integrated nonlinear model predictive control for
TRANSCRIPT
Delft University of Technology
Integrated nonlinear model predictive control for automated driving
Chowdhri, Nishant; Ferranti, Laura; SantafΓ© Iribarren, Felipe; Shyrokau, Barys
DOI10.1016/j.conengprac.2020.104654Publication date2021Document VersionFinal published versionPublished inControl Engineering Practice
Citation (APA)Chowdhri, N., Ferranti, L., SantafΓ© Iribarren, F., & Shyrokau, B. (2021). Integrated nonlinear modelpredictive control for automated driving. Control Engineering Practice, 106, [104654].https://doi.org/10.1016/j.conengprac.2020.104654
Important noteTo cite this publication, please use the final published version (if applicable).Please check the document version above.
CopyrightOther than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consentof the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Takedown policyPlease contact us and provide details if you believe this document breaches copyrights.We will remove access to the work immediately and investigate your claim.
This work is downloaded from Delft University of Technology.For technical reasons the number of authors shown on this cover page is limited to a maximum of 10.
Control Engineering Practice 106 (2021) 104654
INa
b
c
d
A
KMOINVERMC
1
rwtnain(fRct0esR
b
hRA0(
Contents lists available at ScienceDirect
Control Engineering Practice
journal homepage: www.elsevier.com/locate/conengprac
ntegrated nonlinear model predictive control for automated drivingishant Chowdhri a, Laura Ferranti b, Felipe SantafΓ© Iribarren c, Barys Shyrokau d,β
Control Systems Engineer, Chassis & Powertrain Development Department, Toyota Gazoo Racing Europe GmbH, GermanyLearning and Autonomous Control Section, Department of Cognitive Robotics, Delft University of Technology, NetherlandsR&D: Chassis Control, Toyota Motor Europe, BelgiumSection of Intelligent Vehicles, Department of Cognitive Robotics, Delft University of Technology, Netherlands
R T I C L E I N F O
eywords:odel predictive controlptimal control
ntegrated controlonlinear controlehicle controlvasive actionear-end collisionIMO systemollision avoidance
A B S T R A C T
This work presents a Nonlinear Model Predictive Control (NMPC) scheme to perform evasive maneuvers andavoid rear-end collisions. Rear-end collisions are among the most common road fatalities. To reduce therisk of collision, it is necessary for the controller to react as quickly as possible and exploit the full vehiclemaneuverability (i.e., combined control of longitudinal and lateral dynamics). The proposed design relies onthe simultaneous use of steering and braking actions to track the desired reference path and avoid collisionswith the preceding vehicle. A planar vehicle model was used to describe the vehicle dynamics. In addition,the dynamics of the brake system were included in the NMPC prediction model. Furthermore, the controllerincorporates constraints to ensure vehicle stability and account for actuator limitations. In this respect, theconstraints were defined on Kamm circle and Ideal Brake Torque Distribution (IBD) logic for optimal tire forceand brake torque distribution. To evaluate the design, the performance of the proposed NMPC was comparedwith two "more classical" MPC designs that rely on: (i) a linear bicycle model, and (ii) a nonlinear bicyclemodel. The performance of these three controller designs was evaluated in simulation (using a high-fidelityvehicle simulator) via relevant KPIs, such as reference tracking Root Mean Square (RMS) error, controllerβsrise/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 thepresence of lateral wind disturbances, road friction variation, and maneuver aggressiveness.
. Introduction
The U.S. National Highway Traffic Safety Administration (NHTSA)eported that the number of fatal crashes in 2016 increased by 5.6%ith a toll of 37,461 deaths (National Highway Traffic Safety Adminis-
ration, 2017). The European Road Safety Observatory reported similarumbers for the EU (European Road Safety Observatory, 2018). In thettempt to reduce the number of fatalities on the road, the automotivendustry started equipping the vehicles with active vehicle safety tech-ologies, such as Antilock Brake System (ABS), Vehicle Stability ControlVSC) and traction control. These technologies halved the number ofatalities from 20,774 in year 2007 to 11,990 in year 2016 (Europeanoad Safety Observatory, 2018). But, the NHTSA reported that rear-endollisions are the main cause of road fatalities (accounting for morehan 30% of all the road fatalities) (Insurance Information Institute,000; National Highway Traffic Safety Administration, 2016). A rear-nd crash occurs when the difference in relative speeds between theubject vehicle (SV) and lead vehicle (LV) in front causes a collision.ear-end crashes are extremely common in both urban and highway
β Corresponding author.E-mail addresses: [email protected] (N. Chowdhri), [email protected] (L. Ferranti), [email protected] (F.S. Iribarren),
[email protected] (B. Shyrokau).
environments, with a collision rate of 1 accident per 8 s (Gilreath& Associates, 2013). These accidents are caused by the inability ofthe human driver to perform an evasive maneuver to avoid collidingwith the LV successfully. According to Adams (1994), Beal and Gerdes(2009), Markkula, Benderius, Wolff, and Wahde (2012) and Wang, Zhu,Chen, and Tremont (2016), the major reasons behind the human-driverfailure are associated with (i) the driver preference towards brakingrather than steering, (ii) the longer driver reaction time, (iii) the driverinability to control the vehicle during highly nonlinear and criticalmaneuvers, (iv) fear and anxiety. The linear regime of motion (i.e., thelinear handling behavior) comes naturally to the driver. As soon as thevehicle is pushed to the handling limits (for example, during an evasivemaneuver at high speed), the situation becomes challenging for thedriver (Hac & Bodie, 2002).
Automated technologies such as Emergency Driving Support (EDS)can be extremely beneficial in this context. An EDS consists of fivemain components, that are, Risk Monitoring, Driver Monitoring, DecisionMaking, Path Planning, and Control (Choi, Kim, & Yi, 2011). This study
ttps://doi.org/10.1016/j.conengprac.2020.104654eceived 19 March 2020; Received in revised form 17 July 2020; Accepted 6 Octovailable online xxxx967-0661/Β© 2020 The Author(s). Published by Elsevier Ltd. This is an open acceshttp://creativecommons.org/licenses/by/4.0/).
ber 2020
s article under the CC BY license
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
cCJ
fuaooardo2(iicmssbcti
Mlta
focuses on the last component of the EDS design, that is, the designof a tailored control strategy for the rear-end collision scenario. Thedesign of such a controller is an active research area. Extensive reviewson various control strategiesβsuch as PID control, Sliding Mode Con-trol (SMC), Linear Quadratic Regulator (LQR), Nonlinear backsteppingcontrol, etc.βwere presented in Ackermann, Bechtloff, and Isermann(2015), Aripina, Sam, Kumeresan, Ismail, and Kemao (2014), Choi et al.(2011), Mokhiamar and Abe (2004), Shah (2015), Soudbakhsh andEskandarian (2010) and Zhu, Shyrokau, Boulkroune, van Aalst, andHappee (2018).
According to Ackermann et al. (2015), Choi et al. (2011) andMokhiamar and Abe (2004), an effective control design in the contextof EDS should:
1. Involve both steering and VSC via Differential Braking (DB).2. Optimally distribute the steering and brake control actions to
improve the overall vehicle performance.3. Handle tire nonlinearities during highly dynamic situations
(e.g., during an evasive steering maneuver).
Hence, an integrated (i.e., involving both steering and DB), optimal, andnonlinear control design should be able to handle an evasive maneuversuccessfully.
PID and SMC control are not optimal in nature. While LQR doesprovide optimal control, it only works for unconstrained optimiza-tion problem which is a limitation for vehicle control as the vehicledynamics are always bounded within the designed operating range.MPC on the other hand covers all the three conclusions made underone control design and becomes the most suitable control algorithmfor vehicle control. Since MPC is an optimal control technique andis based on the designed prediction model, it can accommodate vehi-cle nonlinearities and Multi-InputβMulti-Output (MIMO) models in itsdesign. Therefore our goal is to design an EDS controller that relieson NMPC to accommodate all control objectives above. By relying onan augmented nonlinear planar vehicle model as prediction model, theproposed NMPC design allows to simultaneously control the lateral andlongitudinal vehicle dynamics via steering and braking, while takingactuator dynamics into account. The proposed solution has brake actu-ator dynamics modeled inside its prediction which allows direct controlof the wheels and does not require any additional control allocationscheme (which would be nontrivial to implement). In addition, NMPCallows to directly account for tire saturation limitations and actuatorlimits in the constraint formulation. Most of the literature in the areaof MPC for evasive maneuvers focuses on lateral control at constantlongitudinal speed and relies on simplified vehicle models (e.g., thebicycle model) (Beal & Gerdes, 2009, 2013; Choi, Kang, & Lee, 2012;Keviczky, Falcone, Asgari, & Hrovat, 2006). The main reasons for thischoice is that more complex vehicle and tire models are challenging toimplement in real-time framework. Using a dynamic bicycle model asprediction model, however, limits the controller to exploit DB. This isbecause one requires control of the left and right wheels for DB to work.But in bicycle model, both the front and rear tires are lumped togetherrespectively as one tire each as a result of which the effect of DB is notwell captured in the dynamics of bicycle model. From the vehicle dy-namics perspective, DB plays a fundamental role to ensure safety duringevasive maneuvers. Compared to the aforementioned controllers, theproposed design exploit the benefits of DB by controlling each of thewheels directly and allows to control longitudinal and lateral dynamics.This is achieved by modeling the brake actuator dynamics inside theprediction model with the planar vehicle model to have an overalloptimal control strategy and removing the need of conventional controlallocation schemes, making it a unique MPC-based controller designthat allows direct control of the vehicleβs wheels.
The designed planar vehicle-based integrated NMPC control wasvalidated in several different scenarios, ranging from highly dynamicsingle-lane-change evasive maneuvers (to replicate scenarios in which
rear-end collisions occur if not properly handled) to normal lane change2
maneuvers. The controller was tested at varying vehicle velocities. Itsperformance was also validated in the presence of external disturbancessuch as lateral wind and parameter uncertainty via varying the roadfriction coefficient. The integrated NMPC control design is not limitedto EDS and can be used as a controller for automated driving for high-way and urban driving environments, provided that a path-planningalgorithm provides a suitable trajectory for the proposed controllerto follow (for example, the vulnerable-road-users-aware path-planningalgorithm proposed in Ferranti et al., 2019). Finally, using specific KeyPerformance Indicators (KPIs), the proposed integrated nonlinear MPCdesign was compared with the following baselines: (i) an MPC designthat uses a linear bicycle model as prediction model (referred to asthe Linear MPC design), and (ii) an MPC design that uses a nonlinearbicycle model as prediction model (referred to as the Nonlinear MPCdesign). In all test cases, the proposed design outperforms the baselinecontrollers.
