Control Engineering Practice 106 (2021) 104654

Contents lists available at ScienceDirect

Control Engineering Practice

journal homepage: www.elsevier.com/locate/conengprac

Integrated nonlinear model predictive control for automated driving

Nishant Chowdhri a , Laura Ferranti b , Felipe Santafé Iribarren c , Barys Shyrokau d ,∗
a
Control Systems Engineer, Chassis & Powertrain Development Department, Toyota Gazoo Racing Europe GmbH, Germany
b
Learning and Autonomous Control Section, Department of Cognitive Robotics, Delft University of Technology, Netherlands
c
R&D: Chassis Control, Toyota Motor Europe, Belgium
d
Section of Intelligent Vehicles, Department of Cognitive Robotics, Delft University of Technology, Netherlands

ARTICLE INFO ABSTRACT

Keywords: This work presents a Nonlinear Model Predictive Control (NMPC) scheme to perform evasive maneuvers and
Model predictive control avoid rear-end collisions. Rear-end collisions are among the most common road fatalities. To reduce the
Optimal control risk of collision, it is necessary for the controller to react as quickly as possible and exploit the full vehicle
Integrated control
maneuverability (i.e., combined control of longitudinal and lateral dynamics). The proposed design relies on
Nonlinear control
the simultaneous use of steering and braking actions to track the desired reference path and avoid collisions
Vehicle control
Evasive action
with the preceding vehicle. A planar vehicle model was used to describe the vehicle dynamics. In addition,
Rear-end collision the dynamics of the brake system were included in the NMPC prediction model. Furthermore, the controller
MIMO system incorporates constraints to ensure vehicle stability and account for actuator limitations. In this respect, the
Collision avoidance constraints were defined on Kamm circle and Ideal Brake Torque Distribution (IBD) logic for optimal tire force
and brake torque distribution. To evaluate the design, the performance of the proposed NMPC was compared
with two "more classical" MPC designs that rely on: (i) a linear bicycle model, and (ii) a nonlinear bicycle
model. The performance of these three controller designs was evaluated in simulation (using a high-fidelity
vehicle simulator) via relevant KPIs, such as reference tracking Root Mean Square (RMS) error, controller’s
rise/settling time, and Distance to Collision (i.e., the lateral distance by which collision was avoided safely).
Different single-lane-change maneuvers were tested and the behavior of the controllers was evaluated in the
presence of lateral wind disturbances, road friction variation, and maneuver aggressiveness.

1. Introduction environments, with a collision rate of 1 accident per 8 s (Gilreath

& Associates, 2013). These accidents are caused by the inability of
The U.S. National Highway Traffic Safety Administration (NHTSA) the human driver to perform an evasive maneuver to avoid colliding
reported that the number of fatal crashes in 2016 increased by 5.6% with the LV successfully. According to Adams (1994), Beal and Gerdes
with a toll of 37,461 deaths (National Highway Traffic Safety Adminis- (2009), Markkula, Benderius, Wolff, and Wahde (2012) and Wang, Zhu,
tration, 2017). The European Road Safety Observatory reported similar Chen, and Tremont (2016), the major reasons behind the human-driver
numbers for the EU (European Road Safety Observatory, 2018). In the failure are associated with (i) the driver preference towards braking
attempt to reduce the number of fatalities on the road, the automotive
rather than steering, (ii) the longer driver reaction time, (iii) the driver
industry started equipping the vehicles with active vehicle safety tech-
inability to control the vehicle during highly nonlinear and critical
nologies, such as Antilock Brake System (ABS), Vehicle Stability Control
maneuvers, (iv) fear and anxiety. The linear regime of motion (i.e., the
(VSC) and traction control. These technologies halved the number of
linear handling behavior) comes naturally to the driver. As soon as the
fatalities from 20,774 in year 2007 to 11,990 in year 2016 (European
Road Safety Observatory, 2018). But, the NHTSA reported that rear-end vehicle is pushed to the handling limits (for example, during an evasive
collisions are the main cause of road fatalities (accounting for more maneuver at high speed), the situation becomes challenging for the
than 30% of all the road fatalities) (Insurance Information Institute, driver (Hac & Bodie, 2002).
0000; National Highway Traffic Safety Administration, 2016). A rear- Automated technologies such as Emergency Driving Support (EDS)
end crash occurs when the difference in relative speeds between the can be extremely beneficial in this context. An EDS consists of five
subject vehicle (SV) and lead vehicle (LV) in front causes a collision. main components, that are, Risk Monitoring, Driver Monitoring, Decision
Rear-end crashes are extremely common in both urban and highway Making, Path Planning, and Control (Choi, Kim, & Yi, 2011). This study

∗ Corresponding author.
E-mail addresses: [email protected] (N. Chowdhri), [email protected] (L. Ferranti), [email protected] (F.S. Iribarren),
[email protected] (B. Shyrokau).

https://doi.org/10.1016/j.conengprac.2020.104654
Received 19 March 2020; Received in revised form 17 July 2020; Accepted 6 October 2020
Available online xxxx
0967-0661/© 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

focuses on the last component of the EDS design, that is, the design maneuvers. The controller was tested at varying vehicle velocities. Its
of a tailored control strategy for the rear-end collision scenario. The performance was also validated in the presence of external disturbances
design of such a controller is an active research area. Extensive reviews such as lateral wind and parameter uncertainty via varying the road
on various control strategies—such as PID control, Sliding Mode Con- friction coefficient. The integrated NMPC control design is not limited
trol (SMC), Linear Quadratic Regulator (LQR), Nonlinear backstepping to EDS and can be used as a controller for automated driving for high-
control, etc.—were presented in Ackermann, Bechtloff, and Isermann way and urban driving environments, provided that a path-planning
(2015), Aripina, Sam, Kumeresan, Ismail, and Kemao (2014), Choi et al. algorithm provides a suitable trajectory for the proposed controller
(2011), Mokhiamar and Abe (2004), Shah (2015), Soudbakhsh and to follow (for example, the vulnerable-road-users-aware path-planning
Eskandarian (2010) and Zhu, Shyrokau, Boulkroune, van Aalst, and algorithm proposed in Ferranti et al., 2019). Finally, using specific Key
Happee (2018). Performance Indicators (KPIs), the proposed integrated nonlinear MPC
According to Ackermann et al. (2015), Choi et al. (2011) and design was compared with the following baselines: (i) an MPC design
Mokhiamar and Abe (2004), an effective control design in the context that uses a linear bicycle model as prediction model (referred to as
of EDS should: the Linear MPC design), and (ii) an MPC design that uses a nonlinear
bicycle model as prediction model (referred to as the Nonlinear MPC
1. Involve both steering and VSC via Differential Braking (DB).
design). In all test cases, the proposed design outperforms the baseline
2. Optimally distribute the steering and brake control actions to
controllers.
improve the overall vehicle performance.
3. Handle tire nonlinearities during highly dynamic situations
1.1. Related work
(e.g., during an evasive steering maneuver).

Hence, an integrated (i.e., involving both steering and DB), optimal, and The literature survey provided limited work in the field of integrated-
nonlinear control design should be able to handle an evasive maneuver control design using MPC (Barbarisi, Palmieri, Scala, & Glielmo, 2009;
successfully. Choi & Choi, 2016; Falcone, Tseng, Borrelli, Asgari, & Hrovat, 2008;
PID and SMC control are not optimal in nature. While LQR does Jalali, Khosravani, Khajepour, Chen, & Litkouhi, 2017; Yi et al., 2016).
provide optimal control, it only works for unconstrained optimiza- The authors in Falcone, Borrelli, Asgari, Tseng, and Hrovat (2007)
tion problem which is a limitation for vehicle control as the vehicle formulated the NMPC problem for a double-lane-change maneuver
dynamics are always bounded within the designed operating range. using a bicycle model as system model with the steering wheel angle
MPC on the other hand covers all the three conclusions made under as control command. The designed NMPC worked successfully at speed
one control design and becomes the most suitable control algorithm of 7 m/s but failed to stabilize the vehicle at 10 m/s. The authors
for vehicle control. Since MPC is an optimal control technique and of Falcone et al. (2007) concluded that integrated control of steering
is based on the designed prediction model, it can accommodate vehi- and braking can improve the performance of the controller. The same
cle nonlinearities and Multi-Input–Multi-Output (MIMO) models in its research team then designed a NMPC based control with 1s as pre-
design. Therefore our goal is to design an EDS controller that relies diction horizon to optimize combination of braking and steering for
on NMPC to accommodate all control objectives above. By relying on obstacle avoidance via double-lane-change maneuver (Falcone et al.,
an augmented nonlinear planar vehicle model as prediction model, the 2008). They used a 10 DoF planar vehicle model as prediction model
proposed NMPC design allows to simultaneously control the lateral and (first six DoF being the vehicle’s longitudinal and lateral velocity, head-
longitudinal vehicle dynamics via steering and braking, while taking ing angle and yaw rate, and the vehicle’s global position coordinates
actuator dynamics into account. The proposed solution has brake actu- in both longitudinal and lateral direction. The four wheel’s dynamics
ator dynamics modeled inside its prediction which allows direct control considered individually are the remaining four DoF) and used a Pacejka
of the wheels and does not require any additional control allocation model to model the tire characteristics. The control action was the front
scheme (which would be nontrivial to implement). In addition, NMPC steering angle and each wheel’s brake torque values. The controller
allows to directly account for tire saturation limitations and actuator successfully passed the test at 14 m/s. The controller, however cannot
limits in the constraint formulation. Most of the literature in the area be applied for real-time applications because it took around 15 min to
of MPC for evasive maneuvers focuses on lateral control at constant complete a 12 s simulation. In addition, high amount of oscillations in
longitudinal speed and relies on simplified vehicle models (e.g., the the steering angle were observed due to improper tuning because of
bicycle model) (Beal & Gerdes, 2009, 2013; Choi, Kang, & Lee, 2012; increased number of model parameters.
Keviczky, Falcone, Asgari, & Hrovat, 2006). The main reasons for this The authors in Jalali et al. (2017) designed an integrated Linear
choice is that more complex vehicle and tire models are challenging to MPC (LMPC) control using Active Front Steering (AFS) and DB for
implement in real-time framework. Using a dynamic bicycle model as lateral stability of the vehicle. The authors use bicycle model as predic-
prediction model, however, limits the controller to exploit DB. This is tion model with a prediction horizon of 0.3 s. The controller provides
because one requires control of the left and right wheels for DB to work. assistance control of ±10 deg on the road wheel angle, satisfying the
But in bicycle model, both the front and rear tires are lumped together side-slip angle 𝛽 constraint to ensure vehicle stability at all times. The
respectively as one tire each as a result of which the effect of DB is not controller, however, is not subjected to robustness tests, such as wind
well captured in the dynamics of bicycle model. From the vehicle dy- disturbance or parameter uncertainty.
namics perspective, DB plays a fundamental role to ensure safety during Similarly, the authors in Choi and Choi (2016) designed MPC con-
evasive maneuvers. Compared to the aforementioned controllers, the troller via an extended bicycle model that utilized AFS and DB for
proposed design exploit the benefits of DB by controlling each of the vehicle stability. In their work, the prediction model encapsulated
wheels directly and allows to control longitudinal and lateral dynamics. the lagged characteristics of actuator dynamics and tire forces, both
This is achieved by modeling the brake actuator dynamics inside the modeled as a first-order lag system. By calculating the control action
prediction model with the planar vehicle model to have an overall as steering wheel angle and yaw moment correction 𝑀𝑧 , another
optimal control strategy and removing the need of conventional control optimization problem was solved to get the optimal tire forces, thereby
allocation schemes, making it a unique MPC-based controller design increasing the overall computational time and loss of performance. By
that allows direct control of the vehicle’s wheels. solving two optimization problems, the idea of having one integrated
The designed planar vehicle-based integrated NMPC control was controller for vehicle control was lost.
validated in several different scenarios, ranging from highly dynamic The authors in Barbarisi et al. (2009) designed a Vehicle Dynamics
single-lane-change evasive maneuvers (to replicate scenarios in which Control using linear time-varying MPC with sampling time and predic-
rear-end collisions occur if not properly handled) to normal lane change tion horizon as 0.25 s and 5 steps, respectively. They assumed constant

2
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

longitudinal dynamics with control on lateral vehicle motion only.

