๐ , ๐ฆ๐ and ๐ค๐ are the sampled values of the input ๐ข(๐ก), the output ๐ฆ(๐ก) and the system disturbance ๐ค(๐ก) at
time instant ๐ , respectively; ๐(๐ง) and ๐(๐ง) are the
denominator and the numerator polynomial of the impulse
response transfer function of the system, which can be derived
from model (2) by using z Transform; ๐ฅ is the sampling time
interval. Then (3) can be written as
๐ด(๐ง)๐ฆ๐ = ๐(๐ง)๐ฅ๐ข๐ + ๐ค๐ (4)
where ๐ด(๐ง) = ๐(๐ง)๐ฅ = 1 + ๐ด1๐งโ1 + โฏ+ ๐ด๐๐งโ๐ , and ๐(z) = ๐1๐งโ1 + โฏ + ๐๐๐งโ๐ , where ๐ and ๐ are the order
and the input number of the system, respectively. Consequently,
the discrete model of the system can be represented by
๐ฆ๐+1 + ๐ด1๐ฆ๐ + โฏ+ ๐ด๐๐ฆ๐โ๐+1 = ๐1ฮ๐ข๐ + ๐2ฮ๐ข๐โ1 + โฏ+๐๐ฮ๐ข๐โ๐+1 (5)
where ฮ๐ข๐ is the input increment. The output ๐ฆ๐ is often
measurable in practice. Equation (5) is the basis of the
one-step-ahead prediction, which can be used recursively to get
an n-step-ahead prediction.
3 MPC CONTROLLER DESIGN
To get the solution of the control, the linear MPC problem
can be converted to a standard QP (Quadratic Programming)
problem, as shown by (6).
{min๐ฅ ๐ฝ = ๐ฅ๐๐๐ฅ + ๐ฅ๐๐ + ๐๐ . ๐ก. ๐๐ฅ โค ๐ (6)
where ๐ is the weight matrix; ๐ and ๐ are the coefficient
matrix of the first-order term, and the constant matrix,
respectively; ๐ and ๐ are the coefficient matrix and the
border matrix of the inequality constraint, respectively.
By minimizing the cost function ๐ฝ the optimal solution ๐ฅ๐๐๐ก can be achieved under the system constraints.
3.1 Output Prediction
Referring to [24], we can express the multi-step output
prediction as
๏ฟฝโ๏ฟฝ๐+1 = ๐ปโ๏ฟฝโโ๏ฟฝ๐ + ๐โ๏ฟฝโโ๏ฟฝ๐โ1 โ ๐๏ฟฝโ๏ฟฝ๐ (7)
where the notation โโ represents โfutureโ, and โโ represents โpastโ in the prediction; ๐ป = ๐ถ๐ดโ1๐ถ๐ , ๐ = ๐ถ๐ดโ1๐ป๐
and ๐ = ๐ถ๐ดโ1๐ป๐ด, with the parameter matrices ๐ถ๐ด, ๐ถ๐, ๐ป๐ด and ๐ป๐ defined as
๐ถ๐ด = [ 1 0 0 โฏ 0๐ด1 1 0 โฏ 0๐ด2 ๐ด1 1 โฏ โฎ โฎ โฎ โฎ โฎ 1๐ด๐๐ฆ โฏ ๐ด2 ๐ด1 1]
, ๐ถ๐ = [๐1 0๐2 ๐1 0 00 0โฎ โฎ๐๐๐ข โฏ ๐1 โฎ๐2 ๐1
],
๐ป๐ด = [ ๐ด1 โฏ ๐ด๐โ๐๐ฆ๐ด2 โฏ ๐ด๐โ๐๐ฆ+1 ๐ด๐โ๐๐ฆ+1 โฏ ๐ด๐๐ด๐โ๐๐ฆ+2 โฏ 0โฎ โฎ โฎ ๐ด๐๐ฆ โฏ ๐ด๐โ1 โฎ โฎ โฎ ๐ด๐ โฏ 0 ]
, and
๐ป๐ = [ ๐2 โฏ ๐๐โ๐๐ข๐3 โฏ ๐๐โ๐๐ข+1 ๐๐โ๐๐ข+1 โฏ ๐๐๐๐โ๐๐ข+2 โฏ 0โฎ โฎ โฎ ๐๐๐ข+1 โฏ ๐๐โ1 โฎ โฎ โฎ ๐๐ โฏ 0 ]
and {๏ฟฝโ๏ฟฝ๐+1 = [๐ฆ๐+1 ๐ฆ๐+2 โฆ ๐ฆ๐+๐๐ฆ]๐ ๏ฟฝโ๏ฟฝ๐= [๐ฆ๐ ๐ฆ๐โ1 โฆ ๐ฆ๐โ๐๐ฆ+1]๐ โ๏ฟฝโโ๏ฟฝ๐ = [โ๐ข๐ โ๐ข๐+1 โฆ โ๐ข๐+๐๐ขโ1]๐
where ๐๐ฆ and ๐๐ข are the prediction horizon and the input
horizon, respectively.