1.1. Related work
The literature survey provided limited work in the field of integrated-ontrol design using MPC (Barbarisi, Palmieri, Scala, & Glielmo, 2009;hoi & Choi, 2016; Falcone, Tseng, Borrelli, Asgari, & Hrovat, 2008;alali, Khosravani, Khajepour, Chen, & Litkouhi, 2017; Yi et al., 2016).
The authors in Falcone, Borrelli, Asgari, Tseng, and Hrovat (2007)ormulated the NMPC problem for a double-lane-change maneuversing a bicycle model as system model with the steering wheel angles control command. The designed NMPC worked successfully at speedf 7 m/s but failed to stabilize the vehicle at 10 m/s. The authorsf Falcone et al. (2007) concluded that integrated control of steeringnd braking can improve the performance of the controller. The sameesearch team then designed a NMPC based control with 1s as pre-iction horizon to optimize combination of braking and steering forbstacle avoidance via double-lane-change maneuver (Falcone et al.,008). They used a 10 DoF planar vehicle model as prediction modelfirst six DoF being the vehicleβs longitudinal and lateral velocity, head-ng angle and yaw rate, and the vehicleβs global position coordinatesn both longitudinal and lateral direction. The four wheelβs dynamicsonsidered individually are the remaining four DoF) and used a Pacejkaodel to model the tire characteristics. The control action was the front
teering angle and each wheelβs brake torque values. The controlleruccessfully passed the test at 14 m/s. The controller, however cannote applied for real-time applications because it took around 15 min toomplete a 12 s simulation. In addition, high amount of oscillations inhe steering angle were observed due to improper tuning because ofncreased number of model parameters.
The authors in Jalali et al. (2017) designed an integrated LinearPC (LMPC) control using Active Front Steering (AFS) and DB for
ateral stability of the vehicle. The authors use bicycle model as predic-ion model with a prediction horizon of 0.3 s. The controller providesssistance control of Β±10 deg on the road wheel angle, satisfying the
side-slip angle π½ constraint to ensure vehicle stability at all times. Thecontroller, however, is not subjected to robustness tests, such as winddisturbance or parameter uncertainty.
Similarly, the authors in Choi and Choi (2016) designed MPC con-troller via an extended bicycle model that utilized AFS and DB forvehicle stability. In their work, the prediction model encapsulatedthe lagged characteristics of actuator dynamics and tire forces, bothmodeled as a first-order lag system. By calculating the control actionas steering wheel angle and yaw moment correction ππ§, anotheroptimization problem was solved to get the optimal tire forces, therebyincreasing the overall computational time and loss of performance. Bysolving two optimization problems, the idea of having one integratedcontroller for vehicle control was lost.
The authors in Barbarisi et al. (2009) designed a Vehicle DynamicsControl using linear time-varying MPC with sampling time and predic-tion horizon as 0.25 s and 5 steps, respectively. They assumed constant
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
Kfcc
swtascsb
1
ceicht
o
t
mo
abapir
2
2
chtrSnohsm
t
i0(vpcicw
trπ
longitudinal dynamics with control on lateral vehicle motion only.While the controller was able to pass the standard ISO 19365:2016Sine with Dwell test, the controller showed oscillatory behavior whenworking close to the constraint boundaries.
Lastly, the authors in Yi et al. (2016) designed two MPC controllersfor collision avoidance control using steering and braking combinedas control action. They used a nonlinear bicycle model as predictionmodel for the NMPC controller design and for the linear MPC design,they linearized the bicycle model around the operating point along withapproximation of the nonlinear constraints into linear form. But insteadof calculating brake torques, their MPC control action is the com-manded longitudinal acceleration which is then used in another logic tocalculate required brake torques. They tested their design on a single-lane-change maneuver at 70 km/h with target lateral displacement 2m. They set the sampling time and prediction horizon at 0.06 s and 25steps, respectively. While the collision was avoided, it was seen withboth the control strategies that an overshoot of about 35% was achievedin lateral position tracking, leading to poor tracking performance. Also,the NMPC designed was not real-time feasible with mean computationtime between 4 to 8 s reported.
Compared to the previous approaches, the proposed integratedNMPC control design relies on a single controller to compute thecontrol action for the four wheels, while taking into account the ve-hicle limitations. The integrated NMPC design solves the optimizationproblem online and in real-time. In addition, the actuator dynamicswas modeled in the prediction model to account for their reactiontime and have more accurate predictions. Furthermore, a kinematicreference path1 (i.e., the reference is generated based on a kinematicdescription of the vehicle, without taking into account dynamics) asreference trajectory was used in all the simulations to further assess thecontrollerβs robustness towards tracking imprecise (from the dynamicpoint of view) reference values. Lastly, this approach in its modelinginvolves the use of dynamic constraints aimed at maximizing thevehicle stability by minimizing the vehicle body slip angle and bodyslip angle rate. Hence, the control design accounts for gβg diagram2 and
amm circle3 constraints by design to ensure that the vehicle and theour tires operate in the stable working regime of motion. In addition,ompared to the state of the art, the integrated NMPC control designonsiders IBD constraints for ideal distribution of brake torques.
The robustness of the proposed integrated NMPC design was exten-ively tested to various disturbances and uncertainties such as lateralind and road friction variation. The maneuverβs nature during these
ests was kept evasive at all times to test controllerβs robustness underggressive and nonlinear vehicle dynamics conditions. For all the testedcenarios, the integrated NMPC design was successful to avoid rear-endollisions at all times thereby guarantying vehicle stability. All theseimulations were tested on a high-fidelity vehicle simulator providedy Toyota and validated by field tests.
.2. Paper structure
The paper is structured as follows. Section 2 introduces preliminaryoncepts used in the paper, such as the designed maneuver, the ref-rence trajectory, and model predictive control. Section 3 details thentegrated NMPC formulation. Section 4 details the two benchmarkontrollers used for performance comparison. Section 5 presents theigh-fidelity vehicle simulator. Section 6 explains the KPIs to quan-itatively assess the controllers performance. This section also gives
1 Usually, model-based path planners rely only on kinematic descriptionsf the vehicle and do not take into account its dynamics.
2 gβg diagram characterizes the vehicleβs lateral and longitudinal accelera-ion performance envelope under a given road surface condition.
3 Also referred as friction circle or friction ellipse, Kamm circle defines theaximum values of the resultant of longitudinal and lateral force that can be
btained under a particular operating condition.
a3
Fig. 1. Single lane change maneuver.
n overview of all the scenarios covered to analyze the performanceetter. Section 7 presents the results for each maneuver performed. Itlso details the KPIs values for every maneuver and provides an com-arative performance analysis between the three controllers designedn this research. Finally, Section 8 concludes the paper and providesecommendations for future work.
. Preliminaries
.1. Single lane change maneuver
To evaluate the designed controller, a representation of a rear-endollision avoidance maneuver is required. To the best of our knowledge,owever, no standard maneuvers for evasive action are available inhe literature (when referred to SAE, NHTSA, ISO and Euro NCAP,espectively). Nevertheless, the NHTSA report (Lee, Llaneras, Klauer, &udweeks, 2007) provides certain key insights concerning real-life sce-arios in which rear-end collisions occur. By collecting the crash dataf 100 different cars, the NHTSA highlighted that collisions frequentlyappen on a straight road with no junctions while driving at constantpeed. Hence, based on the NHTSA conclusions, the single-lane-changeaneuver was used shown in Fig. 1 and
parameter values provided by Ford Motors Research were selectedo represent an evasive scenario (Zegelaar, 2017). The values are πref= 2.5 m and πΏref = 30 m, that is, the maneuver begins when thesubject vehicle is exactly 30 m away from lead vehicle and has tolaterally traverse 2.5 m to avoid the rear-end collision. To ensure thatthe designed maneuver is indeed aggressive and nonlinear in nature,the vehicle speeds were selected to perform the maneuver according tothe following metrics:
TTC =πΏrefπ£π₯
, (1a)
TTB =π£π₯
2ππ₯max
, (1b)
TTS =
β
2πrefππ¦max
, (1c)
where Time to Collision (TTC), Time to Brake (TTB) and Time to Steer(TTS). Here, π£π₯ refers to the subject vehicleβs longitudinal speed, ππ₯maxs the subject vehicleβs maximum longitudinal acceleration (taken as.8 ΞΌg) and ππ¦max is the subject vehicleβs maximum lateral accelerationtaken as 0.6 ΞΌg), with π being the friction coefficient. The accelerationalues mentioned above were chosen based on the baseline valuesrovided by NHTSAβs definition of a Near-Crash which states thatircumstances involving vehicle braking greater than 0.5 g or steeringnput leading to lateral acceleration greater than 0.4 g to avoid a crashonstitutes a rapid maneuver scenario (Lee et al., 2007). Therefore itas ensured that the designed maneuver is evasive at all times.
Eq. (1) was evaluated at different vehicle speeds and ensured thathe inequality TTS β€ TTC β€ TTB is satisfied at all times to designealistic test scenarios. Table 1 reports the speed range for each value of. The small values of TTC highlights that the maneuver is aggressivend captures the real-life evasive situations.
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
tyal
π¦
π
π
Fa
2
poi
wπimmioc
thm
otdPDiMaibqtsPiac
cwwfcvioc
3
Ftppt
3
sticHi
Koihaor
TaTTar
Lt
Table 1Reference maneuver speed range and TTC values.π π£π₯min
TTCmin π£π₯maxTTCmin
β [km/h] [s] [km/h] [s]
1 78 1.38 115 0.930.9 75 1.44 110 0.980.8 70 1.54 102 1.060.7 65 1.66 97 1.130.6 60 1.80 90 1.200.5 55 1.96 82 1.310.4 50 2.16 72 1.500.3 43 2.51 64 1.680.2 35 3.08 50 2.160.1 25 4.32 35 3.08
The single-lane-change maneuver should be translated into a pathhe vehicle can follow in terms of lateral position, heading angle, andaw rate. Hence, the lateral reference position is approximated by usingSigmoid curve (Choi et al., 2011) defined as a function of vehicleβs
ongitudinal position π₯ as follows:
ref =π΅
1 + πβπ(π₯βπ). (2)
In addition, the reference heading angle πref and reference yaw rate ref are defined as follows:
πref = tanβ1(
ππ¦refππ₯
)
(3a)
ref = π 1π£π₯, (3b)
or the readability of the paper, Appendix A details the quantitiesssociated with the definition of the reference signals.
