3612 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO.

8, OCTOBER 2014

Comparison of Feedback Control Techniques

for Torque-Vectoring Control of
Fully Electric Vehicles
Leonardo De Novellis, Member, IEEE, Aldo Sorniotti, Member, IEEE, Patrick Gruber, and Andrew Pennycott

Abstract—Fully electric vehicles (FEVs) with individually con- control system level as the traditional tools for handling tuning
trolled powertrains can significantly enhance vehicle response to (e.g. the adjustment of the suspension parameters) produce a
steering-wheel inputs in both steady-state and transient condi- marginal effect compared with the continuous TV actuation of
tions, thereby improving vehicle handling and, thus, active safety
and the fun-to-drive element. This paper presents a compari- the electric motor drives.
son between different torque-vectoring control structures for the Full exploitation of the benefits of TV control for FEVs
yaw moment control of FEVs. Two second-order sliding-mode is only feasible with the implementation of more advanced
controllers are evaluated against a feedforward controller com- controllers than those used in vehicle stability control systems
bined with either a conventional or an adaptive proportional– today. Current systems are typically designed for friction brake
integral–derivative (PID) controller. Furthermore, the potential
performance and robustness benefits arising from the integration actuation in emergency conditions without any particular re-
of a body sideslip controller with the yaw rate feedback control quirement for smoothness of the intervention and, therefore, ve-
system are assessed. The results show that all the evaluated con- hicle comfort or fun-to-drive [2]. However, stability controllers
trollers are able to significantly change the understeer behavior are evolving toward smooth continuous operation, particularly
with respect to the baseline vehicle. The PID-based controllers during significant braking and acceleration maneuvers. In this
achieve very good vehicle performance in steady-state and tran-
sient conditions, whereas the controllers based on the sliding-mode respect, in [3] alternative control structures for FEV TV con-
approach demonstrate a high level of robustness against variations trol, with particular focus on linear parameter-varying gain-
in the vehicle parameters. The integrated sideslip controller effec- scheduled control, are discussed. The vehicle stability control
tively maintains the sideslip angle within acceptable limits in the systems installed in modern passenger cars are mainly based on
case of an erroneous estimation of the tire–road friction coefficient. feedback yaw rate controllers that intervene to recover a sig-
Index Terms—Fully electric vehicle (FEV), sideslip angle, nificant yaw rate deviation when a tolerance threshold between
torque-vectoring (TV) control, yaw rate. the reference value (mainly dependent on steering-wheel angle,
vehicle velocity, and the estimated friction coefficient) and the
I. I NTRODUCTION measured value is exceeded. The usual method for evaluating
the corrective yaw moment is a feedback controller for which
I NDIVIDUALLY controlled powertrains in fully electric
vehicles (FEVs) allow significant improvements not only in
terms of vehicle architecture, packaging, and energy manage-
the gains can be computed through various methodologies such
as the Riccati equation [2]. In [4], a linear matrix inequality-
ment but also from the viewpoint of vehicle dynamics design. based method for the design of a gain-scheduled yaw rate
The precise and highly responsive torque control of individual controller is presented. Other publications deal with advanced
electric motor drives can have a major impact on the vehicle robust control techniques, such as generalized predictive
steady-state and transient handling response characteristics. control [5].
Torque vectoring (TV) actuated through individual motor con- Furthermore, the controllers can regulate vehicle sideslip
trol is more effective than TV based on active differentials. angle β, which can be estimated through an extended Kalman
In fact, TV differentials have limitations in terms of dynamic filter. The sideslip angle controllers are based on the theory of
response, maximum allowable torque transfer, efficiency, and Shibahata (see the analyses in [6] and [7] for an overall idea),
flexibility in the torque transfer direction in the case of a signif- which demonstrates that the achievable vehicle yaw moment
icant speed difference between the sun gears [1]. The design of Mz , controlled by the driver through the steering-wheel angle,
the cornering response of FEVs can be carried out mainly at the is a decreasing function of the magnitude of β. As a conse-
quence, at high values of β, the driver no longer perceives a
significant effect of the steering-wheel angle on the vehicle
Manuscript received January 19, 2013; revised July 9, 2013 and response. This can lead to driver reactions that can degrade
December 10, 2013; accepted January 28, 2014. Date of publication vehicle stability.
February 10, 2014; date of current version October 14, 2014. This work was The potential of electric motor drives in TV control can be
supported by the European Union’s Seventh Framework Programme FP7/2007-
2013 under Grant 284708. The review of this paper was coordinated by exploited through the adoption of feedforward controllers that
Prof. T. Shim. generate continuous reference yaw moments. The controllers
The authors are with the University of Surrey, Surrey GU2 7XH, U.K. must be coupled with feedback control structures that produce
(e-mail: [email protected]; [email protected]; p.gruber@surrey.
ac.uk; [email protected]). a yaw moment contribution starting from a continuous refer-
Digital Object Identifier 10.1109/TVT.2014.2305475 ence yaw rate, which, in turn, is based on precise tire–road
0018-9545 © 2014 IEEE. Translations and content mining are permitted for academic research only. Personal use is also permitted, but republication/redistribution
requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
DE NOVELLIS et al.: COMPARISON OF FEEDBACK CONTROL TECHNIQUES FOR TV CONTROL OF FEVS 3613

δ are commonly measured by on-board sensors and inertial

platforms. Vehicle speed V and friction coefficient μ at the
tire–road contact are estimated quantities. Since the estimation
of vehicle speed, friction coefficient, and sideslip angle is not
the subject of this paper, it is assumed that this estimation
can be satisfactorily implemented (or unsatisfactorily, in the
case of failure in tire–road friction estimation, as discussed in
Sections IV and VI).