3.2 Standardization of the Cost Function
Typically, the cost function acts as the performance index
of an optimization control process. In our problem, the cost
function consists of four penalty terms, including the terms of
vehicle velocity ๐ฃ, velocity rate โ๐ฃ, acceleration control input ๐ข and its rate โ๐ข. Among them, โ๐ฃ is relative to the velocity
smoothness, ๐ข is the control input and โ๐ข represents โvehicle
jerkingโ. We design our cost function ๐ฝ based on the four
factors, and it is described as
๐ฝ = ๐1(๐ฃ๐โโ โโ ๐+1 โ ๏ฟฝโ๏ฟฝ๐+1)๐(๐ฃ๐โโ โโ ๐+1 โ ๏ฟฝโ๏ฟฝ๐+1) +๐2๏ฟฝโ๏ฟฝ๐+1๐ ๏ฟฝโ๏ฟฝ๐+1 + ๐3๏ฟฝโโ๏ฟฝ๐๐ ๏ฟฝโโ๏ฟฝ๐ + ๐4โ๏ฟฝโโ๏ฟฝ๐๐โ๏ฟฝโโ๏ฟฝ๐ (8)
where ๐๐ (i = 1,2,3,4) are the weights of the four penalty
terms, respectively. Through mathematical deductions, the
optimization objective can be expressed by
minโโ๏ฟฝโโโ๏ฟฝ๐ ๐ฝ =minโโ๏ฟฝโโโ๏ฟฝ๐ โ๏ฟฝโโ๏ฟฝ๐๐ [(๐1 + ๐2)๐ป๐๐ป + (๐3 + ๐4)๐ผ]โ๏ฟฝโโ๏ฟฝ๐ +2โ๏ฟฝโโ๏ฟฝ๐๐[(๐1 + ๐2)๐ป๐(๐โ๏ฟฝโโ๏ฟฝ๐โ1 + ๐๏ฟฝโ๏ฟฝ๐) โ ๐1๐ป๐๐๐+1 +๐3โ๏ฟฝโโ๏ฟฝ๐โ1] (9)
Then, we can convert our problem to a standard QP problem as
defined by (6), with ๐ฅ = โ๏ฟฝโโ๏ฟฝ๐ and
3 Copyright ยฉ 2018 ASME
{๐ = (๐1 + ๐2)๐ป๐๐ป + (๐3 + ๐4)๐ผ ๐ = 2[(๐1 + ๐2)๐ป๐(๐โ๏ฟฝโโ๏ฟฝ๐โ1 + ๐๏ฟฝโ๏ฟฝ๐) โ ๐1๐ป๐๐๐+1 + ๐3โ๏ฟฝโโ๏ฟฝ๐โ1] ๐ = 0
For the space limitation of this paper, the proof of (9) is
omitted here.