While the controller was able to pass the standard ISO 19365:2016
Sine with Dwell test, the controller showed oscillatory behavior when
working close to the constraint boundaries.
Lastly, the authors in Yi et al. (2016) designed two MPC controllers
for collision avoidance control using steering and braking combined
as control action. They used a nonlinear bicycle model as prediction
model for the NMPC controller design and for the linear MPC design,
they linearized the bicycle model around the operating point along with
approximation of the nonlinear constraints into linear form. But instead Fig. 1. Single lane change maneuver.
of calculating brake torques, their MPC control action is the com-
manded longitudinal acceleration which is then used in another logic to
calculate required brake torques. They tested their design on a single- an overview of all the scenarios covered to analyze the performance
lane-change maneuver at 70 km/h with target lateral displacement 2 better. Section 7 presents the results for each maneuver performed. It
m. They set the sampling time and prediction horizon at 0.06 s and 25 also details the KPIs values for every maneuver and provides an com-
steps, respectively. While the collision was avoided, it was seen with parative performance analysis between the three controllers designed
both the control strategies that an overshoot of about 35% was achieved in this research. Finally, Section 8 concludes the paper and provides
in lateral position tracking, leading to poor tracking performance. Also, recommendations for future work.
the NMPC designed was not real-time feasible with mean computation
time between 4 to 8 s reported.
2. Preliminaries
Compared to the previous approaches, the proposed integrated
NMPC control design relies on a single controller to compute the
2.1. Single lane change maneuver
control action for the four wheels, while taking into account the ve-
hicle limitations. The integrated NMPC design solves the optimization
To evaluate the designed controller, a representation of a rear-end
problem online and in real-time. In addition, the actuator dynamics
collision avoidance maneuver is required. To the best of our knowledge,
was modeled in the prediction model to account for their reaction
however, no standard maneuvers for evasive action are available in
time and have more accurate predictions. Furthermore, a kinematic
the literature (when referred to SAE, NHTSA, ISO and Euro NCAP,
reference path1 (i.e., the reference is generated based on a kinematic
respectively). Nevertheless, the NHTSA report (Lee, Llaneras, Klauer, &
description of the vehicle, without taking into account dynamics) as
Sudweeks, 2007) provides certain key insights concerning real-life sce-
reference trajectory was used in all the simulations to further assess the
narios in which rear-end collisions occur. By collecting the crash data
controller’s robustness towards tracking imprecise (from the dynamic
of 100 different cars, the NHTSA highlighted that collisions frequently
point of view) reference values. Lastly, this approach in its modeling
happen on a straight road with no junctions while driving at constant
involves the use of dynamic constraints aimed at maximizing the
speed. Hence, based on the NHTSA conclusions, the single-lane-change
vehicle stability by minimizing the vehicle body slip angle and body
maneuver was used shown in Fig. 1 and
slip angle rate. Hence, the control design accounts for g–g diagram2 and
parameter values provided by Ford Motors Research were selected
Kamm circle3 constraints by design to ensure that the vehicle and the
to represent an evasive scenario (Zegelaar, 2017). The values are 𝑑ref
four tires operate in the stable working regime of motion. In addition,
= 2.5 m and 𝐿ref = 30 m, that is, the maneuver begins when the
compared to the state of the art, the integrated NMPC control design
subject vehicle is exactly 30 m away from lead vehicle and has to
considers IBD constraints for ideal distribution of brake torques.
laterally traverse 2.5 m to avoid the rear-end collision. To ensure that
The robustness of the proposed integrated NMPC design was exten-
the designed maneuver is indeed aggressive and nonlinear in nature,
sively tested to various disturbances and uncertainties such as lateral
the vehicle speeds were selected to perform the maneuver according to
wind and road friction variation. The maneuver’s nature during these
the following metrics:
tests was kept evasive at all times to test controller’s robustness under
aggressive and nonlinear vehicle dynamics conditions. For all the tested 𝐿ref
TTC = , (1a)
scenarios, the integrated NMPC design was successful to avoid rear-end 𝑣𝑥
collisions at all times thereby guarantying vehicle stability. All these 𝑣𝑥
TTB = , (1b)
simulations were tested on a high-fidelity vehicle simulator provided 2𝑎𝑥max
√
by Toyota and validated by field tests. 2𝑑ref
TTS = , (1c)
𝑎𝑦max
1.2. Paper structure
where Time to Collision (TTC), Time to Brake (TTB) and Time to Steer
The paper is structured as follows. Section 2 introduces preliminary (TTS). Here, 𝑣𝑥 refers to the subject vehicle’s longitudinal speed, 𝑎𝑥max
concepts used in the paper, such as the designed maneuver, the ref- is the subject vehicle’s maximum longitudinal acceleration (taken as
erence trajectory, and model predictive control. Section 3 details the 0.8 μg) and 𝑎𝑦max is the subject vehicle’s maximum lateral acceleration
integrated NMPC formulation. Section 4 details the two benchmark (taken as 0.6 μg), with 𝜇 being the friction coefficient. The acceleration
controllers used for performance comparison. Section 5 presents the values mentioned above were chosen based on the baseline values
high-fidelity vehicle simulator. Section 6 explains the KPIs to quan- provided by NHTSA’s definition of a Near-Crash which states that
titatively assess the controllers performance. This section also gives circumstances involving vehicle braking greater than 0.5 g or steering
input leading to lateral acceleration greater than 0.4 g to avoid a crash
1
constitutes a rapid maneuver scenario (Lee et al., 2007). Therefore it
Usually, model-based path planners rely only on kinematic descriptions
was ensured that the designed maneuver is evasive at all times.
of the vehicle and do not take into account its dynamics.
2
g–g diagram characterizes the vehicle’s lateral and longitudinal accelera-
Eq. (1) was evaluated at different vehicle speeds and ensured that
tion performance envelope under a given road surface condition. the inequality TTS ≤ TTC ≤ TTB is satisfied at all times to design
3
Also referred as friction circle or friction ellipse, Kamm circle defines the realistic test scenarios. Table 1 reports the speed range for each value of
maximum values of the resultant of longitudinal and lateral force that can be 𝜇. The small values of TTC highlights that the maneuver is aggressive
obtained under a particular operating condition. and captures the real-life evasive situations.

3
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

Table 1 optimal control problem (OCP) is discretized. ACADO relies on Sequen-

Reference maneuver speed range and TTC values.
tial Quadratic Programming (SQP). The SQP algorithm linearizes the
𝜇 𝑣𝑥min TTCmin 𝑣𝑥max TTCmin
discretized nonlinear control problem to convert it into a Quadratic
– [km/h] [s] [km/h] [s]
Programming (QP) problem (Vukov, Domahidi, Ferreau, Morari, &
1 78 1.38 115 0.93
Diehl, 2013). Using condensing techniques, the state variables are elim-
0.9 75 1.44 110 0.98
0.8 70 1.54 102 1.06 inated (i.e., the overall number of optimization variables is reduced).
0.7 65 1.66 97 1.13 Multiple shooting method was preferred over single shooting method
0.6 60 1.80 90 1.20 as it is more robust for nonlinear systems, such as the vehicle dynamics
0.5 55 1.96 82 1.31
in the non-linear handling area. Once the optimization problem has
0.4 50 2.16 72 1.50
0.3 43 2.51 64 1.68 been linearized, an Active Set Method was used to solve the resulting
0.2 35 3.08 50 2.16 quadratic programming problem. Levenberg–Marquardt Gauss New-
0.1 25 4.32 35 3.08 ton based hessian approximation method was used, and, for active
set method, open-source C++ software qpOASES (Ferreau, Kirches,
Potschka, Bock, & Diehl, 2014) was selected. In addition, ACADO
The single-lane-change maneuver should be translated into a path implements the Real Time Iteration (RTI), which involves performing
the vehicle can follow in terms of lateral position, heading angle, and a single SQP iteration per sampling time, for a more efficient solution
yaw rate. Hence, the lateral reference position is approximated by using calculation (Vukov et al., 2013).
a Sigmoid curve (Choi et al., 2011) defined as a function of vehicle’s In the current study, to mitigate the effects of the unpredictable
longitudinal position 𝑥 as follows: computation times, the maximum amount of iterations of the solver
was fixed. If the optimizer reaches the maximum number of iterations
𝐵
𝑦ref = . (2) without converging to an optimal solution, then previously computed
1 + 𝑒−𝑎(𝑥−𝑐)
feasible prediction (shifted one step forward) is used. This previously
In addition, the reference heading angle 𝜓ref and reference yaw rate
computed prediction is used to warm-start the optimizer. Since the
𝜓̇ ref are defined as follows:
vehicle dynamics (plant model) are continuous in nature, sudden jumps
( )
𝜕𝑦ref in its dynamics are less likely. This gives a higher chance that the next
𝜓ref = tan−1 (3a)
𝜕𝑥 optimal solution is close to the previous one. Both modifications are
𝜓̇ ref = 𝜅1 𝑣𝑥 , (3b) common practice in practical MPC implementations.