Fig. 1. Scheme of the proposed TV-controlled FEV. A. Design Requirements

friction coefficient estimation. Moreover, the feedback part of In conditions of constant velocity, the adoption of a TV
the controller, which is responsible for the generation of the control system allows the achievement of a reference under-
reference yaw moment correction, must be smooth and robust steer characteristic significantly different to that of the baseline
against variations in the vehicle characteristics such as mass vehicle (i.e. the passive vehicle with constant torque split).
and yaw moment of inertia, and tire parameters and operating For example, the understeer gradient at low to medium levels
conditions, such as tire–road friction. Sliding-mode control is of lateral acceleration can be reduced to improve the fun-to-
well-suited for this application since it can cope with systems drive aspect, and the linear region of the vehicle response
characterized by significant uncertainty, and at the same time, (representing an approximately constant understeer gradient)
its implementation does not require highly complex control can be extended to enhance vehicle controllability in cornering
laws. Although several publications have presented different conditions. Moreover, the maximum achievable lateral accel-
formulations of sliding-mode controllers applied to vehicle yaw eration can be increased, as demonstrated by previous studies
moment control (see [8] and [9]), a direct comparison of the (see [13] and [14]). In any case, the controller has been de-
performance of these control structures with conventional feed- signed to work according to different driving modes (“normal,”
back controllers based on the proportional–integral–derivative “sport,” and “economy”), selectable by the driver, each of them
(PID) algorithm is lacking. corresponding to a different set of understeer characteristics.
In this paper, the structure of the yaw rate controller and The results of this paper refer to the “sport” driving mode.
its requirements for vehicle implementation are presented and In conditions of combined cornering and acceleration/
described, together with a preliminary analysis of the yaw rate deceleration, the TV system is required to compensate, as much
frequency response characteristics obtained from a single-track as possible, the variation in vehicle response. This variation
model under various linearization conditions. This aspect re- with traction/braking is caused by the longitudinal load transfer
presents a point of novelty with respect to previous publications between the two axles and the interaction between lateral and
on the same subject [6], [10] and constitutes a useful tool for the longitudinal tire forces, according to the friction ellipse concept
TV control system designer. Furthermore, two different sliding- [15]. On conventional vehicles (without TV), the compensation
mode algorithms for yaw rate control are implemented and of the variation in vehicle behavior is carried out by the driver
assessed against conventional PID and adaptive PID algorithms. through the steering-wheel angle. The parameters of the refer-
The evaluation is carried out through an extensive analysis of ence understeer characteristics, such as the understeer gradient
maneuvers simulated by means of an experimentally validated and the maximum value of lateral acceleration, are selected
vehicle model. Finally, the integration of yaw rate and sideslip according to the longitudinal acceleration of the vehicle [16],
angle controllers based on sliding mode into a single control [17]. Finally, for consistent vehicle behavior in transient ma-
structure is presented, and the effectiveness of this structure in neuvers, the variation in the frequency response characteristic
limiting sideslip angle when the friction estimation required for of the system due to changing operating conditions has to be
yaw rate control fails is demonstrated by means of simulations. compensated by the controller.
To ensure the feasibility of the actual implementation of such
a control system, the design has to cope with actuator satura-
II. YAW R ATE C ONTROL S TRUCTURE
tion, which determines the maximum and minimum threshold
The structure of the yaw rate controller is shown in Fig. 1. values of the applicable yaw moment; these are variable and de-
The outputs of the controller are the wheel torques, which must pend on the operating conditions, such as the tire–road friction
be delivered by the electric motors and the friction brakes (when coefficient (see the simplified scheme represented in Fig. 1), the
necessary) to achieve the reference yaw rate. maximum torque deliverable by the motors, and vehicle speed.
The controllers are designed for implementation on real The dependence on speed is indirect, as the maximum torque
vehicles developed within the European research project of the electric motors is a function of the motor speed beyond
E-VECTOORC (Electric Vehicle Control of Individual Wheel the base speed. Finally, the threshold limit value of the sideslip
Torque for On- and Off-Road Conditions) [11], [12]. Therefore, angle βT H , which is evaluated, e.g. for safety reasons with the
the design specifications and physical limits of the actuators and approaches presented in [17] and [18] or chosen according to
the energy storage unit have been carefully taken into account. the desired application (e.g. to achieve high levels of sustained
Yaw rate r, longitudinal acceleration ax , accelerator pedal and stable sideslip following the specification of the car maker),
position pa , brake pedal position pb , and steering-wheel angle should not be exceeded.
3614 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 8, OCTOBER 2014

torque (calculated through a transfer function), and the windage

losses (experimentally measured maps). The efficiency maps of
the motor drives have been provided by the manufacturer. The
vehicle dynamics model has been validated against extensive
experimental tests in steady-state and transient conditions car-
ried out at the proving ground in Lommel, Belgium [17].
For the design of the feedforward contribution of the con-
troller, a simpler vehicle model has been considered based on a
quasi-static formulation [17]. This consists of a set of algebraic
equations based on kinematic and equilibrium relationships.
The model considers eight degrees of freedom corresponding
to the translational motions in the longitudinal and lateral direc-
tions, yaw and roll motions, and the four rotational motions of
the wheels and the front drivetrains. By assuming that the time
Fig. 2. Functional schematic of the case-study FWD electric vehicle.
derivatives of the main state variables (e.g. vehicle sideslip,
TABLE I longitudinal slip ratio, and roll angle) are zero, the quasi-static
M AIN E LECTRIC V EHICLE PARAMETERS model allows the computationally demanding forward time
integration of the equations of motion to be omitted. This is
particularly beneficial when static maps of the main physical
quantities are needed for the automated derivation of the lookup
tables for the feedforward controller. The quasi-static model has
been validated against simulations with the CarMaker model
and experimental data [14], [17].

C. Preliminary Analysis of the Vehicle Dynamics

The yaw rate transfer function of the vehicle without TV
TABLE II control (the so-called “baseline vehicle”) can be obtained from
T IRE PARAMETERS FOR H IGH -F RICTION C ONDITIONS the equations describing the lateral force and yaw dynamics of
the single-track vehicle model [20]. Thus

Jz ṙ = Nβ β + Nr r + Nδ δw
(1)
mV (r + β̇) = Yβ β + Yr r + Yδ δw .

In this equation, δw is the mean steer angle of the front wheels.

The stability derivatives can be expressed as a function of the
front and rear cornering stiffnesses, i.e. CF and CR , respec-
tively, as
B. Plant and the Simulation Models
aCF − b CR
Yβ = CF + CR ; Yr = ; Yσ = −CF
The case-study vehicle is a high-performance front-wheel- V
drive (FWD) sport utility vehicle. The drivetrain layout, shown a 2 C F + b2 C R
in Fig. 2, consists of one on-board switched reluctance electric Nβ = a C F − b C R ; Nr = ; Nδ = −aCF .
V
motor per wheel. The powertrains are connected to the wheels
through single-speed transmissions and half-shafts with con- By rearranging the equations, the transfer function of the yaw
stant velocity joints. rate dynamics becomes
The main vehicle characteristics are reported in Table I. 
r  Ds + E
Moreover, two different commercially available tires have been (s) = 2 + Bs + C
(2)
considered to assess the robustness of the yaw moment con- δw baseline As
troller: Tire A is comfort oriented and Tire B is sports oriented. where coefficients A − E are functions of the axle cornering
Specific tire parameters are provided in Table II. stiffness (defined as the incremental ratio of the lateral force to
To evaluate the vehicle dynamics behavior, a simulation the mean slip angle of the wheels on the same axle), the yaw
model in the time domain was created through integration of the moment of inertia, the vehicle mass, and the semi-wheelbases.
IPG CarMaker chassis module into the Simulink environment Thus
adopted for the development of the drivetrain model. The
electric drivetrain dynamics have been considered by taking A = mV Jz ; B = −(mV Nr + Yβ Jz )
into account the stiffness of the half-shafts and the backlash
C = mV Nβ + Nr Yβ − Nβ Yr ; D = mV Nδ
in the transmission elements [19]. The electric motor drive has
been modeled by including its torque slew rate, the air gap E = Nβ Y δ − Nδ Y β .
DE NOVELLIS et al.: COMPARISON OF FEEDBACK CONTROL TECHNIQUES FOR TV CONTROL OF FEVS 3615

Fig. 5. Yaw rate frequency response evaluated for the baseline vehicle at ay =
4 m/s2 and different values of longitudinal acceleration, from ax = −2.5 m/s2
to ax = 2.5 m/s2 , in steps of 1.25 m/s2 . The arrows point in the direction of
increasing ax values. Tire A has been used.

Fig. 3. Cornering stiffness of the front axle evaluated with the quasi-static
model as a function of lateral acceleration for different values of longitudinal
acceleration. Tire A has been used. The different trend of CF for positive and
negative values of ax is due to the effect of the longitudinal load transfer in
traction and braking conditions.