3.3 Standardization of the Constraints
In our problem, both the control input ๐ข๐ and its rate โ๐ข๐
are physically limited. They are constrained by {๐ข โค ๐ข๐ โค ๐ข, โ๐ โ๐ข โค โ๐ข๐ โค โ๐ข, โ๐ (10)
where ๐ข and ๐ข are the upper and lower limits of ๐ข๐ ,
respectively; โ๐ข and โ๐ข are the upper and lower limits of โ๐ข๐, respectively. Referring to [25], we can combine the two
constraints by a compact matrix expression as
[๐ถโ๐ข๐ถ๐ข๐ธ] โ๏ฟฝโโ๏ฟฝ๐ โค [๐โ๐ข๐๐ข ] + [ 0โ๐ถ๐ข๐ฟ] ๐ข๐โ1 (11)
where
๐ถ๐ข, ๐ถโ๐ข =[ 1 0 โฏ 0 0 1 โฏ 0 โฎ โฎ โฎ โฎ 0 0 โฏ 1โ1 0 โฏ 0 0 โ1 โฏ 0 โฎ โฎ โฎ โฎ ]
, ๐ธ = [1 01 1 โฏ 0โฏ 0โฎ โฎ1 1 โฎ โฎโฏ 1],
๐ฟ = [11โฎ1], ๐๐ข =[ ๐ขโฎ๐ขโ๐ขโฎโ๐ข]
, and ๐โ๐ข =[ โ๐ขโฎโ๐ขโโ๐ขโฎโโ๐ข]
. 4 STRATEGY OF ADAPTIVE COST FUNCTION
To make the cost function adaptive, a straightforward
method is to adjust the weight value of each penalty term with
the changing of road conditions. Before designing the adaptive
weights, weight priority policies are necessary, which should
reflect the relationship between the four penalty terms and the
road conditions. On the basis of human driving experience, we
propose a set of priority policies as the following.
4.1 Weight Priority Policies with respect of the Road Grade
Four basic policies for road grade are listed below.
(i) For flat roads, the velocity accuracy and the
driving/riding comfort (smoothness) are of high
priority. It is because the flat road is favorable and
also the most common situation.
(ii) For upgrade and downgrade scenarios, the priority of
the velocity accuracy should be decreased, which is
similar in human driving behaviors.
(iii) The fuel economy (acceleration) is less important
except in downgrade scenario.
(iv) Vehicle jerking should be avoided in most of the
scenarios.
4.2 Weight Priority Policies with respect of the Road Friction Coefficient
There are also four basic policies for road friction
coefficient/slip ratio:
(v) For roads with low slip ratio, the velocity accuracy
and the smoothness are both important due to the
advantageous road condition.
(vi) For roads with high slip ratio, e.g. the snowy or icy
road, the velocity accuracy is less important, while
the velocity rate is more important to guarantee the
safety.
(vii) For roads with high slip ratio, the fuel economy is less
important.
(viii)Vehicle jerking should be avoided in most of the
scenarios.
4.3 Determination Method of the Penalty Weights
Define ๐1๐ and ๐2๐ (๐ = 1,2,3,4) as the sub-weights in
terms of the road grade and the friction coefficient/slip ratio,
respectively. The weight priority can be described by Table I
and Table II.
TABLE I WEIGHT PRIORITY WITH RESPECT OF THE ROAD GRADE
Road grade range Typical road Weight priority ฮฃ๐=14 ๐1๐ ๐11 ๐12 ๐13 ๐14 ๐ < โ10ยฐ Downgrade LS LS VR VR 1.0 โ10ยฐ โค ๐ โค 10ยฐ Flat VR LS LS VR 1.0 ๐ > 10ยฐ Upgrade LS LS LS VR 1.0
TABLE II WEIGHT PRIORITY WITH RESPECT OF THE SLIP RATIO
Slip ratio range Typical road Weight priority ฮฃ๐=14 ๐2๐ ๐21 ๐22 ๐23 ๐24 ๐ < 0 Braking LS IM LS VR 1.0 0.0 โค ๐ < 0.2 Newly Constructed IM VR LS VR 1.0 0.2 โค ๐ < 0.5 Dry asphalt LS LS LS VR 1.0 0.5 โค ๐ < 0.8 Snowy/Wet LS LS LS IM 1.0 0.8 โค ๐ โค 1.0 Icy LS VR VR VR 1.0
where the priority notations are defined as
LS: Less important
IM: Important
VR: Very important
The final value of each weight ๐๐ can be calculated by ๐๐(๐, ๐ ) = ๐พ๐ ๐1๐(๐) + (1 โ ๐พ๐ )๐2๐(๐ ), (0 < ๐พ๐ < 1, ๐ = 1,2,3,4) (12)
where ๐พ๐ is the allocation coefficient, which represents the
priorities of the two road condition factors in one problem.