For the readability of the paper, Appendix A details the quantities

3. NMPC controller design
associated with the definition of the reference signals.

2.2. Model predictive control This section describes the proposed integrated NMPC formulation.
First, the various dynamic couplings in the vehicle model are discussed
Model Predictive Control (MPC) solves a constrained optimization that should be well captured in the prediction model to improve the
problem online to compute the optimal sequence of control commands performance of the controller. Then, the planar vehicle model used as
over a finite time window, called prediction horizon. The problem prediction model in the proposed NMPC design is discussed. Finally,
is formulated based on (i) the available plant measurements, (ii) the the control objectives and the constraints are described.
plant-prediction model, (iii) control objectives, and (iv) plant/actuator
limitations. Only the first control command of this sequence is applied 3.1. Vehicle dynamics coupling
to the plant in closed loop in the receding-horizon fashion. The predic-
tion model captures the plant dynamics and gives controller the ability Designing an integrated control is a non-trivial problem due to the
to predict the behavior of plant. The prediction model, as this work also strong couplings in the vehicle dynamics as explained below. Based on
shows, is fundamental for the performance of the controller. the knowledge of vehicle dynamics and its associated coupling effect,
A general formulation of MPC controller is given by it is therefore essential for developing the prediction model of the MPC
𝑁𝑝 −1
∑ ( ) ( ) controller. According to Attia, Orjuela, and Basset (2012) and Lim and
min 𝐽𝑘 𝑋𝑘 , 𝑈𝑘 , 𝑋𝑘ref + 𝐽𝑘 𝑋𝑁𝑝 , 𝑋𝑁
ref
(4a) Hedrick (1999), the following longitudinal and lateral couplings arise
𝑈 𝑝
𝑘=0 in case of vehicle motion:
( )
s.t.: 𝑋𝑘+1 = 𝑓 𝑋𝑘 , 𝑈𝑘 , 𝑘 = 0, … , 𝑁𝑝 − 1 (4b)
( ) Kinematic and dynamic coupling. This coupling arises due to the effect
𝐺 𝑋𝑘 , 𝑈𝑘 ≤ 𝑔𝑏 , 𝑘 = 0, … , 𝑁𝑝 − 1 (4c) of wheel steering on longitudinal dynamics of the vehicle by chang-
( )
𝐺 𝑋𝑁𝑝 ≤ 𝑔𝑝 (4d) ing the tire lateral forces. The tire longitudinal forces on the other
hand affects both the lateral dynamics and yaw motion of vehicle
𝑋0 = 𝑋init , (4e) and subsequently the rate change of lateral position is a function
of longitudinal velocity. Thus both dynamics are always coupled as
where 𝐽𝑘 is the cost function to be minimized for optimal control action
reflected in (B.1)–(B.6) (detailed in Appendix B.1).
𝑈𝑘 . 𝑋𝑘 and 𝑋𝑘ref are the states and the reference values at prediction
instant 𝑘 (𝑘 = 0, … , 𝑁𝑝 ), respectively. Function 𝑓 is the prediction Tire–road coupling. This coupling arises due to the application of lateral
model that captures the plant’s dynamics. 𝑋init is the current state and longitudinal forces by the tire. This coupling is reflected in (17).
measurement from the plant and updated online at every sampling The equations used for the lateral forces are reported in Appendix B.2.
instant. Finally, function 𝐺 comprises of all the constraints defined To model the longitudinal forces, the single corner model was used
on the states and control action with 𝑔𝑏 being the bound value. The as shown in Fig. 3. The equations used for the longitudinal forces are
constraints can be either convex or nonconvex. reported in Appendix B.3.
There are several toolboxes that can be used to solve Problem (4). In
this work the ACADO Toolkit (Quirynen, Vukov, Zanon, & Diehl, 2014) Load transfer phenomenon. This coupling arises because of the load
has been used. ACADO tackles nonlinear optimal control problems and transfer during longitudinal and lateral accelerations. This coupling is
multi-objective optimal control problems efficiently. In ACADO, the reflected in (22a)–(23e) (detailed in Section 3.3).

4
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

Fig. 3. Single corner model.

Fig. 2. Planar vehicle model.

To make the future predictions more accurate, instead of keeping

3.2. Prediction model
the lateral and longitudinal forces (𝐹𝑦ij and 𝐹𝑥ij ) constant throughout
the prediction horizon, they were formulated in terms of the prediction
The designed prediction model for integrated NMPC accounts for
model’s states. The equations used for the lateral forces are reported in
these coupling effects and captures the behavior of vehicle in both
Appendix B.2. To model the longitudinal forces, the single corner model
linear and nonlinear regime of motion. A planar vehicle model was
was used as shown in Fig. 3. The equations used for the longitudinal
selected as prediction model of the integrated NMPC design, because forces are reported in Appendix B.3.
it allows to preserve the effect of the couplings.
Following modeling simplifications were made, which are mostly Remark 1. The designed prediction model (refer for details to (B.8)–
made when using a planar vehicle model: (B.9) in Appendix B) captures the brake actuator’s dynamics. Hence, the
• Roll, pitch, and vertical motion are not modeled. controller can understand how and at what rate will the brake pressure
build up in each wheel’s caliper, allowing accordingly to calculate the
• Suspension and compliance effects are not considered.
control action, that is, the four brake torque rates 𝑑𝑇𝑏 . In addition,
• Ackermann geometry is not considered (i.e., the front left and ij
front right wheel turn by same amount). by including the actuator dynamics, the proposed strategy does not
• Since the actuator delay is small for both the front and rear require a separate controller design for control allocation, making it an
calipers, it has been ignored in the prediction model formulation. integrated approach and theoretically easy to implement as a hardware
on actual vehicle. Instead of only calculating the upper-level control
Fig. 2 provides an overview of the main quantities of the prediction action, the designed integrated NMPC controller can now directly
model. The planar vehicle is described by the following equation: calculate the control action that needs to be applied to the wheels. In
this way, one controller can take care of all the two classical control
̇ = 𝑓 (𝑥(𝑡), 𝑢(𝑡)),
𝑥(𝑡) (5) layers, that are, vehicle level and control allocation. This integrated
where 𝑥(𝑡) ∈ R15 ,
𝑢(𝑡) ∈ and 𝑓 ∶ R5 , × → R15 𝑅5 R15 .
To improve the controller design is only possible with MPC due to its modular design
readability of the paper, Appendix B.1 details 𝑓 (𝑥, 𝑢). The state vector concept.
is given by 𝑥 = [𝑣𝑥 , 𝑣𝑦 , 𝑟, 𝜓, 𝑥𝑝 , 𝑦𝑝 , 𝛿, 𝑇𝑏fl , 𝑇𝑏fr , 𝑇𝑏rl , 𝑇𝑏rr ,
act act act act Remark 2. The brake torques 𝑇𝑏ij represent the actual brake torques
𝑇𝑏fl , 𝑇𝑏fr , 𝑇𝑏rl , 𝑇𝑏rr ] where 𝑣𝑥 , 𝑣𝑦 , 𝑟, 𝜓, 𝑥𝑝 , and 𝑦𝑝 are the lon- act
cal cal cal cal applied to the wheels (i.e., the values after the brake actuator dynam-
gitudinal velocity, lateral velocity, yaw rate, heading angle, global x ics). Brake torques 𝑇𝑏ij represent the values calculated before the
and y positions of the vehicle, respectively. The brake torque 𝑇𝑏ij are cal
act effect of actuator dynamics. By providing the actuator dynamics, the
the actual brake torque value for tire 𝑖𝑗 applied to wheel after the brake controller can compensate for possible performance losses to compute
actuator dynamics. The last four states 𝑇𝑏ij are the brake torque values the final brake torque value 𝑇𝑏ij .
cal act
calculated before the brake actuator dynamics. The control vector is
𝑢 = [𝑑𝛿 𝑑𝑇𝑏 𝑑𝑇𝑏 𝑑𝑇𝑏 𝑑𝑇𝑏 ] where 𝑑𝛿 , 𝑑𝑇𝑏 , 𝑑𝑇𝑏 , 𝑑𝑇𝑏 , and 𝑑𝑇𝑏 are the 3.3. Constraints
fl fr rl rr fl fr rl rr
rate of change of road wheel angle and brake torque rate for each
wheel, respectively. The control action 𝑑𝛿 is applied to the vehicle’s States and actuator limitations should be taken into account by
steering system as steering wheel velocity (SWV) by multiplying the design for a realistic control and to eliminate actuator failure. In
control action with steering ratio 𝑠st . addition, the controller should ensure that, if a feasible solution of the
The considered maneuver is aggressive and nonlinear in nature, MPC problem exists, this will keep the vehicle in the stability envelope,
that is, the vehicle dynamics need to be constrained to a region in which
therefore the model needs to capture the tire nonlinearities. Hence,
control of the vehicle can be ensure at all times. The MPC framework
instead of giving a fixed value to the cornering stiffness 𝐶𝛼 , a Dugoff tire
allows to incorporate these stability-envelope and actuator limits in the
model was used to capture the tire nonlinear behavior. Appendix B.4
form of constraints in the design phase of the controller. Hence, to
details the model formulation. The fitted nonlinear cornering stiffness
account for physical limitation and to define a handling envelope, the
𝐶𝛼non is given by:
following constraints (19 in total) were defined that should be satisfied
𝐶𝛼ij along the length of the prediction horizon of the NMPC controller:
𝐶𝛼non = 𝑓 (𝜆), (6)
ij 1 − 𝜅ij 0 ≤ 𝑣𝑥 ≤ 170 [km/h] (7)
where 𝜅 is the longitudinal slip and 𝑓 (𝜆) is a function of the longitudinal
slip (its definition is reported in (B.17), Appendix B.4). This allows the − 5 ≤ 𝛽 ≤ 5 [deg] (8)
prediction model to preserve the overall tire behavior as well as the
tire dynamics for an accurate and superior control. − 25 ≤ 𝛽̇ ≤ 25 [deg/s] (9)