Fig. 6. Yaw rate frequency response evaluated for the controlled vehicle with
constant-gain PID and feedforward contributions at different values of lateral
acceleration, from ay = 0 m/s2 to ay = 6 m/s2 , in steps of 1.5 m/s2 . The
arrows point in the direction of increasing ay values. Tire A has been used.

In the case of the controlled vehicle subject to a control yaw

moment Mz , the yaw rate in the Laplace domain (s denoting
the Laplace variable) can be expressed as

Fig. 4. Yaw rate frequency response evaluated for the baseline vehicle at r  r
r(s) = (s) δw + (s)Mz (s)
different values of lateral acceleration, from ay = 0 m/s2 to ay = 6 m/s2 , in
steps of 1.5 m/s2 . The arrows point in the direction of increasing ay values.
δw baseline Mz
Tire A has been used. (Ds + E)δw (F2 s + F1 )Mz (s)
= 2
+ (3)
As + Bs + C As2 + Bs + C
The variation of the vehicle frequency response as a function
of the operating conditions in terms of longitudinal and lateral with F2 = mV and F1 = −Yβ . For example, Mz can in-
accelerations is discussed further below (see Figs. 4 and 5). This clude a feedforward contribution Mz,f f (s) and the contribution
is achieved through variation of the terms CF and CR defining Mz,P ID (s) of a PID controller. In this research, the feedfor-
the coefficients of (2). In particular, by employing the quasi- ward contribution is the main element responsible for achieving
static model described in Section II-B, it is possible to evaluate the reference steady-state gain in conjunction with the PID
the cornering forces at the front and rear axles for the FWD controllers. Therefore, in this linearized example, Mz,f f can be
case-study vehicle in different operating conditions. Therefore, obtained starting from the equations of the linearized system.
at discrete points, the front- and rear-axle cornering forces have Thus
 
been locally linearized to evaluate the cornering stiffness of the Crref δw
baseline vehicle for the front and rear axles as a function of the Mz,f f = −E . (4)
δw F1
longitudinal and lateral accelerations in trimmed conditions, as
shown in Fig. 3. The cornering stiffness of the front axle is not The contribution of the feedback controller is Mz,P ID (s) =
subjected to a significant variation in braking conditions, as the kP ID (s)(r(s) − rref (s)), where rref is the reference yaw
reduction induced by the braking force is compensated for by rate. rref can be a constant rref,s in the frequency domain,
w
the increase due to the load transfer from the rear to the front for example, given by rref,s = δw /(V KU,ref + L/V ), where
w
axle. KU,ref = (∂δw /∂ay )ref is the reference value of the under-
The effect of lateral acceleration ay on the yaw rate dynamics steer gradient at the wheel. Alternatively, the reference yaw rate
(i.e. the yaw rate gain) is shown in Fig. 4. The steady-state gain can be expressed as a first-order transfer function [18] such as
decreases from 7.3 to 5.5 s−1 , and the damping ratio of the rref (s) = rref,s /(1 + τ s), where τ is the time constant of the
system varies from 0.9 to 0.8. At constant velocity, phase angle first-order yaw rate reference. The transfer function of the yaw
φ decreases with lateral acceleration, particularly for frequency rate gain for the controlled vehicle including feedforward and
values larger than approximately 0.5 Hz. Vehicle understeer PID is then
  
increases with longitudinal acceleration ax , since the steady- Mz,f f rref 
r f z (s) (s) + (s)k P ID (s) + r
(s) 
state gain decreases from 10.6 to 4.5 s−1 ; at frequencies larger =
δw δw δw baseline
than about 1.2 Hz, the yaw rate transfer functions tend to δw 1 + fz (s)kP ID (s)
overlap (see Fig. 5). Furthermore, for low values of frequency, (5)
the delay between the wheel steer angle and the yaw rate with fz (s) = (r/Mz )(s).
decreases with increasing ax since the phase angle tends toward Figs. 6 and 7 show the frequency response of the overall
larger values. system evaluated through (5) at different values of ax and ay .
3616 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 8, OCTOBER 2014

the control design requirements, as previously mentioned in

Section II-A, as functions of ax and μ. In particular, the same
value of KU at different values of the longitudinal accelera-
tion has been considered, to achieve the compensation of the
variation of the understeer gradient in traction and braking
conditions. Using (6), the steady-state value of the reference
yaw rate is given by
Fig. 7. Yaw rate frequency response evaluated for the controlled vehicle ay
with constant-gain PID and feedforward contributions at different values of
rref,S = . (7)
V
longitudinal acceleration, from ax = −3.75 m/s2 to ay = 3.75 m/s2 , in steps
of 1.25 m/s2 . The arrows point in the direction of increasing ax values. Tire A Equations (6) and (7) allow the generation of a lookup table
has been used. for the reference yaw rate as a function of vehicle speed V ,
The reference yaw rate has been obtained at V = 90 km/h steering-wheel angle, longitudinal acceleration, and friction co-
w
considering the value of KU,ref = 0.04 degs2 /m constant with efficient at the tire–road contact. To prevent excessive control
ax and ay . The following gains have been adopted: kP = action for high-frequency steering-wheel inputs, the output of the
80 kNms/rad, kI = 0.004 Nm/rad, and kD = 0.8 Nms2 /rad. lookup table is multiplied by a first-order factor with τ = 0.3 s.
The variation in the yaw rate transfer function is strongly
attenuated, except at extreme values of longitudinal and lateral E. Control Strategies
accelerations, compared with the results shown in Figs. 4 and 5. Four different approaches for the design of the yaw moment
Moreover, the controlled system is characterized by a general controller generating MZ are considered and evaluated as
reduction in the magnitude of the phase angles. Additionally, follows:
as the variation of the vehicle cornering response is very well
• PID control with feedforward contribution;
compensated (see Figs. 6 and 7), a gain scheduling approach
• adaptive PID control with feedforward contribution [22];
as a function of ax and ay is expected to produce only a small
• second-order sliding-mode (SOSM) control based on the
improvement relative to a constant gain PID controller (with
suboptimal algorithm (without feedforward contribution)
respect to ax and ay ), which will be adopted in the rest of
[23]–[25];
this paper. Some gain scheduling of the PID gains is required
• SOSM control based on the twisting algorithm (without
as a function of vehicle speed in any case (when considering
feedforward contribution) [26], [27].
maneuvers with a significant speed range), but this is already
well known [10] and not relevant to this paper, which focuses To limit the analysis to the objective comparison of the
on maneuvers from an initial vehicle speed of 90 km/h and four yaw moment control strategies, this paper is focused on
covering a relatively limited speed range. These preliminary maneuvers that do not require friction brake intervention (i.e.
conclusions will be verified in the following sections, in which traction conditions), such that the assessment of the controllers
the limitations of the actuated Mz (which can significantly is independent from the wheel torque distribution algorithm and
change the results) and the anti-windup conditions on the the brake-blending algorithm. As a consequence, the drivetrain
controller are introduced. torques at the front wheels, i.e. TLF and TRF , are expressed as
Mz R
TLF = Tm −
D. Reference Yaw Rate c
Mz R
This section describes the methodology for the definition TRF = Tm + (8)
c
of rref,S for the controller implemented on the real vehicle
with nonlinear dynamics. A target understeer characteristic is where R is the laden radius of the tire, and Tm is the mean
defined in terms of an analytical function relating the dynamic wheel torque necessary for vehicle propulsion. Tm is evaluated
steering-wheel angle δdyn = δ − δkin (where δ is the actual by the vehicle drivability controller starting from the accelerator
steering-wheel angle, and δkin is the kinematic steering-wheel pedal position pa defined by the driver model (see Fig. 1). The
angle) to the lateral acceleration ay . Therefore, (6) is proposed, end goal of this paper is to see whether there is any real benefit
which is based on the following three characteristic parameters: that can be derived from the additional robustness of adaptive
the understeer gradient KU = ∂δdyn /∂ay ; the threshold value PID or sliding-mode formulations (recently adopted by several
a∗y , which defines the upper limit of the linear part of the academic authors) against a simple PID with constant gains, for
understeer characteristic; and the maximum lateral acceleration maneuvers in a well-defined speed range.
achievable in trimmed conditions ay,M AX . Thus
⎧ 1
⎪
⎪ δdyn , if δdyn < a∗y KU III. D ESIGN OF THE YAW R ATE C ONTROLLERS
⎨ KU ∗
ay,M AX + ay − ay,M AX A. PID Controller
ay = (6)
⎪
⎪
KU a ∗
y −δdyn
⎩ ∗
The feedforward contribution working together with the PID
× e (ay,M AX −ay )KU , if δdyn ≥ a∗y KU
controller is based on lookup tables in terms of δ, ax , and μ.
where ay,M AX = ay,M AX (μ, ax ), and a∗y = a∗y (μ, ax ). The The tables were obtained with the quasi-static vehicle model
terms KU , a∗y , and ay,M AX can be chosen according to [14] using an optimization routine to derive the control yaw
DE NOVELLIS et al.: COMPARISON OF FEEDBACK CONTROL TECHNIQUES FOR TV CONTROL OF FEVS 3617