.2. Model predictive control
Model Predictive Control (MPC) solves a constrained optimizationroblem online to compute the optimal sequence of control commandsver a finite time window, called prediction horizon. The problems formulated based on (i) the available plant measurements, (ii) the
plant-prediction model, (iii) control objectives, and (iv) plant/actuatorlimitations. Only the first control command of this sequence is appliedto the plant in closed loop in the receding-horizon fashion. The predic-tion model captures the plant dynamics and gives controller the abilityto predict the behavior of plant. The prediction model, as this work alsoshows, is fundamental for the performance of the controller.
A general formulation of MPC controller is given by
minπ
ππβ1β
π=0π½π
(
ππ, ππ, πrefπ
)
+ π½π(
πππ , πrefππ
)
(4a)
s.t.:ππ+1 = π(
ππ, ππ)
, π = 0,β¦ , ππ β 1 (4b)
πΊ(
ππ, ππ)
β€ ππ, π = 0,β¦ , ππ β 1 (4c)
πΊ(
πππ
)
β€ ππ (4d)
π0 = πinit, (4e)
here π½π is the cost function to be minimized for optimal control actionπ. ππ and πref
π are the states and the reference values at predictionnstant π (π = 0,β¦ , ππ), respectively. Function π is the predictionodel that captures the plantβs dynamics. πinit is the current stateeasurement from the plant and updated online at every sampling
nstant. Finally, function πΊ comprises of all the constraints definedn the states and control action with ππ being the bound value. Theonstraints can be either convex or nonconvex.
There are several toolboxes that can be used to solve Problem (4). Inhis work the ACADO Toolkit (Quirynen, Vukov, Zanon, & Diehl, 2014)as been used. ACADO tackles nonlinear optimal control problems andulti-objective optimal control problems efficiently. In ACADO, the
r4
ptimal control problem (OCP) is discretized. ACADO relies on Sequen-ial Quadratic Programming (SQP). The SQP algorithm linearizes theiscretized nonlinear control problem to convert it into a Quadraticrogramming (QP) problem (Vukov, Domahidi, Ferreau, Morari, &iehl, 2013). Using condensing techniques, the state variables are elim-
nated (i.e., the overall number of optimization variables is reduced).ultiple shooting method was preferred over single shooting method
s it is more robust for nonlinear systems, such as the vehicle dynamicsn the non-linear handling area. Once the optimization problem haseen linearized, an Active Set Method was used to solve the resultinguadratic programming problem. LevenbergβMarquardt Gauss New-on based hessian approximation method was used, and, for activeet method, open-source C++ software qpOASES (Ferreau, Kirches,otschka, Bock, & Diehl, 2014) was selected. In addition, ACADOmplements the Real Time Iteration (RTI), which involves performingsingle SQP iteration per sampling time, for a more efficient solution
alculation (Vukov et al., 2013).In the current study, to mitigate the effects of the unpredictable
omputation times, the maximum amount of iterations of the solveras fixed. If the optimizer reaches the maximum number of iterationsithout converging to an optimal solution, then previously computed
easible prediction (shifted one step forward) is used. This previouslyomputed prediction is used to warm-start the optimizer. Since theehicle dynamics (plant model) are continuous in nature, sudden jumpsn its dynamics are less likely. This gives a higher chance that the nextptimal solution is close to the previous one. Both modifications areommon practice in practical MPC implementations.
. NMPC controller design
This section describes the proposed integrated NMPC formulation.irst, the various dynamic couplings in the vehicle model are discussedhat should be well captured in the prediction model to improve theerformance of the controller. Then, the planar vehicle model used asrediction model in the proposed NMPC design is discussed. Finally,he control objectives and the constraints are described.
.1. Vehicle dynamics coupling
Designing an integrated control is a non-trivial problem due to thetrong couplings in the vehicle dynamics as explained below. Based onhe knowledge of vehicle dynamics and its associated coupling effect,t is therefore essential for developing the prediction model of the MPController. According to Attia, Orjuela, and Basset (2012) and Lim andedrick (1999), the following longitudinal and lateral couplings arise
n case of vehicle motion:
inematic and dynamic coupling. This coupling arises due to the effectf wheel steering on longitudinal dynamics of the vehicle by chang-ng the tire lateral forces. The tire longitudinal forces on the otherand affects both the lateral dynamics and yaw motion of vehiclend subsequently the rate change of lateral position is a functionf longitudinal velocity. Thus both dynamics are always coupled aseflected in (B.1)β(B.6) (detailed in Appendix B.1).
ireβroad coupling. This coupling arises due to the application of lateralnd longitudinal forces by the tire. This coupling is reflected in (17).he equations used for the lateral forces are reported in Appendix B.2.o model the longitudinal forces, the single corner model was useds shown in Fig. 3. The equations used for the longitudinal forces areeported in Appendix B.3.
oad transfer phenomenon. This coupling arises because of the loadransfer during longitudinal and lateral accelerations. This coupling is
eflected in (22a)β(23e) (detailed in Section 3.3).N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
π₯
wspt
tmAwf
R(cbcbrioactlcc
Raiect
3
daMtcafafa
0
Fig. 2. Planar vehicle model.
3.2. Prediction model
The designed prediction model for integrated NMPC accounts forthese coupling effects and captures the behavior of vehicle in bothlinear and nonlinear regime of motion. A planar vehicle model wasselected as prediction model of the integrated NMPC design, becauseit allows to preserve the effect of the couplings.
Following modeling simplifications were made, which are mostlymade when using a planar vehicle model:
β’ Roll, pitch, and vertical motion are not modeled.β’ Suspension and compliance effects are not considered.β’ Ackermann geometry is not considered (i.e., the front left and
front right wheel turn by same amount).β’ Since the actuator delay is small for both the front and rear
calipers, it has been ignored in the prediction model formulation.
Fig. 2 provides an overview of the main quantities of the predictionmodel. The planar vehicle is described by the following equation:
(π‘) = π (π₯(π‘), π’(π‘)), (5)
where π₯(π‘) β R15, π’(π‘) β R5, and π βΆ R15 Γ π 5 β R15. To improve thereadability of the paper, Appendix B.1 details π (π₯, π’). The state vectoris given by π₯ = [π£π₯, π£π¦, π, π, π₯π, π¦π, πΏ, ππflact
, ππfract, ππrlact
, ππrract,
ππflcal, ππfrcal
, ππrlcal, ππrrcal
] where π£π₯, π£π¦, π, π, π₯π, and π¦π are the lon-gitudinal velocity, lateral velocity, yaw rate, heading angle, global xand y positions of the vehicle, respectively. The brake torque ππijact
arethe actual brake torque value for tire ππ applied to wheel after the brakeactuator dynamics. The last four states ππijcal
are the brake torque valuescalculated before the brake actuator dynamics. The control vector isπ’ = [ππΏ πππfl πππfr πππrl πππrr ] where ππΏ , πππfl , πππfr , πππrl , and πππrr are therate of change of road wheel angle and brake torque rate for eachwheel, respectively. The control action ππΏ is applied to the vehicleβssteering system as steering wheel velocity (SWV) by multiplying thecontrol action with steering ratio π st.
The considered maneuver is aggressive and nonlinear in nature,therefore the model needs to capture the tire nonlinearities. Hence,instead of giving a fixed value to the cornering stiffness πΆπΌ , a Dugoff tiremodel was used to capture the tire nonlinear behavior. Appendix B.4details the model formulation. The fitted nonlinear cornering stiffnessπΆnonπΌ is given by:
πΆnonπΌij
=πΆπΌij
1 β π ijπ (π), (6)
here π is the longitudinal slip and π (π) is a function of the longitudinallip (its definition is reported in (B.17), Appendix B.4). This allows therediction model to preserve the overall tire behavior as well as theire dynamics for an accurate and superior control.
5
Fig. 3. Single corner model.
To make the future predictions more accurate, instead of keepingthe lateral and longitudinal forces (πΉπ¦ij and πΉπ₯ij ) constant throughouthe prediction horizon, they were formulated in terms of the predictionodelβs states. The equations used for the lateral forces are reported inppendix B.2. To model the longitudinal forces, the single corner modelas used as shown in Fig. 3. The equations used for the longitudinal
orces are reported in Appendix B.3.
emark 1. The designed prediction model (refer for details to (B.8)βB.9) in Appendix B) captures the brake actuatorβs dynamics. Hence, theontroller can understand how and at what rate will the brake pressureuild up in each wheelβs caliper, allowing accordingly to calculate theontrol action, that is, the four brake torque rates πππij . In addition,y including the actuator dynamics, the proposed strategy does notequire a separate controller design for control allocation, making it anntegrated approach and theoretically easy to implement as a hardwaren actual vehicle. Instead of only calculating the upper-level controlction, the designed integrated NMPC controller can now directlyalculate the control action that needs to be applied to the wheels. Inhis way, one controller can take care of all the two classical controlayers, that are, vehicle level and control allocation. This integratedontroller design is only possible with MPC due to its modular designoncept.
emark 2. The brake torques ππijactrepresent the actual brake torques
pplied to the wheels (i.e., the values after the brake actuator dynam-cs). Brake torques ππijcal
represent the values calculated before theffect of actuator dynamics. By providing the actuator dynamics, theontroller can compensate for possible performance losses to computehe final brake torque value ππijact
.