5
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

front
−17 17 𝐹𝑧rr = 𝐹𝑧,𝑔 + 𝐹𝑧𝑥 + 𝐹𝑧𝑦 , (22d)
≤𝛿≤ [deg] (10) 𝑟
𝑠st 𝑠st
−800 800 where
≤ 𝛿̇ ≤ [deg/s] (11) 𝑚𝑔𝑙𝑓
𝑠st 𝑠st front
𝐹𝑧,𝑔 = , (23a)
2 2 2
2𝐿
(𝑣̇ 𝑥 − 𝑣𝑦 𝑟) + (𝑣̇ 𝑦 + 𝑣𝑥 𝑟) ≤ (𝜇𝑔) (12) rear 𝑚𝑔𝑙𝑟
𝐹𝑧,𝑔 = , (23b)
2𝐿
0 ≤ 𝑇𝑏ij ≤ 4900 [N m] , ij = (fl, fr) (13) 𝑚(𝑣̇ 𝑥 − 𝑣𝑦 𝑟)ℎcg
act 𝐹 𝑧𝑥 = , (23c)
2𝐿
( )
0 ≤ 𝑇𝑏ij ≤ 1610 [N m] , ij = (rl, rr) (14) 𝑚(𝑣̇ 𝑦 + 𝑣𝑥 𝑟) 𝑙𝑟 ℎrf 𝐾𝜙,𝑓 ℎ
act 𝐹𝑧𝑦 = + , (23d)
𝑓 𝑡𝑓 𝐿 𝐾𝜙,𝑓 + 𝐾𝜙,𝑟 − 𝑚𝑔ℎ
( )
− 7000 ≤ 𝑇̇ 𝑏ij ≤ 7000 [N m/s] , ij = (fl, fr) (15) 𝑚(𝑣̇ 𝑦 + 𝑣𝑥 𝑟) 𝑙𝑓 ℎrr 𝐾𝜙,𝑟 ℎ
act 𝐹𝑧𝑦 = + , (23e)
𝑟 𝑡𝑟 𝐿 𝐾𝜙,𝑓 + 𝐾𝜙,𝑟 − 𝑚𝑔ℎ
− 5550 ≤ 𝑇̇ 𝑏ij ≤ 5550 [N m/s] , ij = (rl, rr) (16) and ℎ = ℎcg − (𝑙𝑟 ℎrf + 𝑙𝑓 ℎrr )𝐿−1 . Finally, Constraint (18) defines the
act
( )2 ( )2 ( )2 front to rear brake torque distribution ratio based on the parabolic
𝐹𝑥ij + 𝐹𝑦ij ≤ 𝜇ij 𝐹𝑧ij , ij = (fl, fr, rl, rr) (17) curve (right-hand side of Constraint (18) according to Breuer & Bill,
2008) for an ideal brake torque distribution. In straight-line driving,
𝑙𝑓 ℎcg (𝑣̇ 𝑥 −𝑣𝑦 𝑟)
𝑇𝑏rl + 𝑇𝑏rr 𝐿
+ 𝑔𝐿
when a vehicle brakes, it pitches forward, increasing the normal load
act act
≤ (18) of the front tires. Therefore the ability of the front tires to generate
𝑇𝑏fl + 𝑇𝑏fl +𝜖 𝑙𝑓 ℎcg (𝑣̇ 𝑥 −𝑣𝑦 𝑟)
act act 1− 𝐿
− 𝑔𝐿 brake force increases as compared to rear ones. Hence due to vehicle
Constraint (7) limits the vehicle’s speed. To ensure vehicle stability, design, usually in a straight-line driving, the front tires brake more
constraints (8)–(9) limit both the vehicle sideslip angle 𝛽 and the than the rear tires. Since this was not modeled in the prediction model,
sideslip angle gradient 𝛽.̇ Based on the concept of stable 𝛽 - 𝛽̇ reference constraint (18), which is only activated during straight-line driving,
region by He, Crolla, Levesley, and Manning (2006) and the evaluations captures the IBD behavior well and is defined with 𝜖 in denominator
on the same phase plane by European Council Service Framework Pro- equal to 0.001 to ensure mathematical infeasibility is avoided.
gramme (0000) and Shyrokau, Wang, Savitski, Hoepping, and Ivanov
(2015), it was concluded that a bound of 5 deg for 𝛽 and a bound of 3.4. Cost function
25 deg/s for 𝛽̇ were reasonable to define a stable region for vehicle
motion. The constraints defined ensure that the vehicle remains within The cost function incorporates the control objectives of the NMPC
this stable region at all times and does not spin away. The vehicle design. It is designed to keep the tracking error between process output
sideslip angle 𝛽 can be in terms of vehicle states so that the constraint and given reference as small as possible and at the same time, minimize
is dynamic in nature and is always satisfied along the entire prediction the control action along the prediction horizon. Based on Barbarisi et al.
horizon using Eq. (19). (2009), Falcone et al. (2008) and Jalali et al. (2017) a 2-square norm
( ) error minimization function was chosen to model the cost function. The
𝑣𝑦
𝛽 = tan−1 (19) cost function is defined as follows:
𝑣𝑥 𝑁𝑝 −1
∑ [
Since, the bounds in constraints (8)–(9) are small angles, the ap- 𝐽𝑘 = ‖𝑋(𝑘 + 𝑖) − 𝑋ref (𝑘 + 𝑖)‖2𝑄 + (24a)
proximation tan 𝛽 ≈ 𝛽 holds true which gives the final equation for 𝑖=1
]
approximating vehicle slip quantities as shown in Eqs. (20)–(21). ‖𝑈 (𝑘 + 𝑖 − 1)‖2𝑃 + (24b)
𝑣𝑦
𝛽= (20) ‖𝑋(𝑘 + 𝑁𝑝 ) − 𝑋ref (𝑘 + 𝑁𝑝 )‖2𝑆 , (24c)
𝑣𝑥
𝑣̇ 𝑦 where 𝑋 is the state prediction, 𝑋ref = [0, 0, 𝑟ref , 𝜓ref , 0, 𝑦ref , 0, 0, 0, 0,
𝛽̇ = (21) 0, 0, 0, 0, 0]T is the reference prediction, and 𝑈 is the control predic-
𝑣𝑥
tion. The reference trajectories 𝑦ref , 𝜓ref and 𝑟ref are the reference values
Constraints (10)–(11) limits the steering wheel angle and steering for position, heading angle and yaw rate, respectively. In addition,
wheel rate, respectively (note that by using the steering ratio 𝑠st they the cost penalizes 𝛿 to control the magnitude of the Steering Wheel
have been written in the form of road wheel angle to directly bound the Angle (SWA) at higher speeds (high SWA and SWV may lead to vehicle
state). Constraint (12) represents the g–g diagram constraint represent- spinning out). Furthermore, the cost penalizes the brake torques to
ing the working limit of the vehicle. Since the controller is integrated in ensure that minimum control action energy is utilized to perform the
nature and can control both lateral and longitudinal dynamics, there- maneuver. Lastly, the cost penalizes all the five control actions to
fore accordingly the working envelope is defined. Constraints (13)–(16) ensure that the entire maneuver can be performed at minimum control
define the brake actuator limits in terms of maximum brake torque values. This ensures that the control energy cost is minimized, reducing
and rates. Constraint (17) represents the four Kamm circle constraints the actuator wear and improving its service life as well.
(one for each tire). These constraints prevent/minimize the effect of
tire saturation. By making assumptions that the sprung and unsprung 3.5. Controller tuning
masses are lumped as total mass 𝑚, the roll angle 𝜙 is small and the
dynamic terms of roll and pitch motion are ignored, that is, only the The proposed design involves the selection of several tuning param-
contribution from the static terms are taken in modeling, the normal eters. This section details and motivates the design choices.
load (which is the right hand side of the bound) on each tire 𝐹𝑧ij , The sampling time 𝑡𝑠 of the controller to 0.035 s has been chosen
respectively, was defined in equations below: motivated by the cycle update time of all the other ECU’s of the
rear passenger car. This ensures that at each sample, the controller has
𝐹𝑧fl = 𝐹𝑧,𝑔 − 𝐹𝑧𝑥 − 𝐹𝑧𝑦 , (22a)
𝑓 adequate information of all the reference signals and vehicle’s states to
rear
𝐹𝑧fr = 𝐹𝑧,𝑔 − 𝐹𝑧𝑥 + 𝐹𝑧𝑦 , (22b) solve the optimization problem. Furthermore, the prediction horizon
𝑓 𝑁𝑝 of the proposed controller is set to 30 steps (i.e., 1.05 s) to max-
front
𝐹𝑧rl = 𝐹𝑧,𝑔 + 𝐹𝑧𝑥 − 𝐹𝑧𝑦 , (22c) imize its performance (according to the assessment criteria described
𝑟

6
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

4.1. Linear MPC

This controller involves the use of a linear bicycle model as pre-

diction model. This representation requires the additional following
assumptions:

• Small angle approximation, that is, sin 𝜃 ≈ 𝜃, cos 𝜃 ≈ 1.

• Constant longitudinal velocity, that is, 𝑎𝑥 = 0.
• Linear tire model is used to capture the tire dynamics:

𝐹𝑦ij = 𝐶𝛼ij 𝛼ij (25)

The dynamics are detailed in Appendix C. In addition, compared to the

Fig. 4. Bicycle model.
proposed design, this design has only six constraints to ensure vehicle
stability and bounded control action within feasible actuator range.
Specifically, five constraints are similar to constraints (7)–(11), define
in Section 6.2) while ensuring computational feasibility (i.e., real-time for the planar vehicle model. The last constraint bounds the lateral
performance). acceleration of the vehicle as follows:
The remaining tuning parameters are the weight matrices 𝑄 ⪰ 0, ( )
− 0.85𝜇𝑔 ≤ 𝑣̇ 𝑦 + 𝑣𝑥 𝑟 ≤ 0.85𝜇𝑔 (26)
𝑃 ≻ 0, and 𝑆 ⪰ 0 in (24). Tuning these matrices is nontrivial, given the
multiple objectives the controller needs to optimize. Tuning is based The cost function associated with the linear bicycle model is similar
on a trade-off to achieve small tracking error balancing the use of the (with a reduced set of state and control objectives) to the one presented
control actions. Compared to other control techniques (such as rule- in Section 3.4. Finally, 𝑁𝑝 was set to 50 steps to predict 1.75 s in future
based controllers), tuning of these weighting matrices is based on the to improve the control performance. The horizon is longer with respect
dynamics and kinematics of the vehicle. For example, giving a high to the one selected using the planar vehicle model because the opti-
penalty to the control action means that the actuators should stay as mizer has to solve a smaller problem with less constraints and decision
close as possible to zero, which is unrealistic in the scenarios considered variables. While, with a shorter horizon the two baseline controllers
in this work. provided substandard tracking performance, with a longer horizon the
The initial tuning was refined using heuristics and simulations. One integrated design failed to meet the real-time requirements. Hence,
parameter at the time was varied and the KPIs were evaluated. In to have a fair comparison among the controllers, different horizon
addition, given that the maneuver is performed for various speeds and lengths were selected for the integrated and baseline controllers. While
for various values of 𝜇, the tuning was performed at different speeds a shorter horizon (due to real-time constraints) for the integrated design
and for each value of 𝜇. By doing so, following patterns were identified could be seen as a limitation, its performance is still very good thanks
which were used to schedule the weights in the designed cost function: to its ability to use a more accurate model to generate predictions.