moment that yields the desired set of understeer curves [17]. In Depending on the relative degree between the sliding variable
traction conditions, the reference characteristics have a reduced S(x, t) and the control input u(t), v(t) may represent either the
understeer gradient, a wider linear region, and a larger maxi- actual control action or its time derivative [29]; in the former
mum value of lateral acceleration with respect to the baseline case (relative degree 2), the control law will be discontinu-
vehicle. ous, whereas the latter case (relative degree 1), which is also
The constant gains of the conventional PID have been tuned referred to as the antichattering case, leads to a continuous
starting from the frequency response characteristic of the open- control action [23], [24]. Hence, for the purpose of vehicle
loop system through specifications in terms of the phase and yaw moment control, system (10) is described as a relative
gain margins, and the requirements of the closed-loop tracking degree-1 formulation, so that chattering, which is the main issue
bandwidth. An anti-windup system has been included for the of sliding-mode control applications, is prevented. In fact, the
management of the control action saturation [28]. discontinuity is transferred to the time derivative of the yaw
moment and is absent on the yaw moment itself, calculated by
B. Adaptive PID Controller integration in the time domain. This is ideal for the application
as it allows conjugating robustness without chattering. The
The adaptive PID control proposed in [22] is used here single-track model (see [8] and [30] is recalled to define the
for the vehicle yaw moment control. The particular control sliding surface as Sr = r − rref . Thus
formulation derives from the theory of adaptive interaction and
an approximation of the Frechet tuning algorithm. The gains (aFyF (t) − bFyR (t) + Mz (t))
ẏ1 = Ṡr (t) = − ṙref (t)
of the three-term controller (kP , kD , and kI ) are updated as a Jz
function of the yaw rate error: ẏ2 = S̈r (t) = Φ + Γv(t) (12)

k̇P = − γP e
2
with Γ = 1, v(t) = Ṁz (t)/Jz , and Φ = (aḞyF (t)−bḞyR (t))/
k̇I = − γI e e dt Jz − r̈ref (t). According to the suboptimal sliding-mode algo-
rithm, the control law assumes the form
k̇D = − γD eė. (9)
Ṁz (t) = −Jz kr sign [Sr (t) − 0.5Sr (tM k )] (13)
The initial values for the gains are calculated according to
the conventional guidelines and principles for system stability where tM k is defined as the time corresponding to the last
and tracking bandwidth outlined in the previous section. The singular value of Sr (t), i.e. Ṡr (tM k ) = 0, with the constraint
relevant adaptation coefficients were determined through a kr > 2Φ.
sensitivity analysis. The conditions for the Lyapunov stability After algebraic manipulation and using a simple expression
of this control structure are presented in [22]. The system must for the reference yaw rate, Φ becomes
have a structure of the form ay (n) = u, with the following  
conditions: 1) a > 0; 2) dominant order n ≤ 2; 3) availability Φ = β̇(aCF − bCR ) − δ̇w aCF /Jz − δ̈w /(V KUw + L/V ).
of signals y and y (1) ; and 4) the reference trajectory and its (14)
first n derivatives are bounded and piecewise continuous. These In (14), the time derivatives of the wheel steer angle δw depend
properties are satisfied for the specific application as Jz > 0 on the variation of the actual steering-wheel input. These terms
and known (with some uncertainty depending on the payload), can be very large compared with the other terms of (14).
and the overall yaw moment due to vehicle longitudinal and This has been verified through vehicle dynamics simulations,
lateral forces can be considered as the input u to the system. In including the computation of (14) during typical transient ma-
the specific project, the stability of the adaptation scheme has neuvers. In particularly critical situations, such as the step steer,
been empirically verified during the tuning process. the terms in (14) assume very significant values during the
application of the steering-wheel input. However, they tend to
C. Suboptimal SOSM rapidly decrease following the main input application event. As
a result, a neighborhood of the sliding surface Sr = 0 can be
The main advantage of SOSM control is the ability to achieve  is satisfied for relatively
found, in which the condition kr > 2Φ
the robustness typical of conventional sliding-mode control
small values of kr . Such a sliding mode has only a local
while avoiding control input chattering, which can compromise
attraction property in the sense that the sliding motion of the
vehicle comfort and drivability. Considering a double integrator
system on the surface occurs after a certain time delay. On the
system including state variables y1 (t) and y2 (t)
other hand, the limits of the achievable control yaw moment due
ẏ1 (t) = y2 (t) to actuator saturation demand a careful choice of the control
ẏ2 (t) = Φ(·) + Γ(x, t)v(t) (10) gain, i.e. kr , to prevent bang-bang control actuation due to the
frequent actuator saturation. The saturation of Mz is considered
in a neighborhood of the sliding manifold S(x, t) = y1 (t) = 0, by implementing the desaturation strategy presented in [32]
the uncertain terms Φ and Γ are bounded by known positive ⎧
constants, i.e. ⎨ Mz,M AX , Mz (t− s ) > Mz,M AX
Mz ts = Mz,M IN , Mz (t−
+
s ) < Mz,M IN
|Φ(·)| ≤ Φ̃ ⎩
Mz (t−
s ) , M −
z,M AX ≥ Mz (ts ) ≥ Mz,M IN
0 < Γ̃1 ≤ Γ(x, t) ≤ Γ̃2 . (11) (15)
3618 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 8, OCTOBER 2014

with ts being the switching time, namely, Sr (ts ) = 0.5Sr (tM k ).