It should be noted that the notation LS, IM and VR are not
exact values but just priorities. In a practical application, the
penalty weights should can be given as exact values (๐1๐(๐)
and ๐2๐(๐ )) and must be consistent with their priorities. In our
design, the total sum of the four penalty weights, i.e. ๐ด๐=14 ๐๐, must be 1.0, which implies a normalization process. In addition,
4 Copyright ยฉ 2018 ASME
The control input rates of the two strategies are shown in
Figure 2(e). It can be seen that the adaptive-weight strategy
behaves better than the fixed-weight strategy, still with a
smaller peak value.
The most important comparison of the two strategies are
presented by control costs. In the preceding optimization
process of MPC, the two curves of the cost function value
present quite different features in the amplitude. As we can
observe clearly in Figure 2(f), the total control cost of the
adaptive-weight strategy is about 49% lower than that of the
fixed-weight strategy. This percent is calculated by
๐ท๐ฝ = โ {๐ฝFixedโweight(๐)โ๐ฝAdaptiveโweight(๐)}๐๐=1 โ ๐ฝFixedโweight(๐)๐๐=1 ร 100% (13)
where ๐ represents the total number of the time instants; ๐
represents the current time instant.
Therefore, the overall performance of the adaptive-weight
strategy is better than that of the fixed-weight strategy because,
for the same control problem the former has a better tracking
performance and at the same time a much lower control cost
than the latter.
7 CONCLUSIONS
In this paper, a road condition adaptive strategy of MPC
control has been devised based on a mechanism of adaptive
cost function. The proposed approach has been applied to the
longitudinal control problem of autonomous vehicles. The MPC
controller is solved after being converted to a QP problem. To
verify the effectiveness of the proposed approach, simulation
has been carried out on a simulated road with time-varying road
grades and friction coefficients. The results show that the
proposed approach is effective in handling different road
conditions. Using the new strategy, lower control cost can be
achieved compared with the strategy of fixed cost function.
This paper presents the application of the road condition
adaptive MPC method on the longitudinal control of
autonomous vehicles. As for future work, we will study to
apply the proposed approach to more other control problems in
autonomous vehicles such as platooning control and lateral
control, to further leverage the research outcomes.
ACKNOWLEDGEMENT
This work was supported by the National Science
Foundation Grant CRII-1755771.
REFERENCES
[1] Bansal, P., and Kockelman, K. M. โForecasting Americansโ long-term
adoption of connected and autonomous vehicle technologies.โ
Transportation Research Part A: Policy and Practice 95, 2017: 49-63.
[2] Rawlings, J. B., and Mayne, D. Q. โModel predictive control: theory and
design.โ Nob Hill Pub., 2012.
[3] Waluล, K. J., and Olszewski, Z. โAnalysis of tire-road contact under
winter conditions.โ Proceedings of the World Congress on Engineering.
(3), 2011.
[4] Borrelli, F., Falcone, P., Keviczky, T., Asgari, J., and Hrovat, D.
โMPC-based approach to active steering for autonomous vehicle systems.โ
International Journal of Vehicle Autonomous Systems, 2005, 3(2-4):
265-291.
[5] Falcone, P., Borrelli, F., Asgari, J., Tseng, H. E., and Hrovat, D.
โPredictive active steering control for autonomous vehicle systems.โ
IEEE Transactions on Control Systems Technology, 2007, 15(3):566-580.
[6] Falcone, P., Borrelli, F., Tseng, H. E., Asgari, J., and Hrovat, D. โLinear
time-varying model predictive control and its application to active
steering systems: stability analysis and experimental validation.โ
International Journal of Robust and Nonlinear Control, 2008, 18(8),
862-875.
[7] Turri, V., Carvalho, A., Tseng, H. E., Johansson, K. H., and Borrelli, F.
โLinear model predictive control for lane keeping and obstacle avoidance
on low curvature roadsโ, Intelligent Transportation Systems-(ITSC), 2013
16th International IEEE Conference, 2013: 378-383.
[8] Carvalho, A., Gao, Y., Gray, A., Tseng, H. E., and Borrelli, F. โPredictive control of an autonomous ground vehicle using an iterative linearization
approach.โ Intelligent Transportation Systems-(ITSC), 2013 16th
International IEEE Conference, 2013: 2335-2340.