• Increasing the terminal position tracking tuning weight 𝑆𝑦𝑁 lead 4.2. Nonlinear MPC
to corner cutting.
• Increasing value of yaw rate tuning weight 𝑄𝜓̇ improved the This controller relies on a nonlinear bicycle model to capture the
overall reference tracking performance. vehicle’s nonlinearities while performing the maneuver. In addition,
• Reducing tuning weight of wheel angle 𝑄𝛿 improved tracking compared to the previous design and similar to the proposed design,
performance. this controller relies on the Dugoff tire model to capture the tire
• For a fixed 𝜇 value with increasing maneuver speeds, the tracking nonlinearities. The states and control commands are the same of the
was improved by increasing the weights of road wheel angle and linear bicycle model, but they are non-linearly coupled. Appendix D
wheel velocity (𝑄𝛿 and 𝑅𝛿̇ ), and by reducing the weight of lateral details on the nonlinear bicycle model. The constraints, cost function,
position 𝑄𝑦 . and horizon length are those detailed for the linear bicycle model.
• Decreasing the tuning parameter of control action brake torque
rate 𝑅𝑇̇ 𝑏 , 𝑅𝑇̇ 𝑏 , 𝑅𝑇̇ 𝑏 , 𝑅𝑇̇ 𝑏 and keeping other tuning parameters 5. Vehicle simulator
fl fr rl rr
constant lead to increase in overshoot.
The three controllers were tested and compared on an IPG
CarMaker-based simulation platform using a high-fidelity Toyota ve-
4. Benchmark controller design
hicle model. The model has been parametrized based on mass-inertia
parameters obtained from vehicle inertia measuring facility, suspension
This section describes the benchmark controllers designed to com- kinematics and compliance obtained by measurement on a Kinematics
pare it against the proposed integrated NMPC control design. Compared & Compliance test rig for wheel suspension characterization, and fi-
to the proposed controller, these designs rely on a simplified prediction nally, validated by field tests on the proving ground. A high-fidelity
model, that is a bicycle model represented in Fig. 4. This representation 3-DoF steering model with column-based electric power steering logic
is based on the same assumptions made in the planar vehicle based was used as the steering actuator model. This steering system was
MPC control (Section 3.2) with addition of one more assumption that validated with full-vehicle testing and it is implemented in the Toyota’s
no longitudinal or lateral load transfer is considered. high-end driving simulator (Damian, Shyrokau, Ocariz, & Akutain,
Compared to the planar vehicle model, the bicycle model has seven 2019). To simulate tire dynamics, the Delft-Tyre 6.2 was used in
states, that are, 𝑣𝑥 , 𝑣𝑦 , 𝜓, 𝑟, 𝑥𝑝 , 𝑦𝑝 and 𝛿 and only a control action, that combination with a detailed tire property file identified from bench
is, the rate of change of the wheel steering angle 𝑑𝛿 .4 Both benchmark testing (pure and combined slip, transient dynamics).
controllers have the same cost function and constraints (detailed in Sec- The brakes considered in this research are floating point disk brakes
tion 4.1. In the following, additional details about the two benchmark with conventional HAB system. The nonlinear HAB brake dynamic
controllers are provided. model derived from real-life vehicle data is the following (according
to Zhou, Lu, & Peng, 2010):
4
Given that in a bicycle model the two axle tires are clubbed as one, 𝑃act 𝑒−𝑇𝑑 𝑠
differential braking and individual wheel control are not an option. = , 𝑃̇ ≤ 𝛤 (27)
𝑃cal 𝑇𝑙 𝑠 + 1 act

7
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

Table 2 Table 5
Varying velocity scenarios. Varying 𝐶2 scenarios.
𝜇 [–] 𝑣𝑥 [km/h] 𝜇 [–] 𝑣𝑥 [km/h] 𝐶2 [m]
0.9 (dry road) 75 80 85 90 95 100 0.9 90 5 4 3 2 1 0.5

Table 3
Varying 𝜇 scenarios.
𝜇 [–] 𝑣𝑥 [km/h]
0.5 0.6 0.7 0.8 0.9 1.0 80

Table 4
Varying wind speed scenarios.
𝜇 [–] 𝑣𝑥 [km/h] 𝑣𝑤 [km/h]
0.9 90 0 10 30 50 70
Fig. 5. DTC graphical representation.

The parameters for the front axle are 𝑇𝑑 = 0.06 s, 𝑇𝑙 = 0.12 s and 6.2. Assessment criteria
𝛤 = 230 bar/s. For the rear axle, the parameters are 𝑇𝑑 = 0.02 s, 𝑇𝑙
= 0.05 s and 𝛤 = 550 bar/s. The maximum pressure 𝑃max that the To assess the performance of the three controllers for all the scenar-
brakes can achieve is taken as 160 bar. To convert the brake pressure to ios defined in Section 6.1, following KPIs were defined.
brake torque, the following relationship was used according to Limpert The first KPIs selected were Overshoot (𝑀𝑝 ), Settling Time (𝑇𝑠 ), and
(1999): Rise Time (𝑇𝑟 ). These are typically used to assess the performance of a
controller to step reference signals (which is a close approximation of
𝑇𝑏ij = 2𝑃actij 𝐴wcij 𝜂𝑐ij 𝜇𝐿ij 𝑟ij , ij = (fl, fr, rl, rr) (28) the reference trajectory described in Section 2.1 and graphically shown
act

The brake hysteresis effect was neglected as it is assumed that brake in Fig. 6a). To further assess the tracking performance of the controller,
the RMS of the tracking errors over the horizon length of the controllers
hysteresis has a minor influence on the brake performance for a new
was considered, that is:
hydraulic disk brake mechanism (Shyrokau, Wang, Augsburg, & Ivanov, √
√
2013). √1 ∑ 𝑁
( )2
𝑋RMS = √ 𝑋(𝑖) − 𝑋ref (𝑖) , (29)
𝑁 𝑖=1
6. Maneuver scenarios and assessment criteria
where 𝑋 ∈ {𝑦, 𝜓, 𝜓} ̇ and 𝑋ref ∈ {𝑦ref , 𝜓ref , 𝜓̇ ref }, according to the
definition of the reference signals in Section 2.1. Finally, the last KPI
6.1. Maneuver scenarios
considered was Distance to Collision (DTC) (depicted in Fig. 5). This KPI
represents the lateral distance between the left-rear corner of LV and
The designed single lane change maneuver was performed under a right-front corner of SV. The DTC is a safety-based KPI and gives an
variety of conditions to check the controller capabilities and robustness idea of the safety margin the controller can produce.
in different scenarios. This paper presents the results for the most For a good control performance, the DTC should be as high as pos-
relevant scenarios. sible and all other KPIs should be as small as possible. This will ensure
collision avoidance, well tracked trajectories, and quick stabilization of
6.1.1. Set 1 – varying velocity 𝑣𝑥 the vehicle post lane change.
These scenarios involve variations in vehicle speeds at a constant
road friction coefficient. Table 2 summarizes these scenarios. Note 7. Simulation results
that the speed range was selected based on the results of Section 2.1
(Table 1). The controllers were tested in the scenarios described in Section 6.1.
This section presents one specific case in more detail (with comparison
with the benchmark controllers), that is the scenario 𝑣𝑥 = 90 km/h
6.1.2. Set 2 – varying friction coefficient 𝜇
and 𝜇 = 0.9. In addition, this section shows the KPI results for all the
These scenarios involve variations in values of 𝜇 for a given speed.
scenarios using the proposed design. Furthermore, the section shows
Table 3 summarizes these scenarios.
how the proposed controller handles constraints by design and is
real-time feasible.
6.1.3. Set 3 – varying lateral wind velocity 𝑣𝑤
These scenarios involve variation of external lateral wind speeds 𝑣𝑤 Comparison with the benchmark controllers. Fig. 6 compares the three
control strategies with respect to the reference signals. The dashed-blue
for a fixed value of 𝜇 and 𝑣𝑥 . Table 4 summarizes these scenarios. Note
lines represent the reference signals, the red, yellow, and purple lines
the wind is modeled as constant perturbation to flow only in direction
represent the proposed controller, the linear MPC design, and nonlin-
South, directly opposing the vehicle as it turns left (towards direction
ear MPC design, respectively. The first plot of Fig. 6 shows that the
North) according to the defined maneuver.
proposed integrated NMPC control approach significantly reduces the
overshoot compared to the linear MPC design (33% overshoot). In addi-
6.1.4. Set 4 – varying maneuver’s aggressiveness tion, the proposed design provides better tracking performance (𝑦RMS =
These scenarios highlight the ability of the controller to handle 7.39% and DTC = 0.41 m) compared to nonlinear MPC design (𝑦RMS =
various dynamic maneuvers ranging from evasive actions to normal 11.25% and DTC = 0.26 m). The improved tracking performance is due
single-lane changes. In these scenarios, the parameter 𝐶2 in the sigmoid to the more detailed prediction model and integrated control action of
curve decreases gradually. By doing so, the slope of the trajectory steering and braking. In this respect, Fig. 7 highlights how the proposed
was gradually reduced, making the reference trajectory less aggressive. approach provides the necessary steering action. Recall that the single-
Table 5 summarizes these scenarios. lane change maneuver starts with a left turn, followed by a right turn,

8
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

Fig. 6. Lateral displacement, yaw angle and yaw rate comparison for the scenario 𝜇 = 0.9 and 𝑣𝑥 = 90 km/h. (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)

and concludes with a straight-line drive. During the first turn, as the Table 6
figure shows, the left brakes brake while the right brakes are kept KPIs for Set 3 using the proposed approach — varying lateral wind velocity 𝑣𝑤
scenario.
at zero. This gives the required additional yaw moment for better
𝑣𝑤 𝑀𝑝 𝑇𝑠 𝑇𝑟 𝑦RMS 𝜓RMS 𝜓̇ RMS DTC 𝐷of f
tracking. During the second turn, the controller provides the control
[km∕h] [%] [s] [s] [%] [%] [%] [m] [m]
action to steer the steering wheel clockwise (i.e., negative SWA value).
Simultaneously, the controller also reduces the left brakes and increases 0 1.34 3.10 0.53 5.71 55.24 387.90 0.41 0.00
10 0.29 7.34 0.54 6.12 54.70 386.41 0.39 0.01
the right brakes to get desired yaw moment for tracking the reference 30 0.08 7.72 0.58 7.75 53.94 384.16 0.35 0.04
values. Finally, in the last phase of straight-line driving, the SWA goes 50 0.18 7.83 0.64 11.22 53.45 394.12 0.26 0.08
to zero. At the same time, the right and left brake values are modulated 70 0.11 8.06 1.95 15.21 56.21 411.01 0.18 0.12
to ensure the vehicle remains stable and aligned straight. Once done,
the brake torques also go to zero to conclude the maneuver. The most
important KPI for collision avoidance is the DTC value. This is because Table 7
KPI for Set 4 using the proposed approach — varying maneuver’s aggressiveness
the top priority in case of evasive action is collision avoidance which
scenario.
is directly represented by DTC. A positive and non-zero value ensures
𝐶2 𝑀𝑝 𝑇𝑠 𝑇𝑟 𝑦RMS 𝜓RMS 𝜓̇ RMS 𝑎𝑦max
that collision was avoided successfully. The higher the DTC values are, [–] [%] [s] [s] [%] [%] [%] [m/s2 ]
the higher the safety margins are. Figs. 8a and 9a presents the DTC
5.0 1.34 3.10 0.53 7.37 71.30 500.64 5.83
results for scenario sets 1 and 2 (Section 6.1). As the figures show, the 4.0 2.25 3.11 0.66 6.46 72.92 389.33 4.78
designed integrated NMPC control outperforms the benchmark control 3.0 2.78 3.47 0.77 6.38 56.82 252.89 3.18
strategies providing the highest DTC values. Also, both the benchmark 2.0 2.12 4.04 0.89 4.69 33.58 121.79 2.22
controllers fail to avoid the collision at 100 km/h (DTC value is zero) 1.0 0.00 5.01 1.18 2.44 13.48 36.04 1.23
whereas integrated NMPC controller avoids the collision successfully. 0.5 0.00 5.75 1.49 1.37 6.52 12.20 0.77
It is to be noted that DTC values are meaningful when looked along
with trajectory tracking overshoot values. A higher overshoot value
may result in high DTC value. In principle this reflects that collision
IBD constraint satisfaction. To show the IBD constraint (18) is active,
was safely avoided but it does not highlight that trajectory tracking
the lane change scenario with pre-braking maneuver was considered.
was poor. Therefore, for both sets 1 and 2, the percentage overshoot
𝑀𝑝 figures have also been plotted in Figs. 8b and 9b. It can be seen that Two seconds before the subject vehicle is 30 m away from lead vehicle,
linear MPC gives slightly higher DTC than nonlinear MPC. But linear the subject vehicle will brake and decelerate. After this, the single lane
MPC also gives a very high overshoot value as compared to nonlinear change maneuver is performed. This was done as the IBD constraint is
MPC. Therefore, with a marginal difference in DTC value and negligible only activated during straight-line pre-braking maneuver. Fig. 10 shows
overshoot observed, the performance of nonlinear MPC is overall better the brake torques. Due to this brake distribution constraint, front brake
than linear MPC. And integrated NMPC not only gives highest DTC torques are more than the rear.
value but also gives close to zero overshoot value, proving that it indeed
Constraint satisfaction. The controller is real-time feasible during the
performs the best of all.
maneuver. Fig. 11 shows the calculation times for the number of
Lateral wind. Table 6 summarizes the KPI values for scenario set 3 calls for each considered scenario (5–6 simulations per each scenario
(i.e., lateral wind offset scenario) using the proposed controller. For according to the Tables 2–5). In particular, it should be noted that the
wind speeds up to 70 km/h, the controller is able to avoid the collision computation time increases when the reference signal changes. This
successfully. The table reports an additional parameter, namely 𝐷off , behavior is typical of the optimizer used to solve the nonlinear control
that measures the offset distance between reference trajectory and the problem. The computation time tends to increase with the number of
vehicle trajectory at the end of the maneuver. With such high wind active constraints. However, the computation time is still within the
speeds, the proposed controller returns a maximum offset value of 0.12 real-time constraint highlighted by the dashed-red line.
m, judged smaller than existing references (≈ 0.5 m). This highlights The maximum number of iterations of the solver is empirically
the effectiveness and robustness of the proposed control scheme. fixed to 5𝑁𝑝 (nX + nU + nC) with nX, nU and nC representing total
Maneuver’s aggressiveness. Table 7 summarizes the KPI values for sce- number of states of the prediction model, total number of control action
nario set 4 (i.e., varying maneuver’s aggressiveness) using the proposed and total number of constraints respectively. The value is based on
controller. The table reports the maximum lateral acceleration gen- the diagnostic flags of the solver during initial tests. The solver was
erated by the maneuver, that is 𝑎𝑦max , highlighting that for varying warm-started with the prediction computed at the previous time step
dynamic scenarios, the controller is working efficiently. These results (shifted by one in time). This helps mitigate the effect of unpredictable
show that the proposed design, thanks to its integrated ability to computation times. In addition as a backup, if the solver fails to find a
simultaneously steer and brake, provides both lateral and longitudinal feasible solution within the fixed number of iterations, the last feasible
control and avoids the collision in all scenarios successfully. solution is applied to the vehicle in closed loop.