In the actual implementation of the controller, a value kr =
20 rad/s3 is chosen. In the practical implementation, some
residual oscillations in the sliding-mode controller can occur,
particularly if the sampling time of the yaw rate signal is
quite high (e.g. 10 ms) or different from that of the reference
yaw rate. This phenomenon is similar to the so-called “ringing
effect” that is typical of the digital implementation of sliding-
mode control. A steep saturation function has been adopted for
approximating the sign function in (13) and to damp out the
residual oscillations [30].

D. Twisting SOSM Fig. 8. Qualitative trend of the phase plane trajectories. The stability region is
indicated with dashed lines.
A description of the twisting algorithm used in this research
is presented in [26] and [27]. The implementation follows the The sideslip angle controller, which is implemented as a
same approach as the suboptimal sliding mode, with the control sliding-mode algorithm, aims at a sliding motion on the surface
law given by the relationship Sβ = 0, with

−km Jz sign (Sr (t)) , Sr Ṡr (t) ≤ 0 Sβ = β − β T H
Ṁz = (16)
−kM Jz sign (Sr (t)) , Sr (t)Ṡr (t) > 0
FyF (t) + FyR (t)
2 M 2 Ṡβ = − r − β̇T H
with the following constraints: kM > km ; km > (4Γ z,M AX )/ mV
    
(Jz Sr (0)); km > (Φ/Γ1 ); Γ1 kM − Φ > Γ2 km + Φ.
2 
S̈β = Φβ + Γβ vβ (t) (17)
Desaturation strategy (15) satisfies condition |Mz | ≤
Mz,M AX for the application of the twisting algorithm in the where Φβ = ((ḞyF (t) + ḞyR (t))/mV ) − β̈TH −((aFyF (t)−
form (16). Moreover, as Γ 1 = Γ 2 = 1, the same conclusions on bFyR (t))/Jz ), Γβ = 1, and vβ (t) = −Mz,β (t)/Jz . In this
the magnitude of Φ as in Section III-C can be drawn. The values case, the relative degree between the sliding surface Sβ and
of km = 10 rad/s3 and kM = 32 rad/s3 have been selected. the control action Mz,β is 2, and therefore, a resulting discon-
tinuous control law is expected. According to the suboptimal
IV. I NTEGRATED YAW R ATE AND S IDESLIP sliding-mode algorithm, the control law is expressed as
A NGLE C ONTROLLER
Mz,β (t) = Jz kβ sign [Sβ (t) − 0.5Sβ (tM k,β )] . (18)
In general, the value of the sideslip angle can be kept within
the stability limits of the vehicle through a yaw rate controller The choice of βT H and β̇T H can be determined by evaluating
if the friction coefficient at the tire–road contact is accurately the stability regions in the phase plane β − β̇ for the bicycle
estimated and a correct reference yaw rate is generated. Under model (see Fig. 8) and considering a reasonable safety factor.
these conditions, yaw rate feedback control is sufficient to The stability region is then approximated as a rhomboid [18].
ensure safe handling and driving behavior. However, if the To find the boundary of the uncertain term Φβ in (17), the same
friction estimation is erroneous or the calculated reference yaw method adopted in Section III-C for the time derivative of the
rate is excessive for the actual operating conditions, the vehicle cornering forces is applied. The term β̈T H of Φβ , due to the
behavior may become unstable. In such cases, safety can be linear relationship between βT H and β̇T H , is limited, except
ensured if a countermeasure is activated, which prevents the when β = 0 or β̇ = 0, which are verified only for a very short
vehicle sideslip angle from exceeding the stability threshold. amount of time in the actual implementation of the controller.
As previously mentioned, sideslip angle estimation is not the In case of significant deviations from βT H , the vehicle can be
subject of this paper; therefore, it is assumed that this estimation stabilized only by reducing the torque demand, i.e. slowing
can be satisfactorily implemented, even if this is actually a down the vehicle.
challenging task. Some recent publications [35] show potential The final expression of the control yaw moment for the
improvement in sideslip angle estimation achievable through integrated yaw rate and sideslip angle controller has the form
the integration of the information from the Global Positioning
System and the on-board sensors commonly installed in vehi- Mz (t) = ρ1 (Mz,r (t) − Mz,β (t)) + Mz,β (t) (19)
cles equipped with vehicle dynamics control systems.
A novel control algorithm is developed, which combines with Mz,r and Mz,β given, respectively, by (13) and (18), while
yaw rate control through the suboptimal SOSM approach with to ensure a smooth transition between the two different control
sideslip angle control. In contrast to previous research [9], the actions, the term ρ1 is expressed as
central idea here is to have a yaw moment controller based on
ln ρ1 = − ρ2 eβ (20)
yaw rate regulation and a sideslip angle controller that only 
operates when a certain threshold value of the sideslip angle |Sβ |, |β| > |βT H |
eβ = (21)
βT H is exceeded, without decelerating the vehicle. 0, |β| ≤ |βT H |.
DE NOVELLIS et al.: COMPARISON OF FEEDBACK CONTROL TECHNIQUES FOR TV CONTROL OF FEVS 3619

In this analysis, kβ = 5 rad/s2 and ρ2 = 100 rad−1 , and the ver-

tices of the rhomboid region have been chosen as |βT H,M AX | =
5 deg and |β̇T H,M AX | = 24 deg/s [36].