.3. Constraints
States and actuator limitations should be taken into account byesign for a realistic control and to eliminate actuator failure. Inddition, the controller should ensure that, if a feasible solution of thePC problem exists, this will keep the vehicle in the stability envelope,
hat is, the vehicle dynamics need to be constrained to a region in whichontrol of the vehicle can be ensure at all times. The MPC frameworkllows to incorporate these stability-envelope and actuator limits in theorm of constraints in the design phase of the controller. Hence, toccount for physical limitation and to define a handling envelope, theollowing constraints (19 in total) were defined that should be satisfiedlong the length of the prediction horizon of the NMPC controller:
β€ π£π₯ β€ 170 [km/h] (7)
β 5 β€ π½ β€ 5 [deg] (8)
β 25 β€ π½ β€ 25 [deg/s] (9)N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
(
pa
π½
hsinfda(tm
afc2wobdtcce
3
dat(ec
π½
w0tftAsemevt
3
e
mpasπ
β17π st
β€ πΏ β€ 17π st
[deg] (10)
β800π st
β€ οΏ½οΏ½ β€ 800π st
[deg/s] (11)
οΏ½οΏ½π₯ β π£π¦π)2 + (οΏ½οΏ½π¦ + π£π₯π)
2 β€ (ππ)2 (12)
0 β€ ππijactβ€ 4900 [N m] , ij = (fl, fr) (13)
0 β€ ππijactβ€ 1610 [N m] , ij = (rl, rr) (14)
β 7000 β€ οΏ½οΏ½πijactβ€ 7000 [N m/s] , ij = (fl, fr) (15)
β 5550 β€ οΏ½οΏ½πijactβ€ 5550 [N m/s] , ij = (rl, rr) (16)
(
πΉπ₯ij
)2+(
πΉπ¦ij
)2β€(
πijπΉπ§ij
)2, ij = (fl, fr, rl, rr) (17)
ππrlact+ ππrract
ππflact+ ππflact
+ πβ€
πππΏ +
βcg(οΏ½οΏ½π₯βπ£π¦π)ππΏ
1 β πππΏ β
βcg(οΏ½οΏ½π₯βπ£π¦π)ππΏ
(18)
Constraint (7) limits the vehicleβs speed. To ensure vehicle stability,constraints (8)β(9) limit both the vehicle sideslip angle π½ and thesideslip angle gradient οΏ½οΏ½. Based on the concept of stable π½ - οΏ½οΏ½ referenceregion by He, Crolla, Levesley, and Manning (2006) and the evaluationson the same phase plane by European Council Service Framework Pro-gramme (0000) and Shyrokau, Wang, Savitski, Hoepping, and Ivanov(2015), it was concluded that a bound of 5 deg for π½ and a bound of25 deg/s for οΏ½οΏ½ were reasonable to define a stable region for vehiclemotion. The constraints defined ensure that the vehicle remains withinthis stable region at all times and does not spin away. The vehiclesideslip angle π½ can be in terms of vehicle states so that the constraintis dynamic in nature and is always satisfied along the entire predictionhorizon using Eq. (19).
π½ = tanβ1( π£π¦π£π₯
)
(19)
Since, the bounds in constraints (8)β(9) are small angles, the ap-roximation tan π½ β π½ holds true which gives the final equation forpproximating vehicle slip quantities as shown in Eqs. (20)β(21).
=π£π¦π£π₯
(20)
οΏ½οΏ½ =οΏ½οΏ½π¦π£π₯
(21)
Constraints (10)β(11) limits the steering wheel angle and steeringwheel rate, respectively (note that by using the steering ratio π st theyave been written in the form of road wheel angle to directly bound thetate). Constraint (12) represents the gβg diagram constraint represent-ng the working limit of the vehicle. Since the controller is integrated inature and can control both lateral and longitudinal dynamics, there-ore accordingly the working envelope is defined. Constraints (13)β(16)efine the brake actuator limits in terms of maximum brake torquend rates. Constraint (17) represents the four Kamm circle constraintsone for each tire). These constraints prevent/minimize the effect ofire saturation. By making assumptions that the sprung and unsprungasses are lumped as total mass π, the roll angle π is small and the
dynamic terms of roll and pitch motion are ignored, that is, only thecontribution from the static terms are taken in modeling, the normalload (which is the right hand side of the bound) on each tire πΉπ§ij ,respectively, was defined in equations below:
πΉπ§fl = πΉ rearπ§,π β πΉπ§π₯ β πΉπ§π¦π , (22a)
πΉπ§fr = πΉ rearπ§,π β πΉπ§π₯ + πΉπ§π¦π , (22b)
πΉ = πΉ front + πΉ β πΉ , (22c)
π§rl π§,π π§π₯ π§π¦πi6
πΉπ§rr = πΉ frontπ§,π + πΉπ§π₯ + πΉπ§π¦π , (22d)
where
πΉ frontπ§,π =
ππππ2πΏ
, (23a)
πΉ rearπ§,π =
ππππ2πΏ
, (23b)
πΉπ§π₯ =π(οΏ½οΏ½π₯ β π£π¦π)βcg
2πΏ, (23c)
πΉπ§π¦π =π(οΏ½οΏ½π¦ + π£π₯π)
π‘π
(
ππβrfπΏ
+πΎπ,πβ
πΎπ,π +πΎπ,π β ππβ
)
, (23d)
πΉπ§π¦π =π(οΏ½οΏ½π¦ + π£π₯π)
π‘π
( ππβrrπΏ
+πΎπ,πβ
πΎπ,π +πΎπ,π β ππβ
)
, (23e)
nd β = βcg β (ππβrf + ππβrr)πΏβ1. Finally, Constraint (18) defines theront to rear brake torque distribution ratio based on the parabolicurve (right-hand side of Constraint (18) according to Breuer & Bill,008) for an ideal brake torque distribution. In straight-line driving,hen a vehicle brakes, it pitches forward, increasing the normal loadf the front tires. Therefore the ability of the front tires to generaterake force increases as compared to rear ones. Hence due to vehicleesign, usually in a straight-line driving, the front tires brake morehan the rear tires. Since this was not modeled in the prediction model,onstraint (18), which is only activated during straight-line driving,aptures the IBD behavior well and is defined with π in denominatorqual to 0.001 to ensure mathematical infeasibility is avoided.
.4. Cost function
The cost function incorporates the control objectives of the NMPCesign. It is designed to keep the tracking error between process outputnd given reference as small as possible and at the same time, minimizehe control action along the prediction horizon. Based on Barbarisi et al.2009), Falcone et al. (2008) and Jalali et al. (2017) a 2-square normrror minimization function was chosen to model the cost function. Theost function is defined as follows:
π =ππβ1β
π=1
[
βπ(π + π) βπref(π + π)β2π+ (24a)
βπ (π + π β 1)β2π]
+ (24b)
βπ(π +ππ) βπref(π +ππ)β2π , (24c)
here π is the state prediction, πref = [0, 0, πref, πref, 0, π¦ref, 0, 0, 0, 0,, 0, 0, 0, 0]T is the reference prediction, and π is the control predic-ion. The reference trajectories π¦ref, πref and πref are the reference valuesor position, heading angle and yaw rate, respectively. In addition,he cost penalizes πΏ to control the magnitude of the Steering Wheelngle (SWA) at higher speeds (high SWA and SWV may lead to vehiclepinning out). Furthermore, the cost penalizes the brake torques tonsure that minimum control action energy is utilized to perform theaneuver. Lastly, the cost penalizes all the five control actions to
nsure that the entire maneuver can be performed at minimum controlalues. This ensures that the control energy cost is minimized, reducinghe actuator wear and improving its service life as well.
.5. Controller tuning
The proposed design involves the selection of several tuning param-ters. This section details and motivates the design choices.
The sampling time π‘π of the controller to 0.035 s has been chosenotivated by the cycle update time of all the other ECUβs of theassenger car. This ensures that at each sample, the controller hasdequate information of all the reference signals and vehicleβs states toolve the optimization problem. Furthermore, the prediction horizonπ of the proposed controller is set to 30 steps (i.e., 1.05 s) to max-
mize its performance (according to the assessment criteria described
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
ip
d
Fig. 4. Bicycle model.
n Section 6.2) while ensuring computational feasibility (i.e., real-timeerformance).
The remaining tuning parameters are the weight matrices π βͺ° 0,π β» 0, and π βͺ° 0 in (24). Tuning these matrices is nontrivial, given themultiple objectives the controller needs to optimize. Tuning is basedon a trade-off to achieve small tracking error balancing the use of thecontrol actions. Compared to other control techniques (such as rule-based controllers), tuning of these weighting matrices is based on thedynamics and kinematics of the vehicle. For example, giving a highpenalty to the control action means that the actuators should stay asclose as possible to zero, which is unrealistic in the scenarios consideredin this work.
The initial tuning was refined using heuristics and simulations. Oneparameter at the time was varied and the KPIs were evaluated. Inaddition, given that the maneuver is performed for various speeds andfor various values of π, the tuning was performed at different speedsand for each value of π. By doing so, following patterns were identifiedwhich were used to schedule the weights in the designed cost function:
β’ Increasing the terminal position tracking tuning weight ππ¦π leadto corner cutting.
β’ Increasing value of yaw rate tuning weight ποΏ½οΏ½ improved theoverall reference tracking performance.
β’ Reducing tuning weight of wheel angle ππΏ improved trackingperformance.
β’ For a fixed π value with increasing maneuver speeds, the trackingwas improved by increasing the weights of road wheel angle andwheel velocity (ππΏ and π οΏ½οΏ½), and by reducing the weight of lateralposition ππ¦.
β’ Decreasing the tuning parameter of control action brake torquerate π οΏ½οΏ½πfl , π οΏ½οΏ½πfr , π οΏ½οΏ½πrl , π οΏ½οΏ½πrr and keeping other tuning parametersconstant lead to increase in overshoot.
4. Benchmark controller design
This section describes the benchmark controllers designed to com-pare it against the proposed integrated NMPC control design. Comparedto the proposed controller, these designs rely on a simplified predictionmodel, that is a bicycle model represented in Fig. 4. This representationis based on the same assumptions made in the planar vehicle basedMPC control (Section 3.2) with addition of one more assumption thatno longitudinal or lateral load transfer is considered.
Compared to the planar vehicle model, the bicycle model has sevenstates, that are, π£π₯, π£π¦, π, π, π₯π, π¦π and πΏ and only a control action, thatis, the rate of change of the wheel steering angle ππΏ .4 Both benchmarkcontrollers have the same cost function and constraints (detailed in Sec-tion 4.1. In the following, additional details about the two benchmarkcontrollers are provided.