9
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

Fig. 7. Scenario 𝜇 = 0.9 and 𝑣𝑥 = 90 km/h for proposed control strategy: steering wheel angle (SWA) and brake torques 𝑇𝑏 .

Fig. 8. DTC and percentage overshoot 𝑀𝑝 for Set 1 – varying velocity 𝑣𝑥 scenario.

Fig. 9. DTC and percentage overshoot 𝑀𝑝 for Set 2 – varying friction coefficient 𝜇 scenario.

The proposed design is able to deal with constraints even when the values, respectively, for 𝑣𝑥 = 90 km/h and 𝜇 = 0.9. It can be seen
maneuver is performed in the nonlinear regime of motion, as Figs. 12 that the values are within the defined stable envelope and at the
and 13 show. The figures depict the g–g diagram and Kamm circle same time, while preserving tracking performance. It can be seen in

10
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

Fig. 10. Brake torques for straight-line braking case: 𝑣𝑥 = 90 km/h and 𝜇 = 0.9. Fig. 12. g–g diagram for case: 𝑣𝑥 = 90 km/h and 𝜇 = 0.9.

Fig. 11. MPC computation time for different scenarios.

Fig. 13 that the Rear Right (RR) tire plot shifts towards the left side
i.e. towards negative longitudinal force direction as the maneuver is
being performed. This is because the test vehicle used in the simulation
Fig. 13. Kamm circle of each tire for case: 𝑣𝑥 = 90 km/h and 𝜇 = 0.9.
is a Front Wheel Drive (FWD) car. Even though all the four wheels are
braking (as seen in Fig. 7), the drive torque from the engine is being
transferred to the front wheels as a result of which the overall tire
longitudinal forces in the front tires are mostly in the positive region to take into account that a vehicle’s motion is always coupled in both
of Kamm circle. But during the maneuver, the rear tires brakes as well the lateral and longitudinal direction. Conventional hierarchical control
and the overall tire longitudinal force becomes negative as a result strategies implemented on vehicles, often consider the design of lateral
of which the rear tire’s longitudinal force are mostly in the negative and longitudinal control separately, making the simplified assumptions.
side of Kamm circle. Since the RR tire brakes the most among the two In contrast to the classical approaches, we proposed an integrated
rear wheels, the longitudinal force shift towards the negative half is control strategy based on nonlinear model predictive control (NMPC)
more as compared to other wheels. Nevertheless, it can be seen that taking into account the coupling between lateral and longitudinal
the lateral force ratio is very high and close to the limits, suggesting control by design. It was shown how this strategy is able to effectively
that the controller is able to control the nonlinearities of the maneuver control the vehicle when the maneuver is in the nonlinear range of
effectively at all times. motion and in the presence of lateral wind. The controller did not show
any oscillatory behavior and overshoots while tracking the desired
8. Conclusion and future work reference trajectory for the various conditions tested.
We compared the proposed design with two benchmark controllers
The goal of this work is to design an integrated nonlinear MPC con- based on model predictive control. Compared to the proposed MPC con-
troller to provide effective vehicle control in both linear and nonlinear troller design, the other two baseline MPC control approaches rely on
regime of motion and to reduce the high number of accidents caused in simplifying assumptions on the prediction model (linear and nonlinear
a rear-end collision scenario. In this scenario, it is important to show the bicycle model) and definition of the constraints. Our integrated NMPC
controller’s ability to guarantee vehicle stability and passenger safety design outperformed the other two control strategies in all considered
for various conditions. Also in this scenario, it is of foremost importance scenarios. Furthermore, the designed strategy showed robustness to

11
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

external disturbances and parameter uncertainty, while being real-time 𝑡𝑠 MPC controller sampling time, [s]
feasible. 𝑉𝑥ij Wheel longitudinal velocity, [m/s]
The recommendations for future work are (i) to include the tire 𝑣𝑥 Chassis longitudinal velocity, [m/s]
slip dynamics inside the prediction model to control the phenomenon 𝑣𝑦 Chassis lateral velocity, [m/s]
of wheel locking and the ABS activation while braking; (ii) better
𝑥𝑝 Vehicle global position in longitudinal direction, [m]
trajectory generation methods can be used to make the reference values
more realistic and practical to follow; (iii) incorporation of tire model 𝑦𝑝 Vehicle global position in lateral direction, [m]
inside the controller’s prediction model can make the MPC more robust 𝑦ref Reference lateral position, [m]
to parameter variation and improve the dynamic capabilities of the
controller to capture vehicle dynamics. Declaration of competing interest

The authors declare that they have no known competing finan-

9. Nomenclature according to ISO 8855:2011
cial interests or personal relationships that could have appeared to
influence the work reported in this paper.
𝛼ij Tire slip angle, [rad]
𝛽 Vehicle sideslip angle, [rad] Acknowledgment
𝛽̇ Sideslip angle gradient, [rad/s]
𝜓 Chassis yaw/heading angle, [rad] L. Ferranti is supported by the Dutch Science Foundation NWO-TTW
𝜓ref Reference yaw angle, [rad] within the SafeVRU project (nr. 14667).
𝜓̇ ref Reference yaw rate, [rad/s]
𝛿sw Steering-wheel angle, [rad] Appendix A. Reference signal definition
𝛿ij Road wheel steer angle, [rad]
𝜅ij Longitudinal slip, [−] The main quantities associated with the definition of the reference
𝜔̇ ij Wheel angular acceleration, [rad/s2 ] signals presented in Section 2.1 are defined below.
𝑎𝑥 Longitudinal acceleration, [m/s2 ] 𝜕𝑦ref 𝑎𝐵𝑒−𝑎(𝑥−𝑐)
= , (A.1a)
𝑎𝑦 Lateral acceleration, [m/s2 ] 𝜕𝑥 (1 + 𝑒−𝑎(𝑥−𝑐) )2
( 2 )
𝜕 𝑦ref
𝐶𝛼ij Tire cornering stiffness, [N/rad]
𝜕𝑥2
𝜅1 = (A.1b)
𝐶𝛼𝑓 Front axle cornering stiffness, [N/rad] ( ( )2 ) 23
𝜕𝑦ref
𝐶𝛼𝑟 Rear axle cornering stiffness, [N/rad] 1+ 𝜕𝑥
𝐶𝜅ij Longitudinal slip stiffness, [N/[–]] √
−𝑘2 + 𝑘22 − 4𝑘1 𝑘3
𝑑𝛿 Road wheel steer rate, [rad/s] 𝑎= , (A.1c)
𝑑𝑇𝑏 Brake torque rate per wheel, [N m/s] 2𝑘1
ij 𝐶1
𝑑ref Target lateral displacement by SV, [m] 𝑐= (A.1d)
𝑎
𝐹𝑥ij Tire longitudinal force, [N]
and
𝐹𝑥𝑓 Front axle longitudinal force, [N]
(𝐵𝑥1 )2 (𝐵𝐶2 )2
𝐹𝑥𝑟 Rear axle longitudinal force, [N] 𝑘1 = − (A.2a)
16 16
𝐹𝑦ij Tire lateral force, [N] 𝐵 2 𝑥1 𝐶1 𝐵𝑦1 𝑥1 𝐵 2 𝑥1
𝑘2 = − − + (A.2b)
𝐹𝑧ij Tire normal force, [N] 8 2 4
(𝐵𝐶1 ) 2 2 𝐵𝑦1 𝐶1 𝐵 2 𝐶1
𝑔 Acceleration due to gravity, [m/s2 ] 2 𝐵
𝑘3 = + 𝑦1 + + − 𝐵𝑦1 − − 𝐶22 (A.2c)
ℎrf Front roll center height, [m] 16
( ) 4 2 4
ℎrr Rear roll center height, [m] 𝐵
𝐶1 = log −1 (A.2d)
𝑦tol
𝐼𝑧𝑧 Vehicle inertia around z-axis, [kg m2 ]
𝐽wij Wheel moment of inertia, [kg m2 ] 𝐵 refers to lateral displacement to be achieved by the subject vehicle,
𝑎 is the slope of the Sigmoid curve, (𝑥1 , 𝑦1 ) are the coordinates of the
𝐾𝜙,𝑓 Front roll stiffness, [N m/rad]
obstacle vehicle’s rear-left corner, 𝑦tol is the initial lateral displacement
𝐾𝜙,𝑟 Rear roll stiffness, [N m/rad]
of the subject vehicle at the beginning of the maneuver, 𝐶2 is the pre-
𝑙𝑓 Distance from front axle to CoG, [m] defined minimum length which is a tuning parameter and 𝜅1 is the
𝑙𝑟 Distance from rear axle to CoG, [m] trajectory curvature.
𝐿 Wheelbase, [m]
𝐿ref Distance to LV, [m] Appendix B. Prediction model equations
𝑚 Total vehicle mass, [kg]
𝑁𝑐 Control horizon, [−]
𝑁𝑝 Prediction horizon, [−] B.1. Planar vehicle model
𝜓,
̇ 𝑟 Yaw rate, [rad/s]
𝑟eff Effective rolling radius, [m] The 15 equations representing the planar vehicle NMPC model is
𝑠st Steering ratio, [−] shown in Eqs. (B.1)–(B.11).
𝑇𝑒ij Traction torque, [N m] (𝐹𝑥fl + 𝐹𝑥fr ) cos (𝛿) − (𝐹𝑦fl + 𝐹𝑦fr ) sin (𝛿) + (𝐹𝑥rl + 𝐹𝑥rr )
𝑣̇ 𝑥 =
𝑇𝑏ij Applied brake torque to wheel, [N m] 𝑚 (B.1)
act
+𝑣𝑦 𝑟
𝑇𝑏ij Calculated brake torque before actuator dynamics, [N m]
cal (𝐹𝑥fl + 𝐹𝑥fr ) sin (𝛿) + (𝐹𝑦fl + 𝐹𝑦fr ) cos (𝛿) + (𝐹𝑦rl + 𝐹𝑦rr )
𝑡𝑓 Front track, [m] 𝑣̇ 𝑦 =
𝑚 (B.2)
𝑡𝑟 Rear track, [m] −𝑣𝑥 𝑟