V. A SSESSMENT OF THE P ERFORMANCE OF

THE YAW R ATE C ONTROLLERS

The performance of the yaw rate control systems is assessed

through indexes related to the control error e(t) = r − rref
and the control action u(t) = Mz . In particular, the following
indexes are used during the relevant phases of the selected
maneuvers:
• integral
 tm of the absolute value of the error: IAE = Fig. 9. Understeer characteristics evaluated at V = 90 km/h for (solid line)
0 |e(t)|dt; the baseline vehicle and for the controlled vehicle equipped with different yaw
• integral ofthe time-weighted absolute value of the error: rate controllers in high-friction conditions. Tire A has been used.
t
IT AE = 0 m t|e(t)|dt;
TABLE III
• integral of the absolute value of the control action: P W F VALUES FOR R AMP S TEER T ESTS
t
IACA = 0 m |u(t)|dt.
The IT AE criterion complements the IAE criterion as, for
example, the minimization of the IAE during a step steer can
result in a response with a relatively small overshoot but a
long settling time because the IAE weights all errors equally,
independently of time [33].
To assess the performance of the different controllers in a
simple and objective manner, a dimensionless performance-
weighted function P W F has been defined as
As indicated in Fig. 9, for the majority of the lateral-
PWF = w 2 IT AE + w
1 IAE + w 3 IACA (22) acceleration range, the sliding-mode algorithms perform better
where w 2 = w2 /(rT H t2m ), and w
1 = w1 /(rT H tm ), w 3 = w3 / than the conventional PID and adaptive PID controllers in
(Mz,M AX tm ). Mz,M AX is the maximum feasible absolute terms of tracking the reference cornering behavior. In particular,
value of the control yaw moment. A reasonable threshold value the suboptimal SOSM algorithm allows following the target
of the yaw rate rT H = 0.02 rad/s has been considered; tm is the understeer curve (see dotted-dashed line in Fig. 9) very closely.
time duration of the relevant part of the considered maneuver. In addition, the overall tracking performance is satisfactory for
With such a formulation, w1 + w2 + w3 = 1. The weights have the twisting SOSM algorithm, although a deviation from the
been chosen to prioritize the achievement of the reference yaw reference characteristic is noticeable between δ = 90 deg and
rate with respect to the minimization of the control action. δ = 110 deg, which can lead to an undesired loss of vehicle
Different tests have been simulated with the CarMaker vehi- stability.
cle model, in both steady-state and transient conditions. In the Table III compares the performance of the different con-
model, the yaw moment is generated by the different drivetrain trollers and their robustness during ramp steer maneuvers. In the
torques on the front axle. Robustness is assessed by varying the case of the PID and adaptive PID controllers, close matching is
vehicle weight and the friction coefficient at the tire–road con- achieved due to the efficacy of the feedforward contribution,
tact and considering the two different tire typologies mentioned obtained from the simplified quasi-static model, which repre-
in Table II. sents the overall vehicle with satisfactory accuracy. The PID
controllers without the feedforward contribution would not be
able to provide a good tracking performance of the reference
A. Ramp Steer Maneuver
understeer characteristic during the maneuver. The adaptive
The understeer characteristics of the investigated vehicles PID does not deliver any advantage when compared with the
(i.e. baseline vehicle and the vehicle setups with the four basic PID controller. The PID controller results for μ = 0.5
different feedback controllers) are evaluated with ramp steer were obtained by modifying the feedforward contribution as a
maneuvers (according to [34]) simulated with the Simulink- function of the tire–road friction coefficient (see Section III-A).
CarMaker model at a constant vehicle speed of V = 90 km/h. No adaptation of control parameters is required by the SOSM
For the simulations, the target understeer gradient (at the steer- controllers (which do not require any feedforward contribution).
ing wheel) is selected to be smaller than that of the baseline The suboptimal SOSM shows better performance also in
vehicle, i.e. KU = 12 deg/g. Moreover, the linear section of the asymptotic region of the understeer characteristic, leading
the steering-wheel angle against the lateral-acceleration curve to a smaller error with respect to the reference yaw rate, as
is selected to extend up to a∗y = 7.5 m/s2 , and the asymp- shown in Table III. The suboptimal SOSM is insensitive to the
totic value of maximum lateral acceleration is increased to variation of the tire friction coefficient. However, the tracking
ay,M AX = 9.5 m/s2 . performance is significantly influenced (P W F increases by
3620 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 8, OCTOBER 2014

Fig. 10. Yaw rate r evaluated at V = 90 km/h during the sequence of step Fig. 11. Sideslip angle β evaluated at V = 90 km/h during the sequence of
steer tests for different yaw rate controllers. (Solid line) Actual yaw rate. step steer tests for (solid line) different yaw rate controllers and (dashed line)
(Dashed line) Trend of the reference yaw rate. (Dotted-dashed line) Baseline the baseline vehicle. Tire A has been used.
vehicle behavior. Tire A has been used.

TABLE IV
P W F VALUES FOR S EQUENCE OF S TEP S TEER T ESTS
large steering-wheel angle steps, leading to an increase in the
P W F parameter compared with the other controllers, giving
rise to potential problems in terms of real vehicle drivability and
safety. However, when the tire type is changed or when the ve-
hicle mass is increased (μ = 1, m = 2363 kg), the feedforward
contribution (used in the case of the PID controllers but absent
in the sliding-mode controllers) loses its effectiveness, whereas
the suboptimal SOSM algorithm generally shows good adap-
tation capabilities. On average, P W F is marginally higher for
the suboptimal SOSM than for PID; however, the variance of
a factor of 6 for Tire B) by the variation of vehicle mass, the P W F among the six test cases is lower for the suboptimal
which also affects the control performance when different tires SOSM controller.
are used. In any case, even when Tire B is considered with Fig. 11 shows that the value of the sideslip angle for the
the increased vehicle mass, the tracking performance of the controlled vehicle is moderate and smaller than that of the
suboptimal SOSM remains marginally better than that achieved baseline vehicle at large yaw rate values. In contrast, at smaller
by the PID controllers. No significant effect of the tire type on yaw rates, the controlled vehicle shows larger values of sideslip
the suboptimal controller performance can be observed for the angle than the baseline vehicle since the relevant target under-
nominal value of vehicle mass. steer behavior is different (i.e. KU,T V < KU,baseline ).

B. Sequence of Step Steer Maneuvers C. Tip-In During Cornering

The vehicle behavior in transient conditions over a wide The tip-in during cornering maneuver consists of a step
range of yaw rate values, and up to the saturation point cor- applied to the accelerator pedal while the vehicle is negotiating
responding to the asymptote of the understeer characteristic, is a turn. This test is used to investigate the variation in yaw
evaluated with a sequence of fast-steering inputs of increasing rate and lateral acceleration after a step in the torque demand.
amplitude. The values of the actual yaw rate and the reference As previously explained, one of the main tasks of this control
yaw rate for each feedback controller are shown in Fig. 10. The system is to reduce the steady-state and dynamic variations in
trend in the yaw rate of the baseline vehicle is also overlapped the understeer characteristic with ax as much as possible.
and plotted as dotted-dashed lines. Fig. 12 provides a comparison between suboptimal SOSM,
Table IV presents the values of P W F for the sequence of PID, and the baseline vehicle in terms of ay , r, and β, eval-
step steer tests carried out with different tire–road friction co- uated for an initial speed of V = 90 km/h with a step in
efficients (0.5 and 1), vehicle mass values (1963 and 2363 kg), torque demand, causing a sudden increase in the longitudinal
and two tire types (A and B). As previously specified, the refer- acceleration of the vehicle of nearly 3 m/s2 . This provokes a
ence yaw rate and the feedforward contribution of the reference significant perturbation in the yaw rate, lateral acceleration, and
yaw moment were updated according to the different fric- sideslip response of the baseline vehicle between 4 and 5 s. The
tion conditions. Under nominal operating conditions (μ = 1, phenomenon significantly affects the vehicle safety and comfort
m = 1963 kg), the PID and adaptive PID show smaller values aspects. In real driving conditions, the driver would need to
of P W F than the sliding-mode algorithms. Indeed, in the apply corrective actions via the steering wheel. The suboptimal
case of the sliding-mode controllers, large overshoots (sub- SOSM allows the full compensation of the transient, with a
optimal SOSM) and oscillations (twisting SOSM) occur at performance benefit over the PID controller.
DE NOVELLIS et al.: COMPARISON OF FEEDBACK CONTROL TECHNIQUES FOR TV CONTROL OF FEVS 3621

Fig. 15. Yaw rate r evaluated at V = 90 km/h and μ = 0.5 during the
sequence of step steer tests with (left) the suboptimal SOSM yaw rate controller
and (right) the integrated yaw rate-sideslip angle controller. (Solid line) Actual
yaw rate. (Dashed line) Trend of the reference yaw rate. Tire A has been used.

Fig. 12. Time histories of ax , ay , r, and β evaluated during the tip-in test at
V = 90 km/h for (dotted-dashed line) the baseline vehicle, (dashed line) the
vehicle with the PID controller, and (solid line) the vehicle with the suboptimal
SOSM controller. Tire A has been used.
Fig. 16. Sideslip angle β and sideslip rate β̇ evaluated at V = 90 km/h
and μ = 0.5 during the sequence of step steer tests with (dashed lines) the
suboptimal SOSM yaw rate controller and (solid lines) the integrated yaw rate-
sideslip angle controller. Tire A has been used.

the magnitude and the phase angle response for the whole range
of input frequencies.