4 Given that in a bicycle model the two axle tires are clubbed as one,ifferential braking and individual wheel control are not an option.
7
4.1. Linear MPC
This controller involves the use of a linear bicycle model as pre-diction model. This representation requires the additional followingassumptions:
β’ Small angle approximation, that is, sin π β π, cos π β 1.β’ Constant longitudinal velocity, that is, ππ₯ = 0.β’ Linear tire model is used to capture the tire dynamics:
πΉπ¦ij = πΆπΌijπΌij (25)
The dynamics are detailed in Appendix C. In addition, compared to theproposed design, this design has only six constraints to ensure vehiclestability and bounded control action within feasible actuator range.Specifically, five constraints are similar to constraints (7)β(11), definefor the planar vehicle model. The last constraint bounds the lateralacceleration of the vehicle as follows:
β 0.85ππ β€(
οΏ½οΏ½π¦ + π£π₯π)
β€ 0.85ππ (26)
The cost function associated with the linear bicycle model is similar(with a reduced set of state and control objectives) to the one presentedin Section 3.4. Finally, ππ was set to 50 steps to predict 1.75 s in futureto improve the control performance. The horizon is longer with respectto the one selected using the planar vehicle model because the opti-mizer has to solve a smaller problem with less constraints and decisionvariables. While, with a shorter horizon the two baseline controllersprovided substandard tracking performance, with a longer horizon theintegrated design failed to meet the real-time requirements. Hence,to have a fair comparison among the controllers, different horizonlengths were selected for the integrated and baseline controllers. Whilea shorter horizon (due to real-time constraints) for the integrated designcould be seen as a limitation, its performance is still very good thanksto its ability to use a more accurate model to generate predictions.
4.2. Nonlinear MPC
This controller relies on a nonlinear bicycle model to capture thevehicleβs nonlinearities while performing the maneuver. In addition,compared to the previous design and similar to the proposed design,this controller relies on the Dugoff tire model to capture the tirenonlinearities. The states and control commands are the same of thelinear bicycle model, but they are non-linearly coupled. Appendix Ddetails on the nonlinear bicycle model. The constraints, cost function,and horizon length are those detailed for the linear bicycle model.
5. Vehicle simulator
The three controllers were tested and compared on an IPGCarMaker-based simulation platform using a high-fidelity Toyota ve-hicle model. The model has been parametrized based on mass-inertiaparameters obtained from vehicle inertia measuring facility, suspensionkinematics and compliance obtained by measurement on a Kinematics& Compliance test rig for wheel suspension characterization, and fi-nally, validated by field tests on the proving ground. A high-fidelity3-DoF steering model with column-based electric power steering logicwas used as the steering actuator model. This steering system wasvalidated with full-vehicle testing and it is implemented in the Toyotaβshigh-end driving simulator (Damian, Shyrokau, Ocariz, & Akutain,2019). To simulate tire dynamics, the Delft-Tyre 6.2 was used incombination with a detailed tire property file identified from benchtesting (pure and combined slip, transient dynamics).
The brakes considered in this research are floating point disk brakeswith conventional HAB system. The nonlinear HAB brake dynamicmodel derived from real-life vehicle data is the following (accordingto Zhou, Lu, & Peng, 2010):πact = πβππ π , οΏ½οΏ½act β€ π€ (27)
πcal πππ + 1N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
π€=bb(
π
hh2
6
6
vir
6
rt(
6
T
6
ftSN
6
vscwT
Table 2Varying velocity scenarios.π [β] π£π₯ [km/h]
0.9 (dry road) 75 80 85 90 95 100
Table 3Varying π scenarios.π [β] π£π₯ [km/h]
0.5 0.6 0.7 0.8 0.9 1.0 80
Table 4Varying wind speed scenarios.π [β] π£π₯ [km/h] π£π€ [km/h]
0.9 90 0 10 30 50 70
The parameters for the front axle are ππ = 0.06 s, ππ = 0.12 s and= 230 bar/s. For the rear axle, the parameters are ππ = 0.02 s, ππ0.05 s and π€ = 550 bar/s. The maximum pressure πmax that the
rakes can achieve is taken as 160 bar. To convert the brake pressure torake torque, the following relationship was used according to Limpert1999):
πijact= 2πactijπ΄wcijππijππΏij πij , ij = (fl, fr, rl, rr) (28)
The brake hysteresis effect was neglected as it is assumed that brakeysteresis has a minor influence on the brake performance for a newydraulic disk brake mechanism (Shyrokau, Wang, Augsburg, & Ivanov,013).
. Maneuver scenarios and assessment criteria
.1. Maneuver scenarios
The designed single lane change maneuver was performed under aariety of conditions to check the controller capabilities and robustnessn different scenarios. This paper presents the results for the mostelevant scenarios.
.1.1. Set 1 β varying velocity π£π₯These scenarios involve variations in vehicle speeds at a constant
oad friction coefficient. Table 2 summarizes these scenarios. Notehat the speed range was selected based on the results of Section 2.1Table 1).
.1.2. Set 2 β varying friction coefficient πThese scenarios involve variations in values of π for a given speed.
able 3 summarizes these scenarios.
.1.3. Set 3 β varying lateral wind velocity π£π€These scenarios involve variation of external lateral wind speeds π£π€
or a fixed value of π and π£π₯. Table 4 summarizes these scenarios. Notehe wind is modeled as constant perturbation to flow only in directionouth, directly opposing the vehicle as it turns left (towards directionorth) according to the defined maneuver.
.1.4. Set 4 β varying maneuverβs aggressivenessThese scenarios highlight the ability of the controller to handle
arious dynamic maneuvers ranging from evasive actions to normalingle-lane changes. In these scenarios, the parameter πΆ2 in the sigmoidurve decreases gradually. By doing so, the slope of the trajectoryas gradually reduced, making the reference trajectory less aggressive.
able 5 summarizes these scenarios.8
Table 5Varying πΆ2 scenarios.π [β] π£π₯ [km/h] πΆ2 [m]
0.9 90 5 4 3 2 1 0.5
Fig. 5. DTC graphical representation.
6.2. Assessment criteria
To assess the performance of the three controllers for all the scenar-ios defined in Section 6.1, following KPIs were defined.
The first KPIs selected were Overshoot (ππ), Settling Time (ππ ), andRise Time (ππ). These are typically used to assess the performance of acontroller to step reference signals (which is a close approximation ofthe reference trajectory described in Section 2.1 and graphically shownin Fig. 6a). To further assess the tracking performance of the controller,the RMS of the tracking errors over the horizon length of the controllerswas considered, that is:
πRMS =
β
β
β
β1π
πβ
π=1
(
π(π) βπref(π))2, (29)
where π β {π¦, π, οΏ½οΏ½} and πref β {π¦ref, πref, οΏ½οΏ½ref}, according to thedefinition of the reference signals in Section 2.1. Finally, the last KPIconsidered was Distance to Collision (DTC) (depicted in Fig. 5). This KPIrepresents the lateral distance between the left-rear corner of LV andright-front corner of SV. The DTC is a safety-based KPI and gives anidea of the safety margin the controller can produce.
For a good control performance, the DTC should be as high as pos-sible and all other KPIs should be as small as possible. This will ensurecollision avoidance, well tracked trajectories, and quick stabilization ofthe vehicle post lane change.
7. Simulation results
The controllers were tested in the scenarios described in Section 6.1.This section presents one specific case in more detail (with comparisonwith the benchmark controllers), that is the scenario π£π₯ = 90 km/hand π = 0.9. In addition, this section shows the KPI results for all thescenarios using the proposed design. Furthermore, the section showshow the proposed controller handles constraints by design and isreal-time feasible.
Comparison with the benchmark controllers. Fig. 6 compares the threecontrol strategies with respect to the reference signals. The dashed-bluelines represent the reference signals, the red, yellow, and purple linesrepresent the proposed controller, the linear MPC design, and nonlin-ear MPC design, respectively. The first plot of Fig. 6 shows that theproposed integrated NMPC control approach significantly reduces theovershoot compared to the linear MPC design (33% overshoot). In addi-tion, the proposed design provides better tracking performance (π¦RMS =7.39% and DTC = 0.41 m) compared to nonlinear MPC design (π¦RMS =11.25% and DTC = 0.26 m). The improved tracking performance is dueto the more detailed prediction model and integrated control action ofsteering and braking. In this respect, Fig. 7 highlights how the proposedapproach provides the necessary steering action. Recall that the single-lane change maneuver starts with a left turn, followed by a right turn,
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 π = 0.9 and π£π₯ = 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 thefigure shows, the left brakes brake while the right brakes are keptat zero. This gives the required additional yaw moment for bettertracking. During the second turn, the controller provides the controlaction to steer the steering wheel clockwise (i.e., negative SWA value).Simultaneously, the controller also reduces the left brakes and increasesthe right brakes to get desired yaw moment for tracking the referencevalues. Finally, in the last phase of straight-line driving, the SWA goesto zero. At the same time, the right and left brake values are modulatedto ensure the vehicle remains stable and aligned straight. Once done,the brake torques also go to zero to conclude the maneuver. The mostimportant KPI for collision avoidance is the DTC value. This is becausethe top priority in case of evasive action is collision avoidance whichis directly represented by DTC. A positive and non-zero value ensuresthat collision was avoided successfully. The higher the DTC values are,the higher the safety margins are. Figs. 8a and 9a presents the DTCresults for scenario sets 1 and 2 (Section 6.1). As the figures show, thedesigned integrated NMPC control outperforms the benchmark controlstrategies providing the highest DTC values. Also, both the benchmarkcontrollers fail to avoid the collision at 100 km/h (DTC value is zero)whereas integrated NMPC controller avoids the collision successfully.It is to be noted that DTC values are meaningful when looked alongwith trajectory tracking overshoot values. A higher overshoot valuemay result in high DTC value. In principle this reflects that collisionwas safely avoided but it does not highlight that trajectory trackingwas poor. Therefore, for both sets 1 and 2, the percentage overshootππ figures have also been plotted in Figs. 8b and 9b. It can be seen thatlinear MPC gives slightly higher DTC than nonlinear MPC. But linearMPC also gives a very high overshoot value as compared to nonlinearMPC. Therefore, with a marginal difference in DTC value and negligibleovershoot observed, the performance of nonlinear MPC is overall betterthan linear MPC. And integrated NMPC not only gives highest DTCvalue but also gives close to zero overshoot value, proving that it indeedperforms the best of all.