12
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

[ ( )
𝑟̇ = (𝐹𝑥fl + 𝐹𝑥fr ) sin (𝛿)𝑙𝑓 + (𝐹𝑦fl + 𝐹𝑦fr ) cos (𝛿)𝑙𝑓 𝜇𝐹𝑧𝑡 1 − 𝜅ij
ij
−(𝐹𝑦rl + 𝐹𝑦rr )𝑙𝑟 + 0.5𝑡𝑓 (𝐹𝑥fr − 𝐹𝑥fl ) cos (𝛿) + 0.5𝑡𝑓 (𝐹𝑦fl (B.3) 𝜆= √ (B.16)
] ( )2 ( ( ))2
−𝐹𝑦fr ) sin (𝛿) + 0.5𝑡𝑟 (𝐹𝑥rr − 𝐹𝑥rl ) ∕𝐼𝑧𝑧 2 𝐶𝜅ij 𝜅ij + 𝐶𝛼ij tan 𝛼ij
{
𝜓̇ = 𝑟 (B.4) 𝜆(2 − 𝜆), 𝜆 < 1
𝑓 (𝜆) = (B.17)
1, 𝜆 ≥ 1
𝑥̇ 𝑝 = 𝑣𝑥 cos (𝜓) − 𝑣𝑦 sin (𝜓) (B.5) 𝐶𝜅ij 𝜅ij
𝐹𝑥ij = 𝑓 (𝜆) (B.18)
1 − 𝜅ij
𝑦̇ 𝑝 = 𝑣𝑥 sin (𝜓) + 𝑣𝑦 cos (𝜓) (B.6) ( )
𝐶𝛼ij tan 𝛼ij
𝐹𝑦ij = 𝑓 (𝜆) (B.19)
1 − 𝜅ij
𝛿̇ = 𝑑𝛿 (B.7)
𝜇𝑜 and 𝑒𝑟 are tuning parameters. 𝑉𝑥ij is the wheel’s longitudinal velocity
𝑇𝑏ij − 𝑇𝑏ij
cal act and was calculated using the equations below:
𝑇̇ 𝑏ij = , ij = (fl, fr) (B.8)
act 0.12 ( ) ( )
𝑉𝑥fl = 𝑣𝑦 + 𝑙𝑓 𝑟 sin (𝛿) + 𝑣𝑥 − 0.5𝑡𝑓 𝑟 cos (𝛿) (B.20)
𝑇𝑏ij − 𝑇𝑏ij
cal act
𝑇̇ 𝑏ij = , ij = (rl, rr) (B.9) ( ) ( )
act 0.05 𝑉𝑥fr = 𝑣𝑦 + 𝑙𝑓 𝑟 sin (𝛿) + 𝑣𝑥 + 0.5𝑡𝑓 𝑟 cos (𝛿) (B.21)
𝑇̇ 𝑏ij = 𝑑𝑇𝑏 , ij = (fl, fr) (B.10)
cal ij
𝑉𝑥rl = 𝑣𝑥 − 0.5𝑡𝑟 𝑟 (B.22)
𝑇̇ 𝑏ij = 𝑑𝑇𝑏 , ij = (rl, rr) (B.11)
cal ij
𝑉𝑥rr = 𝑣𝑥 + 0.5𝑡𝑟 𝑟 (B.23)
B.2. Lateral forces
The tire slip angle 𝛼ij is calculated as follows:
(( ) ( ) )
The equations used for the calculation of lateral forces in the above- 𝑣𝑦 + 𝑙𝑓 𝑟 cos 𝛿𝑓 𝑙 − 𝑣𝑥 − 0.5𝑡𝑓 𝑟 sin 𝛿𝑓 𝑙
mentioned prediction model are mention in Eqs. (B.12a)–(B.12d). 𝛼𝑓 𝑙 = −tan−1 ( ) ( ) (B.24a)
𝑣𝑦 + 𝑙𝑓 𝑟 sin 𝛿𝑓 𝑙 + 𝑣𝑥 − 0.5𝑡𝑓 𝑟 cos 𝛿𝑓 𝑙
( ) (( ) ( ) )
𝐶𝛼non 𝛿 − (𝑣𝑦 + 𝑙𝑓 𝑟) 𝑣𝑦 + 𝑙𝑓 𝑟 cos 𝛿𝑓 𝑟 − 𝑣𝑥 + 0.5𝑡𝑓 𝑟 sin 𝛿𝑓 𝑟
𝐹𝑦fl = fl (B.12a) 𝛼𝑓 𝑟 = −tan−1
( ) ( ) (B.24b)
𝑣𝑥 − 0.5𝑡𝑓 𝑟 𝑣𝑦 + 𝑙𝑓 𝑟 sin 𝛿𝑓 𝑟 + 𝑣𝑥 + 0.5𝑡𝑓 𝑟 cos 𝛿𝑓 𝑟
( ) ( )
𝐶𝛼non 𝛿 − (𝑣𝑦 + 𝑙𝑓 𝑟) 𝑣𝑦 − 𝑙 𝑟 𝑟
𝐹𝑦fr = fr (B.12b) 𝛼𝑟𝑙 = −tan−1 (B.24c)
𝑣𝑥 + 0.5𝑡𝑓 𝑟 𝑣𝑥 − 0.5𝑡𝑟 𝑟
( ) ( )
𝑣𝑦 − 𝑙 𝑟 𝑟
𝐶𝛼non −(𝑣𝑦 − 𝑙𝑟 𝑟) 𝛼𝑟𝑟 = −tan−1 (B.24d)
𝐹𝑦rl = rl (B.12c) 𝑣𝑥 + 0.5𝑡𝑟 𝑟
𝑣𝑥 − 0.5𝑡𝑟 𝑟
( ) The wheel slip 𝜅ij for each wheel is given by:
𝐶𝛼non −(𝑣𝑦 − 𝑙𝑟 𝑟)
𝐹𝑦rr = rr (B.12d) | |
𝑣𝑥 + 0.5𝑡𝑟 𝑟 |𝜔ij 𝑟eff − 𝑉𝑥ij |
𝜅ij = | | . (B.25)
max{𝜔ij 𝑟eff , 𝑉𝑥ij }
B.3. Longitudinal forces
The Dugoff model based tire longitudinal and lateral force calcu-
The mathematical equation that describes the wheel dynamics in lation is used in the nonlinear bicycle model based MPC design. By
y-axis in the single corner model described below (note that the rolling capturing the tire nonlinearity during the maneuver, the overshoot
resistance moment has been neglected here): (as seen in the case of linear bicycle model based MPC) was reduced
substantially, thereby improving the overall performance.
𝐽wij 𝜔̇ ij = 𝑇𝑒ij − 𝑇𝑏ij ij = (fl, fr, rl, rr)
− 𝐹𝑥ij 𝑟effij (B.13)
,
act
( )
Appendix C. Linear bicycle model details
Neglecting wheel dynamics, the term 𝐽wij 𝜔̇ ij from the LHS of Eq.
(B.13) was dropped. After rearranging the equation in terms of lon-
gitudinal force 𝐹𝑥 , the final equation used to approximate the tire
longitudinal force is given by: 𝑣̇ 𝑥 = 𝑣𝑦 𝑟 (C.1a)
𝑇𝑒ij − 𝑇𝑏ij 𝐶𝛼 𝑓 + 𝐶𝛼 𝑟 𝑙 𝑟 𝐶𝛼 𝑟 − 𝑙 𝑓 𝐶𝛼 𝑓
act
𝐹𝑥ij = , ij = (fl, fr, rl, rr) (B.14) 𝑣̇ 𝑦 = − 𝑣𝑦 + 𝑟 − 𝑣𝑥 𝑟
𝑟effij 𝑚𝑣𝑥 𝑚𝑣𝑥
𝐶𝛼 𝑓
Substituting (B.14) and (B.12a)–(B.12d) in the respective tire force + 𝛿 (C.1b)
terms of (B.1)–(B.3) gives the final formulation for the 15 state pre- 𝑚
diction model. 𝑙 𝑟 𝐶𝛼 𝑟 − 𝑙 𝑓 𝐶𝛼 𝑓 𝑙𝑟2 𝐶𝛼𝑟 + 𝑙𝑓2 𝐶𝛼 𝑓
𝑟̇ = 𝑣𝑦 − 𝑟
𝐼𝑧𝑧 𝑣𝑥 𝐼𝑧𝑧 𝑣𝑥
B.4. Dugoff tire model 𝑙 𝑓 𝐶𝛼 𝑓
+ 𝛿 (C.1c)
𝐼𝑧𝑧
The equations to model the Dugoff tire model are shown in Eqs.
𝜓̇ = 𝑟 (C.1d)
(B.15)–(B.25). An example of 2D force-slip characteristics and Kamm
circle using Dugoff tire model for normal load 𝐹𝑧 = 5000 N has been 𝑥̇ 𝑝 = 𝑣𝑥 cos (𝜓) − 𝑣𝑦 sin (𝜓) (C.1e)
plotted in Fig. B.14.
𝑦̇ 𝑝 = 𝑣𝑥 sin (𝜓) + 𝑣𝑦 cos (𝜓) (C.1f)
( √ )
𝜇 = 𝜇 𝑜 1 − 𝑒𝑟 𝑉𝑥ij 𝜅ij 2 + tan2 𝛼ij (B.15) 𝛿̇ = 𝑑𝛿 (C.1g)