VI. A SSESSMENT OF THE I NTEGRATED YAW R ATE

Fig. 13. Frequency response evaluated with sinusoidal steering inputs at AND S IDESLIP C ONTROLLER
different frequencies and amplitude of 10 deg for (dotted-dashed line) the
baseline vehicle, (dashed line) the vehicle with the PID controller, and (solid To test the integrated yaw rate and sideslip angle controller, a
line) the vehicle with suboptimal SOSM. Tire A has been used.
sequence of step steer maneuvers with increasing amplitude has
been simulated. If the friction estimation algorithm correctly
works (condition omitted for brevity in the graphs of this
section), the values of the yaw-rate-based P W F parameter are
the same for the SOSM yaw rate controller and the integrated
yaw rate and sideslip controller. In fact, the sideslip controller
works as an additional safety feature of the yaw rate controller.
This arrangement of the sideslip controller operating only in
Fig. 14. Frequency response evaluated with sinusoidal steering inputs at extreme conditions is particularly effective, as it does not
different frequencies and amplitude of 20 deg for (dotted-dashed line) the require a precise estimation of sideslip angle and sideslip rate
baseline vehicle, (dashed line) the vehicle with the PID controller, and (solid
line) the vehicle with suboptimal SOSM. Tire A has been used. during normal driving, which is difficult even with the best state
estimators available at the moment. Conversely, the existing
state estimator technology is capable of detecting significant
D. Frequency Response Analysis
peaks of sideslip angle and sideslip rate, typical of extreme
The yaw rate frequency response characteristics have been cornering maneuvers. As a consequence, for assessing the
evaluated with sinusoidal steering inputs of amplitudes 10 performance of the proposed integrated controller, a tire–road
and 20 deg (Figs. 13 and 14, respectively) at V = 90 km/h friction coefficient of μ = 0.5 is considered, and a fault in the
and varying frequency for the PID, suboptimal SOSM, and friction estimation is simulated so that the reference yaw rate
the baseline vehicle, through simulations with the CarMaker- is generated for μ = 1 (otherwise, the sideslip controller would
Simulink vehicle model. not be useful).
The calculations show that the frequency response of the Fig. 15 shows the beneficial effect in terms of the yaw
controlled vehicle with suboptimal SOSM varies to a large rate limitation when the sideslip angle controller is activated.
extent according to the amplitude of the steering inputs due Fig. 16 highlights that the sideslip angle is maintained within
to the high nonlinearity of the system. However, the subop- the specified threshold limits when the sideslip angle con-
timal SOSM guarantees a marginally faster response (smaller troller is active (solid line). The vehicle with the SOSM yaw
magnitude of the phase angle) than the PID controller at least rate controller only is characterized by peaks of sideslip an-
up to 2.7 Hz (see Fig. 14). On the other hand, for the PID gle in excess of 10 deg, which are beyond the limits that
algorithm, the variation in the yaw rate gain as a function of can be managed by nonprofessional drivers. Simulations at
the input frequency is significantly smoother, showing mono- μ = 0.3 have shown similar results and are omitted here for
tonically decreasing behavior. Both controllers largely improve brevity.
3622 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 63, NO. 8, OCTOBER 2014