Lateral wind. Table 6 summarizes the KPI values for scenario set 3(i.e., lateral wind offset scenario) using the proposed controller. Forwind speeds up to 70 km/h, the controller is able to avoid the collisionsuccessfully. The table reports an additional parameter, namely π·off,that measures the offset distance between reference trajectory and thevehicle trajectory at the end of the maneuver. With such high windspeeds, the proposed controller returns a maximum offset value of 0.12m, judged smaller than existing references (β 0.5 m). This highlightsthe effectiveness and robustness of the proposed control scheme.
Maneuverβs aggressiveness. Table 7 summarizes the KPI values for sce-nario set 4 (i.e., varying maneuverβs aggressiveness) using the proposedcontroller. The table reports the maximum lateral acceleration gen-erated by the maneuver, that is ππ¦max , highlighting that for varyingdynamic scenarios, the controller is working efficiently. These resultsshow that the proposed design, thanks to its integrated ability tosimultaneously steer and brake, provides both lateral and longitudinal
control and avoids the collision in all scenarios successfully.9
Table 6KPIs for Set 3 using the proposed approach β varying lateral wind velocity π£π€scenario.π£π€ ππ ππ ππ π¦RMS πRMS οΏ½οΏ½RMS DTC π·off[kmβh] [%] [s] [s] [%] [%] [%] [m] [m]
0 1.34 3.10 0.53 5.71 55.24 387.90 0.41 0.0010 0.29 7.34 0.54 6.12 54.70 386.41 0.39 0.0130 0.08 7.72 0.58 7.75 53.94 384.16 0.35 0.0450 0.18 7.83 0.64 11.22 53.45 394.12 0.26 0.0870 0.11 8.06 1.95 15.21 56.21 411.01 0.18 0.12
Table 7KPI for Set 4 using the proposed approach β varying maneuverβs aggressivenessscenario.πΆ2 ππ ππ ππ π¦RMS πRMS οΏ½οΏ½RMS ππ¦max
[β] [%] [s] [s] [%] [%] [%] [m/s2]
5.0 1.34 3.10 0.53 7.37 71.30 500.64 5.834.0 2.25 3.11 0.66 6.46 72.92 389.33 4.783.0 2.78 3.47 0.77 6.38 56.82 252.89 3.182.0 2.12 4.04 0.89 4.69 33.58 121.79 2.221.0 0.00 5.01 1.18 2.44 13.48 36.04 1.230.5 0.00 5.75 1.49 1.37 6.52 12.20 0.77
IBD constraint satisfaction. To show the IBD constraint (18) is active,the lane change scenario with pre-braking maneuver was considered.Two seconds before the subject vehicle is 30 m away from lead vehicle,the subject vehicle will brake and decelerate. After this, the single lanechange maneuver is performed. This was done as the IBD constraint isonly activated during straight-line pre-braking maneuver. Fig. 10 showsthe brake torques. Due to this brake distribution constraint, front braketorques are more than the rear.
Constraint satisfaction. The controller is real-time feasible during themaneuver. Fig. 11 shows the calculation times for the number ofcalls for each considered scenario (5β6 simulations per each scenarioaccording to the Tables 2β5). In particular, it should be noted that thecomputation time increases when the reference signal changes. Thisbehavior is typical of the optimizer used to solve the nonlinear controlproblem. The computation time tends to increase with the number ofactive constraints. However, the computation time is still within thereal-time constraint highlighted by the dashed-red line.
The maximum number of iterations of the solver is empiricallyfixed to 5ππ (nX + nU + nC) with nX, nU and nC representing totalnumber of states of the prediction model, total number of control actionand total number of constraints respectively. The value is based onthe diagnostic flags of the solver during initial tests. The solver waswarm-started with the prediction computed at the previous time step(shifted by one in time). This helps mitigate the effect of unpredictablecomputation times. In addition as a backup, if the solver fails to find afeasible solution within the fixed number of iterations, the last feasible
solution is applied to the vehicle in closed loop.N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
Fig. 7. Scenario π = 0.9 and π£π₯ = 90 km/h for proposed control strategy: steering wheel angle (SWA) and brake torques ππ.
Fig. 8. DTC and percentage overshoot ππ for Set 1 β varying velocity π£π₯ scenario.
Fig. 9. DTC and percentage overshoot ππ for Set 2 β varying friction coefficient π scenario.
The proposed design is able to deal with constraints even when themaneuver is performed in the nonlinear regime of motion, as Figs. 12and 13 show. The figures depict the gβg diagram and Kamm circle
10
values, respectively, for π£π₯ = 90 km/h and π = 0.9. It can be seenthat the values are within the defined stable envelope and at thesame time, while preserving tracking performance. It can be seen in
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
Fibibtloaosrmtte
8
tracf
ttsaIctccmar
btsbd
Fig. 10. Brake torques for straight-line braking case: π£π₯ = 90 km/h and π = 0.9.
Fig. 11. MPC computation time for different scenarios.
ig. 13 that the Rear Right (RR) tire plot shifts towards the left side.e. towards negative longitudinal force direction as the maneuver iseing performed. This is because the test vehicle used in the simulations a Front Wheel Drive (FWD) car. Even though all the four wheels areraking (as seen in Fig. 7), the drive torque from the engine is beingransferred to the front wheels as a result of which the overall tireongitudinal forces in the front tires are mostly in the positive regionf Kamm circle. But during the maneuver, the rear tires brakes as wellnd the overall tire longitudinal force becomes negative as a resultf which the rear tireβs longitudinal force are mostly in the negativeide of Kamm circle. Since the RR tire brakes the most among the twoear wheels, the longitudinal force shift towards the negative half isore as compared to other wheels. Nevertheless, it can be seen that
he lateral force ratio is very high and close to the limits, suggestinghat the controller is able to control the nonlinearities of the maneuverffectively at all times.
. Conclusion and future work
The goal of this work is to design an integrated nonlinear MPC con-roller to provide effective vehicle control in both linear and nonlinearegime of motion and to reduce the high number of accidents caused inrear-end collision scenario. In this scenario, it is important to show theontrollerβs ability to guarantee vehicle stability and passenger safetyor various conditions. Also in this scenario, it is of foremost importance
s11
Fig. 12. gβg diagram for case: π£π₯ = 90 km/h and π = 0.9.
Fig. 13. Kamm circle of each tire for case: π£π₯ = 90 km/h and π = 0.9.
o take into account that a vehicleβs motion is always coupled in bothhe lateral and longitudinal direction. Conventional hierarchical controltrategies implemented on vehicles, often consider the design of lateralnd longitudinal control separately, making the simplified assumptions.n contrast to the classical approaches, we proposed an integratedontrol strategy based on nonlinear model predictive control (NMPC)aking into account the coupling between lateral and longitudinalontrol by design. It was shown how this strategy is able to effectivelyontrol the vehicle when the maneuver is in the nonlinear range ofotion and in the presence of lateral wind. The controller did not show
ny oscillatory behavior and overshoots while tracking the desiredeference trajectory for the various conditions tested.
We compared the proposed design with two benchmark controllersased on model predictive control. Compared to the proposed MPC con-roller design, the other two baseline MPC control approaches rely onimplifying assumptions on the prediction model (linear and nonlinearicycle model) and definition of the constraints. Our integrated NMPCesign outperformed the other two control strategies in all consideredcenarios. Furthermore, the designed strategy showed robustness to
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
πΎ
π
π
πΆ
external disturbances and parameter uncertainty, while being real-timefeasible.
The recommendations for future work are (i) to include the tireslip dynamics inside the prediction model to control the phenomenonof wheel locking and the ABS activation while braking; (ii) bettertrajectory generation methods can be used to make the reference valuesmore realistic and practical to follow; (iii) incorporation of tire modelinside the controllerβs prediction model can make the MPC more robustto parameter variation and improve the dynamic capabilities of thecontroller to capture vehicle dynamics.
9. Nomenclature according to ISO 8855:2011
πΌij Tire slip angle, [rad]π½ Vehicle sideslip angle, [rad]οΏ½οΏ½ Sideslip angle gradient, [rad/s]π Chassis yaw/heading angle, [rad]
πref Reference yaw angle, [rad]οΏ½οΏ½ref Reference yaw rate, [rad/s]πΏsw Steering-wheel angle, [rad]πΏij Road wheel steer angle, [rad]π ij Longitudinal slip, [β]οΏ½οΏ½ij Wheel angular acceleration, [rad/s2]ππ₯ Longitudinal acceleration, [m/s2]ππ¦ Lateral acceleration, [m/s2]πΆπΌij Tire cornering stiffness, [N/rad]πΆπΌπ Front axle cornering stiffness, [N/rad]πΆπΌπ Rear axle cornering stiffness, [N/rad]πΆπ ij Longitudinal slip stiffness, [N/[β]]ππΏ Road wheel steer rate, [rad/s]
πππijBrake torque rate per wheel, [N m/s]
πref Target lateral displacement by SV, [m]πΉπ₯ij Tire longitudinal force, [N]πΉπ₯π Front axle longitudinal force, [N]πΉπ₯π Rear axle longitudinal force, [N]πΉπ¦ij Tire lateral force, [N]πΉπ§ij Tire normal force, [N]π Acceleration due to gravity, [m/s2]βrf Front roll center height, [m]βrr Rear roll center height, [m]πΌπ§π§ Vehicle inertia around z-axis, [kg m2]π½wij Wheel moment of inertia, [kg m2]
π,π Front roll stiffness, [N m/rad]πΎπ,π Rear roll stiffness, [N m/rad]ππ Distance from front axle to CoG, [m]ππ Distance from rear axle to CoG, [m]πΏ Wheelbase, [m]
πΏref Distance to LV, [m]π Total vehicle mass, [kg]ππ Control horizon, [β]ππ Prediction horizon, [β]οΏ½οΏ½ , π Yaw rate, [rad/s]πeff Effective rolling radius, [m]π st Steering ratio, [β]ππij Traction torque, [N m]
πijactApplied brake torque to wheel, [N m]
πijcalCalculated brake torque before actuator dynamics, [N m]
π‘π Front track, [m]
π‘π Rear track, [m]12
π‘π MPC controller sampling time, [s]ππ₯ij Wheel longitudinal velocity, [m/s]π£π₯ Chassis longitudinal velocity, [m/s]π£π¦ Chassis lateral velocity, [m/s]π₯π Vehicle global position in longitudinal direction, [m]π¦π Vehicle global position in lateral direction, [m]π¦ref Reference lateral position, [m]
Declaration of competing interest
The authors declare that they have no known competing finan-cial interests or personal relationships that could have appeared toinfluence the work reported in this paper.