13
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

Attia, R., Orjuela, R., & Basset, M. (2012). Longitudinal control for automated vehicle
guidance. The International Federation of Automatic Control.
Barbarisi, O., Palmieri, G., Scala, S., & Glielmo, L. (2009). LTV-MPC for yaw rate control
and side slip control with dynamically constrained differential braking. European
Journal of Control, 3(4), 468–479.
Beal, C. E., & Gerdes, J. C. (2009). Enhancing vehicle stability through model predictive
control. In ASME 2009 dynamic systems and control conference, Vol. 1 (pp. 197–204).
http://dx.doi.org/10.1115/DSCC2009-2659.
Beal, C. E., & Gerdes, J. C. (2013). Model predictive control for vehicle stabilization
at the limits of handling. IEEE Transactions on Control Systems Technology, 21(4),
1258–1269.
Breuer, B., & Bill, K. H. (2008). Brake technology handbook. SAE International.
Choi, M., & Choi, S. B. (2016). MPC for vehicle lateral stability via differential
braking and active front steering considering practical aspects. Journal of Automobile
Engineering, 230(4), 459–469. http://dx.doi.org/10.1177/0954407015586895.
Choi, C., Kang, Y., & Lee, S. (2012). Emergency collision avoidance maneuver based
on nonlinear model predictive control. In: IEEE international conference on vehicular
electronics and safety, Turkey (pp. 393–398).
Choi, J., Kim, K., & Yi, K. (2011). Emergency driving support algorithm with steering
torque overlay and differential braking. In 2011 14th international IEEE conference
on intelligent transportation systems (pp. 1433–1439). Washington, DC, USA: doi:
978-1-4577-2197-7/11/.
Damian, M., Shyrokau, B., Ocariz, A., & Akutain, X. C. (2019). Torque control for more
realistic hand-wheel haptics in a driving simulator. In Driving simulation conference
europe.
European Council Service Framework Programme, 0000. Electric-VEhicle control of
individual wheel torque for on- and off-road conditions (E-VECTOORC), european
funded project, FP7 (2007-2013).
European Road Safety Observatory (2018). Annual accident report 2018: Tech. rep.,
European Commission.
Falcone, P., Borrelli, F., Asgari, J., Tseng, H. E., & Hrovat, D. (2007). Predictive
active steering control for autonomous vehicle systems. IEEE Transactions on Control
Systems Technology, 15(3), 566–580.
Falcone, P., Tseng, H. E., Borrelli, F., Asgari, J., & Hrovat, D. (2008). MPC-based
Fig. B.14. Tire’s 2D force-slip characteristics and Kamm circle using Dugoff tire model. yaw and lateral stabilisation via active front steering and braking. Vehicle System
Dynamics, 46(S1), 611–628. http://dx.doi.org/10.1080/00423110802018297.
Ferranti, L., Brito, B., Pool, E., Zheng, Y., Ensing, R., Happee, R., et al. (2019). SafeVRU:
A research platform for the interaction of self-driving vehicles with vulnerable road
Appendix D. Nonlinear bicycle model details users. In: Proceedings of the intelligent vehicles symposium.
Ferreau, H., Kirches, C., Potschka, A., Bock, H., & Diehl, M. (2014). Qpoases: A para-
metric active-set algorithm for quadratic programming. Mathematical Programming
( 𝑣𝑦 +𝑙𝑓 𝑟
) Computation, 6(4), 327–363.
𝐹𝑥𝑓 cos (𝛿) − 𝐶𝛼non 𝛿 − sin (𝛿) + 𝐹𝑥𝑟 Gilreath & Associates (2013). Rear-end collisions: The most common type of auto
𝑓 𝑣𝑥 accident in the U.S.. https://tinyurl.com/rear-end-collisions.
𝑣̇ 𝑥 = + 𝑣𝑦 𝑟 (D.1a)
( 𝑚 𝑣𝑦 +𝑙𝑓 𝑟
) Hac, A., & Bodie, M. O. (2002). Improvements in vehicle handling through integrated
control of chassis systems. International Journal of Vehicle Autonomous Systems, 1(1),
𝐹 𝑥𝑓 sin (𝛿) + 𝐶𝛼non 𝛿 − 𝑣𝑥
cos (𝛿)
𝑓 83–110.
𝑣̇ 𝑦 =
He, J., Crolla, D. A., Levesley, M. C., & Manning, W. J. (2006). Coordination of
(𝑣 ) 𝑚 active steering, driveline, and braking for integrated vehicle dynamics control.
𝑦 −𝑙𝑟 𝑟
𝐶𝛼non 𝑣 Journal of Automobile Engineering, 220, 1401–1421. http://dx.doi.org/10.1243/
𝑟 𝑥
− − 𝑣𝑥 𝑟 (D.1b) 09544070JAUTO265.
( 𝑚 ( ) )
𝑣𝑦 +𝑙𝑓 𝑟 Insurance Information Institute, (0000) Facts + statistics: Highway safety, https://www.
𝐹𝑥𝑓 sin (𝛿) + 𝐶𝛼non 𝛿 − 𝑣𝑥
cos (𝛿) 𝑙𝑓 iii.org/fact-statistic/facts-statistics-highway-safety.
𝑓
𝑟̇ = Jalali, M., Khosravani, S., Khajepour, A., Chen, S., & Litkouhi, B. (2017). Model
𝐼𝑧𝑧
(𝑣 ) predictive control of vehicle stability using coordinated active steering and dif-
𝑦 −𝑙𝑟 𝑟
𝐶𝛼non 𝑣𝑥
𝑙𝑟 ferential brakes. IFAC Mechatronics: The Science of Intelligent Machines, 48, 30–41.
𝑟
+ (D.1c) http://dx.doi.org/10.1016/j.mechatronics.2017.10.003.
𝐼𝑧𝑧 Keviczky, T., Falcone, P., Asgari, F. B. J., & Hrovat, D. (2006). Predictive control
approach to autonomous vehicle steering. In 2006 American control conference
𝜓̇ = 𝑟 (D.1d)
Minneapolis. Minnesota, USA: IEEE, http://dx.doi.org/10.1109/ACC.2006.1657458.
𝑥̇ 𝑝 = 𝑣𝑥 cos (𝜓) − 𝑣𝑦 sin (𝜓) (D.1e) Lee, S. E., Llaneras, E., Klauer, S., & Sudweeks, J. (2007). Analyses of rear-end
crashes and near-crashes in the 100-car naturalistic driving study to support rear-
𝑦̇ 𝑝 = 𝑣𝑥 sin (𝜓) + 𝑣𝑦 cos (𝜓) (D.1f) signaling countermeasure development. USDOT/National Highway Traffic Safety
Administration.
𝛿̇ = 𝑑𝛿 (D.1g) Lim, E. H. M., & Hedrick, J. K. (1999). Lateral and longitudinal vehicle control coupling
for automated vehicle operation. In Proceedings of the American control conference,
San Diego, California (pp. 3676–3680).
Limpert, R. (1999). Brake design and safety. Society of Automotive Engineers.
References Markkula, G., Benderius, O., Wolff, K., & Wahde, M. (2012). A review of near-collision
driver behavior models. Human Factors, 54(6), 1117–1143. http://dx.doi.org/10.
Ackermann, C., Bechtloff, J., & Isermann, R. (2015). Collision avoidance with combined 1177/0018720812448474.
braking and steering. In P. Pfeffer (Ed.), Springer fachmedien wiesbaden 2015, 6th Mokhiamar, O., & Abe, M. (2004). Simultaneous optimal distribution of lateral and
international munich chassis symposium 2015 (pp. 199–213). http://dx.doi.org/10. longitudinal tire forces for the model following control. Journal of Dynamic Systems,
1007/978-3-658-09711-0_16. Measurement, and Control by ASME, 126, 753–763. http://dx.doi.org/10.1115/1.
Adams, L. D. (1994). Review of the literature on obstacle avoidance maneuvers: Braking 1850533.
versus steering: Tech. rep. UMTRI-94-19,, Ann Arbor, Michigan 48 109-2 150, U.S.A: National Highway Traffic Safety Administration (2016). Table 29 - crashes by first
The University of Michigan, Transportation Research Institute. harmful event, manner of collision, and crash severity. https://cdan.nhtsa.gov/
Aripina, M., Sam, Y., Kumeresan, A. D., Ismail, M., & Kemao, P. (2014). A review tsftables/tsfar.htm#.
on integrated active steering and braking control for vehicle yaw stability system. National Highway Traffic Safety Administration (2017). 2016 fatal motor vehicle crashes:
Journal Technology, 71(2), 105–111. Overview: Tech. rep., U.S. Department of Transportation, DOT HS 812 456.

14
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654

Quirynen, R., Vukov, M., Zanon, M., & Diehl, M. (2014). Autogenerating microsecond Wang, X., Zhu, M., Chen, M., & Tremont, P. (2016). Drivers’ rear end collision
solvers for nonlinear MPC: a tutorial using ACADO integrators. Optimal Control avoidance behaviors under different levels of situational urgency. Transportation
Applications & Methods. Research Part C (Emerging Technologies), (71), 419–433.
Shah, J. (2015). Development and control of evasive steer assist using rear wheel steering: Yi, B., Gottschling, S., Ferdinand, J., Simm, N., Bonarens, F., & Stiller, C. (2016). Real
SAE technical paper 2015-26-0004, http://dx.doi.org/10.4271/2015-26-0004. time integrated vehicle dynamics control and trajectory planning with MPC for
Shyrokau, B., Wang, D., Augsburg, K., & Ivanov, V. (2013). Vehicle dynamics with brake critical maneuver. In: IEEE intelligent vehicles symposium (IV) Gothenburg, Sweden.
hysteresis. Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Zegelaar, P. (2017). For evasive steering assist ’steering you out of trouble’. In 11th
automobile engineering, 227(2), 139–150. international IQPC conference steering systems Berlin.
Shyrokau, B., Wang, D., Savitski, D., Hoepping, K., & Ivanov, V. (2015). Vehicle Zhou, J., Lu, J., & Peng, H. (2010). Vehicle stabilisation in response to exogenous im-
motion control with subsystem prioritization. Mechatronics, 30, 297–315. http: pulsive disturbances to the vehicle body. International Journal of Vehicle Autonomous
//dx.doi.org/10.1016/j.mechatronics.2014.11.004. Systems, 8(2/3/4), 242–262. http://dx.doi.org/10.1504/IJVAS.2010.035798.
Soudbakhsh, D., & Eskandarian, A. (2010). A collision avoidance steering controller using Zhu, L., Shyrokau, B., Boulkroune, B., van Aalst, S., & Happee, R. (2018). Performance
linear quadratic regulator: SAE international 2010-01-0459. benchmark of state-of-the-art lateral path-following controllers. In 15th international
Vukov, M., Domahidi, A., Ferreau, H. J., Morari, M., & Diehl, M. (2013). Auto-generated workshop on advanced motion control, Tokyo, Japan.
algorithms for nonlinear model predicitive control on long and on short horizons.
In: Proceedings of the 52nd conference on decision and control (CDC).

15