VII. C ONCLUSION [10] H. B. Pacejka, Tyre and Vehicle Dynamics, 2nd ed. Oxford, U.K.:
Butterworth-Heinemann, 2006.
In this paper, a comparison of different yaw moment [11] L. De Novellis, A. Sorniotti, P. Gruber, V. Ivanov, and K. Hoepping,
controllers has been presented, focusing on PID-based and “Torque vectoring control for electric vehicles with individually con-
trolled motors: State-of-the-art and future developments,” presented at the
sliding-mode algorithms. These are evaluated in the time and Electric Vehicle Symposium, Los Angeles, CA, USA, 2012, Paper 26.
frequency domains. Furthermore, the effectiveness of an inte- [12] E-VECTOORC FP7 project. [Online]. Available: http:www.e-vectoorc.eu
grated sideslip controller within the overall control structure [13] K. Sawase, Y. Ushiroda, and T. Miura, “Left–right torque vectoring tech-
nology as the core of Super All Wheel Control (S-AWC),” Mitsubishi
applied to maintain stability in emergency conditions has been Motors Tech. Rev., vol. 18, pp. 18–24, Apr. 2006.
investigated. [14] L. De Novellis, A. Sorniotti, and P. Gruber, “Optimal wheel torque distri-
The results highlight the ability of the controllers to sig- bution for a four-wheel-drive fully electric vehicle,” SAE Int. J. Passenger
Cars, vol. 6, no. 1, pp. 128–136, May 2013.
nificantly change the understeer characteristic compared with [15] G. Genta, Motor Vehicle Dynamics: Modelling and Simulation.
the baseline vehicle. In the case of the PID controllers, a Singapore: World Scientific, 1997.
preliminary analysis in the frequency domain has shown that [16] W. F. Milliken and D. L. Milliken, Chassis Design—Principles and Anal-
ysis. Warrendale, PA, USA: SAE, 2002.
gain scheduling is not necessary for compensating variations [17] L. De Novellis, A. Sorniotti, and P. Gruber, “Design and compari-
in the vehicle yaw rate response at different operating points. son of the handling performance of different electric vehicle layouts,”
Furthermore, in both quasi-steady-state maneuvers and typical Proc. Inst. Mech. Eng. D, Autom. Eng., vol. 228, no. 2, pp. 218–232,
Feb. 2014.
transient tests, the PID algorithms allow good tracking perfor- [18] T. Chung and K. Yi, “Design and evaluation of side slip angle-based
mance and acceptable robustness against variations in the main vehicle stability control scheme on a virtual test track,” IEEE Trans.
vehicle parameters and operating conditions. The tracking per- Control Syst. Technol., vol. 14, no. 2, pp. 224–234, Mar. 2006.
[19] F. Bottiglione, A. Sorniotti, and L. Shead, “The effect of half-shaft torsion
formance in quasi-steady-state conditions is further enhanced dynamics on the performance of a traction control system for electric ve-
by the suboptimal sliding-mode approach, which also achieves hicles,” Proc. Inst. Mech. Eng. D, Autom. Eng., vol. 226, no. 9, pp. 1145–
the objective of minimizing the variation in the vehicle yaw 1159, Sep. 2012.
[20] W. F. Milliken and D. L. Milliken, Race Car Vehicle Dynamics.
rate in acceleration conditions during tip-in maneuvers while Warrendale, PA, USA: SAE, 1995.
cornering. However, the sliding-mode controllers can provoke [21] O. Mokhiamar and M. Abe, “Active wheel steering and yaw moment
undesirable oscillations in yaw rate during step steer maneuvers control combination to maximize stability as well as vehicle respon-
siveness during quick lane change for active vehicle handling safety,”
at high steering amplitudes. Overall, ease of implementation, Proc. Inst. Mech. Eng. D, Autom. Eng., vol. 216, no. 2, pp. 115–124,
predictable behavior, and good frequency response are key Feb. 2002.
characteristics in favor of the PID controllers for real vehicle [22] K. El Rifai, “Nonlinearly parameterized adaptive PID control for
parallel and series realization,” in Proc. Amer. Control Conf., 2009,
applications. pp. 5150–5155.
The integration of yaw rate and sideslip angle controllers [23] G. Bartolini, “Applications of a sub-optimal discontinuous control algo-
creates a nested structure in which the yaw rate controller rithm for uncertain second order systems,” Int. J. Robust Nonlin. Control,
vol. 7, no. 4, pp. 299–313, Apr. 1997.
continuously works to achieve the reference cornering behavior, [24] G. Bartolini, A. Ferrara, and E. Usai, “Chattering avoidance by second-
and the sideslip angle controller is used in case of inaccurate order sliding mode control,” IEEE Trans. Autom. Control, vol. 43, no. 2,
friction coefficient estimation or emergency conditions. Sim- pp. 241–246, Feb. 1998.
[25] G. Bartolini, A. Ferrara, A. Pisano, and E. Usai, “On the con-
ulations have demonstrated the effectiveness of the proposed vergence properties of a 2-sliding control algorithm for non-linear
control structure. uncertain systems,” Int. J. Control, vol. 74, no. 7, pp. 718–731,
Jan. 2001.
[26] A. Levant, “Sliding order and sliding accuracy in sliding mode control,”
R EFERENCES Int. J. Control, vol. 58, no. 6, pp. 1247–1263, Dec. 1993.
[1] M. J. Hancock, R. A. Williams, E. Fina, and M. C. Best, “Yaw motion [27] L. V. Levantovsky, “Second order sliding algorithms: Their realiza-
control via active differentials,” Trans. Inst. Meas. Control, vol. 29, no. 2, tion,” in Proc. Dyn. Heterogeneous Syst. Conf., Moscow, Russia, 1985,
pp. 137–158, Jun. 2007. pp. 32–43.
[2] A. van Zanten, R. Erhardt, and G. Pfaff, “VDC, The vehicle dynamics [28] C. A. Smith and A. Corripio, Principles and Practice of Automatic Pro-
control system of Bosch,” presented at the International Congress Expo- cess Control. Hoboken, NJ, USA: Wiley, 1997.
sition, Detroit, MI, USA, 1995, SAE Tech. Paper 950 749. [29] C. Vecchio, “Sliding mode control: Theoretical developments and applica-
[3] G. Kaiser, Q. Liu, C. Hoffmann, M. Korte, and H. Werner, “Torque tions to uncertain mechanical systems,” Ph.D. dissertation, Dept. Comput.
vectoring for an electric vehicle using an LPV drive controller and a torque Eng. Syst. Sci., Univ. Pavia, Pavia, Italy, 2008.
and slip limiter,” in Proc. 51st IEEE Conf. Decision Control, 2012, [30] M. Canale, L. Fagiano, A. Ferrara, and C. Vecchio, “Comparing inter-
pp. 5016–5021. nal model control and sliding-mode,” IEEE Trans. Intell. Transp. Syst.,
[4] C. Poussot-Vassal, O. Sename, L. Dugard, and M. Savaresi, “Vehicle dy- vol. 10, no. 1, pp. 31–41, Mar. 2009.
namic stability improvements through gain-scheduled steering and brak- [31] J. J. E. Slotine and W. Li, Applied Nonlinear Control. Englewood Cliffs,
ing control,” Veh. Syst. Dyn., vol. 49, no. 10, pp. 1597–1621, Oct. 2011. NJ, USA: Prentice-Hall, 1991.
[5] S. Anwar, “Yaw stability control of an automotive vehicle via generalized [32] A. Ferrara and M. Rubagotti, “A sub-optimal second order sliding mode
predictive algorithm,” in Proc. Amer. Control Conf., 2005, pp. 435–440. controller with saturating actuators,” in Proc. Amer. Control Conf., Seattle,
[6] Y. Shibahata, K. Shimada, and T. Tomari, “Improvement of vehicle ma- WA, USA, 2008, pp. 1082–1087.
neuverability by direct yaw moment control,” Veh. Syst. Dyn., vol. 22, [33] Z. L. Gaing, “A particle swarm optimization approach for optimum design
no. 5/6, pp. 465–481, Jan. 1993. of PID controller in AVR system,” IEEE Trans. Energy Convers., vol. 19,
[7] M. Abe, “A theoretical analysis on vehicle cornering behaviors in ac- no. 2, pp. 384–391, Jun. 2004.
celeration and in braking,” Veh. Syst. Dyn., vol. 15, no. S1, pp. 1–14, [34] Passenger Cars—Steady-State Circular Driving Behaviour—Open-Loop
Jan. 1986. Test Methods, ISO Std. 4138:2012, 2012.
[8] M. Canale, L. Fagiano, A. Ferrara, and C. Vecchio, “Vehicle yaw control [35] M. Bauer, C. Ackermann, and R. Isermann, “Integrated state estimation
via second-order sliding-mode technique,” IEEE Trans. Ind. Electron., with driving dynamic sensors and GPS data to evaluate driving dynamics
vol. 33, no. 11, pp. 3908–3916, Nov. 2005. control functions,” presented at the FISITA World Automotive Congress,
[9] B. Kwak and Y. Park, “Robust vehicle stability controller based on multi- Beijing, China, 2012, Vol. 8, Paper F2012-E15-013, Vol. 8.
ple sliding mode control,” presented at the SAE World Congress, Detroit, [36] E. Ono, T. Abe, and Y. Muragishi, “Vehicle stabilizing control apparatus,”
MI, USA, 2001, Paper 2001-01-1060. U.S.A, U.S. Patent 2006/0041 367 A1, Feb. 23, 2006.
DE NOVELLIS et al.: COMPARISON OF FEEDBACK CONTROL TECHNIQUES FOR TV CONTROL OF FEVS 3623

Leonardo De Novellis (M’12) received the M.Sc. Patrick Gruber received the M.Sc. degree in motor-
degree in mechanical engineering and the Ph.D. de- sport engineering and management from Cranfield
gree in mechanical and biomechanical design from University, Cranfield, U.K., in 2005 and the Ph.D.
Politecnico di Bari, Bari, Italy, in 2006 and 2010, degree in mechanical engineering from the Univer-
respectively. sity of Surrey, Surrey, U.K., in 2009.
Since 2011, he has been a Research Fellow with Since 2009, he has been a Lecturer in advanced
the University of Surrey, Surrey, U.K. His main re- vehicle systems engineering with the University of
search interests include continuously variable trans- Surrey. His current research is in the field of tire
missions and vehicle dynamics. dynamics and the development of novel tire models.

Aldo Sorniotti (M’12) received the M.Sc. degree in Andrew Pennycott received the M.Eng. and the
mechanical engineering and the Ph.D. degree in ap-
Ph.D. degrees in mechanical engineering from the
plied mechanics from Politecnico di Torino, Torino,
University of Glasgow, Glasgow, U.K., in 2004 and
Italy, in 2001 and 2005, respectively.
2008, respectively.
He is a Senior Lecturer in advanced vehicle engi-
Since 2012, he has been a Research Fellow with
neering with the University of Surrey, Surrey, U.K.
the University of Surrey, Surrey, U.K. His research
He is the E-VECTOORC (Electric Vehicle Control
is focused on control engineering for automotive and
of Individual Wheel Torque for On- and Off-Road rehabilitation applications.
Conditions) project coordinator. His main research
interests include vehicle dynamics control and trans-
missions for electric vehicles.