Acknowledgment
L. Ferranti is supported by the Dutch Science Foundation NWO-TTWwithin the SafeVRU project (nr. 14667).
Appendix A. Reference signal definition
The main quantities associated with the definition of the referencesignals presented in Section 2.1 are defined below.ππ¦refππ₯
= ππ΅πβπ(π₯βπ)
(1 + πβπ(π₯βπ))2, (A.1a)
π 1 =
(
π2π¦refππ₯2
)
(
1 +(
ππ¦refππ₯
)2)
32
(A.1b)
π =βπ2 +
β
π22 β 4π1π32π1
, (A.1c)
π =πΆ1π
(A.1d)
and
π1 =(π΅π₯1)2
16β
(π΅πΆ2)2
16(A.2a)
π2 = βπ΅2π₯1πΆ1
8βπ΅π¦1π₯1
2+π΅2π₯14
(A.2b)
π3 =(π΅πΆ1)2
16+ π¦21 +
π΅2
4+π΅π¦1πΆ1
2β π΅π¦1 β
π΅2πΆ14
β πΆ22 (A.2c)
1 = log(
π΅π¦tol
β 1)
(A.2d)
π΅ refers to lateral displacement to be achieved by the subject vehicle,π is the slope of the Sigmoid curve, (π₯1, π¦1) are the coordinates of theobstacle vehicleβs rear-left corner, π¦tol is the initial lateral displacementof the subject vehicle at the beginning of the maneuver, πΆ2 is the pre-defined minimum length which is a tuning parameter and π 1 is thetrajectory curvature.
Appendix B. Prediction model equations
B.1. Planar vehicle model
The 15 equations representing the planar vehicle NMPC model isshown in Eqs. (B.1)β(B.11).
οΏ½οΏ½π₯ =(πΉπ₯fl + πΉπ₯fr ) cos (πΏ) β (πΉπ¦fl + πΉπ¦fr ) sin (πΏ) + (πΉπ₯rl + πΉπ₯rr )
π+π£π¦π
(B.1)
οΏ½οΏ½π¦ =(πΉπ₯fl + πΉπ₯fr ) sin (πΏ) + (πΉπ¦fl + πΉπ¦fr ) cos (πΏ) + (πΉπ¦rl + πΉπ¦rr )
π (B.2)
βπ£π₯πN. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
π₯
πΉ
πΉ
B
yr
π½
Std
B
(cp
π
πΉ
π
π
T
οΏ½οΏ½ =[
(πΉπ₯fl + πΉπ₯fr ) sin (πΏ)ππ + (πΉπ¦fl + πΉπ¦fr ) cos (πΏ)ππβ(πΉπ¦rl + πΉπ¦rr )ππ + 0.5π‘π (πΉπ₯fr β πΉπ₯fl ) cos (πΏ) + 0.5π‘π (πΉπ¦fl
βπΉπ¦fr ) sin (πΏ) + 0.5π‘π(πΉπ₯rr β πΉπ₯rl )]
βπΌπ§π§
(B.3)
οΏ½οΏ½ = π (B.4)
π = π£π₯ cos (π) β π£π¦ sin (π) (B.5)
οΏ½οΏ½π = π£π₯ sin (π) + π£π¦ cos (π) (B.6)
οΏ½οΏ½ = ππΏ (B.7)
οΏ½οΏ½πijact=ππijcal
β ππijact
0.12, ij = (fl, fr) (B.8)
οΏ½οΏ½πijact=ππijcal
β ππijact
0.05, ij = (rl, rr) (B.9)
οΏ½οΏ½πijcal= πππij
, ij = (fl, fr) (B.10)
οΏ½οΏ½πijcal= πππij
, ij = (rl, rr) (B.11)
B.2. Lateral forces
The equations used for the calculation of lateral forces in the above-mentioned prediction model are mention in Eqs. (B.12a)β(B.12d).
πΉπ¦fl =πΆnonπΌfl
(
πΏ β (π£π¦ + ππ π))
π£π₯ β 0.5π‘π π(B.12a)
π¦fr =πΆnonπΌfr
(
πΏ β (π£π¦ + ππ π))
π£π₯ + 0.5π‘π π(B.12b)
π¦rl =πΆnonπΌrl
(
β(π£π¦ β πππ))
π£π₯ β 0.5π‘ππ(B.12c)
πΉπ¦rr =πΆnonπΌrr
(
β(π£π¦ β πππ))
π£π₯ + 0.5π‘ππ(B.12d)
.3. Longitudinal forces
The mathematical equation that describes the wheel dynamics in-axis in the single corner model described below (note that the rollingesistance moment has been neglected here):
wij οΏ½οΏ½ij = ππij β ππijactβ πΉπ₯ij πeffij , ij = (fl, fr, rl, rr) (B.13)
Neglecting wheel dynamics, the term(
π½wij οΏ½οΏ½ij
)
from the LHS of Eq.(B.13) was dropped. After rearranging the equation in terms of lon-gitudinal force πΉπ₯, the final equation used to approximate the tirelongitudinal force is given by:
πΉπ₯ij =ππij β ππijact
πeffij, ij = (fl, fr, rl, rr) (B.14)
ubstituting (B.14) and (B.12a)β(B.12d) in the respective tire forceerms of (B.1)β(B.3) gives the final formulation for the 15 state pre-iction model.
.4. Dugoff tire model
The equations to model the Dugoff tire model are shown in Eqs.B.15)β(B.25). An example of 2D force-slip characteristics and Kammircle using Dugoff tire model for normal load πΉπ§ = 5000 N has beenlotted in Fig. B.14.
= ππ
(
1 β ππππ₯β
π ij2 + tan2 πΌij
)
(B.15)
ij13
π =ππΉπ§π‘ij
(
1 β π ij)
2β
(
πΆπ ijπ ij
)2+(
πΆπΌij tan(
πΌij)
)2(B.16)
π (π) ={
π(2 β π), π < 11, π β₯ 1
(B.17)
π₯ij =πΆπ ijπ ij
1 β π ijπ (π) (B.18)
πΉπ¦ij =πΆπΌij tan
(
πΌij)
1 β π ijπ (π) (B.19)
ππ and ππ are tuning parameters. ππ₯ij is the wheelβs longitudinal velocityand was calculated using the equations below:
ππ₯fl =(
π£π¦ + ππ π)
sin (πΏ) +(
π£π₯ β 0.5π‘π π)
cos (πΏ) (B.20)
ππ₯fr =(
π£π¦ + ππ π)
sin (πΏ) +(
π£π₯ + 0.5π‘π π)
cos (πΏ) (B.21)
π₯rl = π£π₯ β 0.5π‘ππ (B.22)
π₯rr = π£π₯ + 0.5π‘ππ (B.23)
he tire slip angle πΌij is calculated as follows:
πΌππ = βtanβ1((
π£π¦ + ππ π)
cos πΏππ β(
π£π₯ β 0.5π‘π π)
sin πΏππ(
π£π¦ + ππ π)
sin πΏππ +(
π£π₯ β 0.5π‘π π)
cos πΏππ
)
(B.24a)
πΌππ = βtanβ1((
π£π¦ + ππ π)
cos πΏππ β(
π£π₯ + 0.5π‘π π)
sin πΏππ(
π£π¦ + ππ π)
sin πΏππ +(
π£π₯ + 0.5π‘π π)
cos πΏππ
)
(B.24b)
πΌππ = βtanβ1( π£π¦ β ππππ£π₯ β 0.5π‘ππ
)
(B.24c)
πΌππ = βtanβ1( π£π¦ β ππππ£π₯ + 0.5π‘ππ
)
(B.24d)
The wheel slip π ij for each wheel is given by:
π ij =|
|
|
πijπeff β ππ₯ij|
|
|
max{πijπeff, ππ₯ij}. (B.25)
The Dugoff model based tire longitudinal and lateral force calcu-lation is used in the nonlinear bicycle model based MPC design. Bycapturing the tire nonlinearity during the maneuver, the overshoot(as seen in the case of linear bicycle model based MPC) was reducedsubstantially, thereby improving the overall performance.
Appendix C. Linear bicycle model details
οΏ½οΏ½π₯ = π£π¦π (C.1a)
οΏ½οΏ½π¦ = βπΆπΌπ + πΆπΌπππ£π₯
π£π¦ +πππΆπΌπ β πππΆπΌπ
ππ£π₯π β π£π₯π
+πΆπΌππ
πΏ (C.1b)
οΏ½οΏ½ =πππΆπΌπ β πππΆπΌπ
πΌπ§π§π£π₯π£π¦ β
π2ππΆπΌπ + π2ππΆπΌπ
πΌπ§π§π£π₯π
+πππΆπΌππΌπ§π§
πΏ (C.1c)
οΏ½οΏ½ = π (C.1d)
οΏ½οΏ½π = π£π₯ cos (π) β π£π¦ sin (π) (C.1e)
οΏ½οΏ½π = π£π₯ sin (π) + π£π¦ cos (π) (C.1f)
οΏ½οΏ½ = ππΏ (C.1g)
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
π₯
R
A
A
A
B
B
B
BC
C
C
D
E
E
F
F
F
F
G
H
H
I
J
K
L
L
LM
M
N
N
Fig. B.14. Tireβs 2D force-slip characteristics and Kamm circle using Dugoff tire model.