[9] Kamal, M. A. S., Mukai, M., Murata, J., and Kawabe, T. โEcological driver assistance system using model based anticipation of
vehicle-road-traffic information.โ IET Journal of Intelligent
Transportation Systems, 2010, 4(4): 244-251.
[10] Yi, B., Gottschling, S., Ferdinand, J., Simm, N., Bonarens, F., and Stiller,
C. โReal time integrated vehicle dynamics control and trajectory planning
with MPC for critical maneuversโ. Intelligent Vehicles Symposium (IV),
2016 IEEE 584-589.
[11] Jalalmaab, M., Pirani, M., Fidan, B., and Jeon, S. โCooperative road
condition estimation for an adaptive model predictive collision avoidance
control strategy.โ Intelligent Vehicles Symposium (IV), 2016 IEEE:
1072-1077.
[12] Beal, C. E., and Gerdes, J. C. โModel predictive control for vehicle
stabilization at the limits of handling.โ IEEE Transactions on Control
Systems Technology, 2013, 21(4): 1258-1269.
[13] Chen, Y., and Wang J. โAdaptive vehicle speed control with input
injections for longitudinal motion independent road frictional condition
estimation.โ IEEE Trans. on Vehicular Technology, 2011, 60(3): 839-848.
[14] Kidambi, N., Harne, R. L., Fujii, Y., Pietron, G. M., and Wang, K.
โMethods in vehicle mass and road grade estimation.โ SAE International
Journal of Passenger Cars-Mechanical Systems, 2014, 7: 981-991.
[15] Han, K., Hwang, Y., Lee, E., and Choi, S. โRobust estimation of
maximum tire-road friction coefficient considering road surface
irregularity.โ International Journal of Automotive Technology, 2016,17(3):
415-425.
[16] Mahyuddin, M. N., Na, J., Herrmann, G., Ren, X., and Barber, P.
โAdaptive observer-based parameter estimation with application to road
gradient and vehicle mass estimation.โ IEEE Transactions on Industrial
Electronics, 2014, 61(6): 2851-2863.
[17] Rajamani, R., Phanomchoeng, G., Piyabongkarn, D., and Lew, J. Y.
โAlgorithms for real-time estimation of individual wheel tire-road friction
coefficients.โ IEEE/ASME Transactions on Mechatronics, 2012, 17(6): 1183-1195.
[18] Liu, Y. H., Li, T., Yang, Y. Y., Ji, X. W., and Wu, J. โEstimation of tire-road friction coefficient based on combined APF-IEKF and iteration
algorithm.โ Mechanical Systems and Signal Processing, 2017, 88: 25-35.
[19] Sahlholm, P., and Johansson, K. H. โRoad grade estimation for
look-ahead vehicle control using multiple measurement runs.โ Control
Engineering Practice, 2010, 18(11): 1328-1341.
[20] Son, Y. S., Kim, W., Lee, S. H., and Chung, C. C. โRobust multirate
control scheme with predictive virtual lanes for lane-keeping system of
autonomous highway driving.โ IEEE Transactions on Vehicular
Technology, 2015, 64(8): 3378-3391.
[21] Li, S., Li, K., Rajamani, R., and Wang, J. โModel predictive
multi-objective vehicular adaptive cruise control. IEEE Transactions on
Control Systems Technology.โ 2011, 19(3): 556-566.
[22] Biron, Z. A., and Pisu, P. โDistributed fault detection and estimation for
cooperative adaptive cruise control system in a platoon.โ In PHM
Conference, 2015.
[23] Shakouri, P., Ordys, A., and Askari, M. R. โAdaptive cruise control with
stop&go function using the state-dependent nonlinear model predictive
control approach.โ ISA Transactions, 2012, 51(5): 622-631.
[24] Rossiter, J. A., and Kouvaritakis, B. โConstrained stable generalised
predictive control.โ In IEE Proceedings D-control Theory and
Applications 1993, 140 (4): 243-254.
[25] Rossiter J. A. Model-based predictive control: a practical approach. CRC
press, 2003.
6 Copyright ยฉ 2018 ASME