Appendix D. Nonlinear bicycle model details
οΏ½οΏ½π₯ =πΉπ₯π cos (πΏ) β πΆ
nonπΌπ
(
πΏ β π£π¦+ππ ππ£π₯
)
sin (πΏ) + πΉπ₯ππ
+ π£π¦π (D.1a)
οΏ½οΏ½π¦ =πΉπ₯π sin (πΏ) + πΆ
nonπΌπ
(
πΏ β π£π¦+ππ ππ£π₯
)
cos (πΏ)
π
βπΆnonπΌπ
( π£π¦βππππ£π₯
)
πβ π£π₯π (D.1b)
οΏ½οΏ½ =
(
πΉπ₯π sin (πΏ) + πΆnonπΌπ
(
πΏ β π£π¦+ππ ππ£π₯
)
cos (πΏ))
πππΌπ§π§
+πΆnonπΌπ
( π£π¦βππππ£π₯
)
πππΌπ§π§
(D.1c)
οΏ½οΏ½ = π (D.1d)
π = π£π₯ cos (π) β π£π¦ sin (π) (D.1e)
οΏ½οΏ½π = π£π₯ sin (π) + π£π¦ cos (π) (D.1f)
οΏ½οΏ½ = ππΏ (D.1g)
eferences
ckermann, C., Bechtloff, J., & Isermann, R. (2015). Collision avoidance with combinedbraking and steering. In P. Pfeffer (Ed.), Springer fachmedien wiesbaden 2015, 6thinternational munich chassis symposium 2015 (pp. 199β213). http://dx.doi.org/10.1007/978-3-658-09711-0_16.
dams, L. D. (1994). Review of the literature on obstacle avoidance maneuvers: Brakingversus steering: Tech. rep. UMTRI-94-19,, Ann Arbor, Michigan 48 109-2 150, U.S.A:The University of Michigan, Transportation Research Institute.
ripina, M., Sam, Y., Kumeresan, A. D., Ismail, M., & Kemao, P. (2014). A reviewon integrated active steering and braking control for vehicle yaw stability system.Journal Technology, 71(2), 105β111.
14
Attia, R., Orjuela, R., & Basset, M. (2012). Longitudinal control for automated vehicleguidance. The International Federation of Automatic Control.
arbarisi, O., Palmieri, G., Scala, S., & Glielmo, L. (2009). LTV-MPC for yaw rate controland side slip control with dynamically constrained differential braking. EuropeanJournal of Control, 3(4), 468β479.
eal, C. E., & Gerdes, J. C. (2009). Enhancing vehicle stability through model predictivecontrol. In ASME 2009 dynamic systems and control conference, Vol. 1 (pp. 197β204).http://dx.doi.org/10.1115/DSCC2009-2659.
eal, C. E., & Gerdes, J. C. (2013). Model predictive control for vehicle stabilizationat the limits of handling. IEEE Transactions on Control Systems Technology, 21(4),1258β1269.
reuer, B., & Bill, K. H. (2008). Brake technology handbook. SAE International.hoi, M., & Choi, S. B. (2016). MPC for vehicle lateral stability via differential
braking and active front steering considering practical aspects. Journal of AutomobileEngineering, 230(4), 459β469. http://dx.doi.org/10.1177/0954407015586895.
hoi, C., Kang, Y., & Lee, S. (2012). Emergency collision avoidance maneuver basedon nonlinear model predictive control. In: IEEE international conference on vehicularelectronics and safety, Turkey (pp. 393β398).
hoi, J., Kim, K., & Yi, K. (2011). Emergency driving support algorithm with steeringtorque overlay and differential braking. In 2011 14th international IEEE conferenceon intelligent transportation systems (pp. 1433β1439). Washington, DC, USA: doi:978-1-4577-2197-7/11/.
amian, M., Shyrokau, B., Ocariz, A., & Akutain, X. C. (2019). Torque control for morerealistic hand-wheel haptics in a driving simulator. In Driving simulation conferenceeurope.
uropean Council Service Framework Programme, 0000. Electric-VEhicle control ofindividual wheel torque for on- and off-road conditions (E-VECTOORC), europeanfunded project, FP7 (2007-2013).
uropean Road Safety Observatory (2018). Annual accident report 2018: Tech. rep.,European Commission.
alcone, P., Borrelli, F., Asgari, J., Tseng, H. E., & Hrovat, D. (2007). Predictiveactive steering control for autonomous vehicle systems. IEEE Transactions on ControlSystems Technology, 15(3), 566β580.
alcone, P., Tseng, H. E., Borrelli, F., Asgari, J., & Hrovat, D. (2008). MPC-basedyaw and lateral stabilisation via active front steering and braking. Vehicle SystemDynamics, 46(S1), 611β628. http://dx.doi.org/10.1080/00423110802018297.
erranti, 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 roadusers. In: Proceedings of the intelligent vehicles symposium.
erreau, H., Kirches, C., Potschka, A., Bock, H., & Diehl, M. (2014). Qpoases: A para-metric active-set algorithm for quadratic programming. Mathematical ProgrammingComputation, 6(4), 327β363.
ilreath & Associates (2013). Rear-end collisions: The most common type of autoaccident in the U.S.. https://tinyurl.com/rear-end-collisions.
ac, A., & Bodie, M. O. (2002). Improvements in vehicle handling through integratedcontrol of chassis systems. International Journal of Vehicle Autonomous Systems, 1(1),83β110.
e, J., Crolla, D. A., Levesley, M. C., & Manning, W. J. (2006). Coordination ofactive steering, driveline, and braking for integrated vehicle dynamics control.Journal of Automobile Engineering, 220, 1401β1421. http://dx.doi.org/10.1243/09544070JAUTO265.
nsurance Information Institute, (0000) Facts + statistics: Highway safety, https://www.iii.org/fact-statistic/facts-statistics-highway-safety.
alali, M., Khosravani, S., Khajepour, A., Chen, S., & Litkouhi, B. (2017). Modelpredictive control of vehicle stability using coordinated active steering and dif-ferential brakes. IFAC Mechatronics: The Science of Intelligent Machines, 48, 30β41.http://dx.doi.org/10.1016/j.mechatronics.2017.10.003.
eviczky, T., Falcone, P., Asgari, F. B. J., & Hrovat, D. (2006). Predictive controlapproach to autonomous vehicle steering. In 2006 American control conferenceMinneapolis. Minnesota, USA: IEEE, http://dx.doi.org/10.1109/ACC.2006.1657458.
ee, S. E., Llaneras, E., Klauer, S., & Sudweeks, J. (2007). Analyses of rear-endcrashes and near-crashes in the 100-car naturalistic driving study to support rear-signaling countermeasure development. USDOT/National Highway Traffic SafetyAdministration.
im, E. H. M., & Hedrick, J. K. (1999). Lateral and longitudinal vehicle control couplingfor automated vehicle operation. In Proceedings of the American control conference,San Diego, California (pp. 3676β3680).
impert, R. (1999). Brake design and safety. Society of Automotive Engineers.arkkula, G., Benderius, O., Wolff, K., & Wahde, M. (2012). A review of near-collision
driver behavior models. Human Factors, 54(6), 1117β1143. http://dx.doi.org/10.1177/0018720812448474.
okhiamar, O., & Abe, M. (2004). Simultaneous optimal distribution of lateral andlongitudinal tire forces for the model following control. Journal of Dynamic Systems,Measurement, and Control by ASME, 126, 753β763. http://dx.doi.org/10.1115/1.1850533.
ational Highway Traffic Safety Administration (2016). Table 29 - crashes by firstharmful event, manner of collision, and crash severity. https://cdan.nhtsa.gov/tsftables/tsfar.htm#.
ational Highway Traffic Safety Administration (2017). 2016 fatal motor vehicle crashes:Overview: Tech. rep., U.S. Department of Transportation, DOT HS 812 456.
N. Chowdhri, L. Ferranti, F.S. Iribarren et al. Control Engineering Practice 106 (2021) 104654
S
S
S
S
V
Quirynen, R., Vukov, M., Zanon, M., & Diehl, M. (2014). Autogenerating microsecondsolvers for nonlinear MPC: a tutorial using ACADO integrators. Optimal ControlApplications & Methods.
hah, J. (2015). Development and control of evasive steer assist using rear wheel steering:SAE technical paper 2015-26-0004, http://dx.doi.org/10.4271/2015-26-0004.
hyrokau, B., Wang, D., Augsburg, K., & Ivanov, V. (2013). Vehicle dynamics with brakehysteresis. Proceedings of the Institution of Mechanical Engineers, Part D: Journal ofautomobile engineering, 227(2), 139β150.
hyrokau, B., Wang, D., Savitski, D., Hoepping, K., & Ivanov, V. (2015). Vehiclemotion control with subsystem prioritization. Mechatronics, 30, 297β315. http://dx.doi.org/10.1016/j.mechatronics.2014.11.004.
oudbakhsh, D., & Eskandarian, A. (2010). A collision avoidance steering controller usinglinear quadratic regulator: SAE international 2010-01-0459.
ukov, M., Domahidi, A., Ferreau, H. J., Morari, M., & Diehl, M. (2013). Auto-generatedalgorithms for nonlinear model predicitive control on long and on short horizons.In: Proceedings of the 52nd conference on decision and control (CDC).
15
Wang, X., Zhu, M., Chen, M., & Tremont, P. (2016). Driversβ rear end collisionavoidance behaviors under different levels of situational urgency. TransportationResearch Part C (Emerging Technologies), (71), 419β433.
Yi, B., Gottschling, S., Ferdinand, J., Simm, N., Bonarens, F., & Stiller, C. (2016). Realtime integrated vehicle dynamics control and trajectory planning with MPC forcritical maneuver. In: IEEE intelligent vehicles symposium (IV) Gothenburg, Sweden.
Zegelaar, P. (2017). For evasive steering assist βsteering you out of troubleβ. In 11thinternational IQPC conference steering systems Berlin.
Zhou, J., Lu, J., & Peng, H. (2010). Vehicle stabilisation in response to exogenous im-pulsive disturbances to the vehicle body. International Journal of Vehicle AutonomousSystems, 8(2/3/4), 242β262. http://dx.doi.org/10.1504/IJVAS.2010.035798.
Zhu, L., Shyrokau, B., Boulkroune, B., van Aalst, S., & Happee, R. (2018). Performancebenchmark of state-of-the-art lateral path-following controllers. In 15th internationalworkshop on advanced motion control, Tokyo, Japan.