intelligent vehicle automatic stop-and-go task based on

Research ArticleIntelligent Vehicle Automatic Stop-and-Go Task Based onHumanized Learning Control Model

Tianjun Sun12 Zhenhai Gao12 Fei Gao2 Tianyao Zhang2 Di Ji3 and Siyan Chen 12

1College of Automotive Engineering Jilin University Changchun 130012 China2State Key Laboratory of Automotive Simulation and Control Jilin University Changchun 130012 China3College of Communication Engineering Jilin University Changchun 130012 China

Correspondence should be addressed to Siyan Chen chensiyanjlueducn

Received 29 September 2020 Revised 12 November 2020 Accepted 6 January 2021 Published 18 January 2021

Academic Editor Songtao Lv

Copyright copy 2021 Tianjun Sun et al (is is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

(e automatic stop-and-go task of intelligent vehicles can make the adaptive cruise control system achieve a full-speed rangeHowever the conventional designmethodsmostly focus on functional safety without considering driversrsquo behaviors thereby leadingto a poor driving experience To improve the situation a humanized learning control model is used instead of mechanical switchinglogic (erefore first the common characteristics of human drivers with different driving styles are found by analyzing real driversrsquoexperiments (en the vehicle automatic starting function is designed based on iterative learning control with the fast Fouriertransform for acceleration fitting Next the vehicle automatic braking function is designed based on dynamic time to collisionFinally the simulation of the stop-and-go scenario is shown in CARSIM and the real vehicle test is performed under the urbanoverpass driving condition Results show that the proposed model can improve the humanization in the vehicle stop-and-go task

1 Introduction

With the high pace of modern life the increasing number ofcitizens will cause traffic congestion Although conventionaladaptive cruise control (ACC) systems can provide partialdrive assistance such systems could still be improved Onthe one hand most ACC systems may not cover the full-speed range [1] On the other hand the current automaticstop-and-go control methods of vehicles do not considerdriversrsquo behavior [2]

Research on stop-and-go tasks is presently one of themost important topics in the field of ACC systems One ofthe reasons is that the conventional ACC will frequentlywithdraw from the system when the velocity is less than30 kmh Moreover the velocity is very low to enablecruise control (CC) function [3] Furthermore the currentcontrol methods mostly focus on functional safetyHowever the poor driving experience will be increasinglyevident when driversrsquo behaviors are not considered Forinstance some drivers used to have a violent start such asa sense of push the back but the control logic may limit

the acceleration by considering safety and comfort (eother drivers used to make a double brake when a vehiclein front suddenly slows down but the control logic mayprovide a great deceleration such that the host vehicle canstop by only braking once (erefore when the driver isaware of the relative motion of the vehicle in the field ofview the driver will make a series of habitual actions tocatch up with the target vehicle and keep a safe distance(us these actions reflect the real driversrsquo driving be-haviors in car following but the existing decision-makingalgorithms or control methods are compliant with theestablished rules

(e challenges of the research comprehensively regarddriversrsquo behavior characteristics vehiclersquos dynamic charac-teristics safety and humanization in vehicle automaticcontrol Moreover in this study an automatic stop-and-gocontrol method is designed by analyzing the real driversrsquostarting and braking behaviors based on an intelligent ve-hicle platform as shown in Figure 1

According to previous studies the conventional controlmethods for vehicle stop-and-go scenario can be

HindawiAdvances in Civil EngineeringVolume 2021 Article ID 8867091 11 pageshttpsdoiorg10115520218867091

summarized as model-based control For example Junchen[4] established a car-following model based on driversrsquobehavior and dynamic vehicle characteristics Howevermodel-based control methods are usually regarded ascomplicated and hard-to-debug systems In the abovemethods while also considering safety the calculation oftime to collision (TTC) plays an important role in theprocess of braking As a quintessential example Jahandidehet al [5] found that pedestrians prefer to cross the road at theintersection point at the average TTC of approximately 62 sand 54 of men and 39 of women face TTC risk for lessthan 3 s Zhenhai andWu [6 7] established models based onTTC for warning and braking time prediction on the basis ofdriversrsquo habitual braking and braking force characteristicsHowever the conventional TTC algorithm cannot meet thechange of acceleration Compared with the conventionalmethods some new trends toward learning control exist Forexample Lu et al [8] designed a learning module based onneural reinforcement learning before the PID control (ismodel can learn from human drivers online based on the on-board sensing information and realize human-like longi-tudinal speed control by learning from the demonstrationparadigm Xing et al [9] provided a personalized leadingvehicle trajectory prediction method based on joint timeseries modeling (is method can generate three differentdriving styles for the leading vehicle through the JTSMalgorithm and achieve better results Lv et al [10 11]proposed a CPS-based framework for codesign optimizationof an automated electric vehicle with different driving styleswhich used unsupervised machine learning to develop thedriving style recognition algorithm

By analyzing the above results this study finds thatconventional methods are applicable but humanization isnot considered due to its inherent logic strategies(erefore based on previous studies we propose a hu-manized vehicle automatic stop-and-go control methodbased on the iterative learning (IL) algorithm and thedynamic time to collision (DTTC) model (is study aimsto develop a humanized learning control model based on ILand DTTC while focusing on the stop-and-go task forautonomous driving (e main contributions of this studyare as follows (1) the driversrsquo control behavior

characteristics are extracted through the real vehicle testswith different types of drivers (2) An automatic drivecontrol method based on the IL algorithm is proposed tolearn the driver behavior and realize the human-likecontrol Based on the IL algorithm complex dynamiccalculations are not required and the control method canbe learned directly from the acceleration curve equation(3) An automatic brake control method based on the DTTCmodel is proposed in this study to learn the driver behaviorand realize the human-like control Based on the DTTCmodel on the one hand the safety of the braking process isimproved by the consideration of dynamic deceleration onthe other hand different types of braking can be learneddirectly from the real vehicle tests (4) Automatic stop-and-go control is implemented by combining the automaticdrive control and automatic brake control methods (evalidity and feasibility of the proposed method are verifiedthrough simulation and real tests

(e remainder of this paper is organized as follows thesecond part is ldquoextraction of driversrsquo control behaviorcharacteristicsrdquo (e third part is the ldquolearning model forvehicle automatic starting controlrdquo (e fourth part is theldquolearning model for vehicle automatic braking controlrdquo (efifth part is ldquosimulation and experimental testsrdquo (e sixthpart is ldquoresults and comparative analysisrdquo (e final part isldquoconclusions and future workrdquo

2 Extraction of Driversrsquo ControlBehavior Characteristics

(e three types of typical drivers are examined based on theprevious classification to obtain the common characteristicsof different drivers with evident driving styles [12ndash14] Inthis study we choose three types namely aggressive drivernormal driver and steady driver Many factors causingdifferent characteristics should be neglected because thispart aims to find the humanized common characteristics ofhuman drivers compared with the conventional model-based design (erefore we assume that the target vehiclestops at the stop line 200m away from the host vehicle(enthe host vehicle starts to catch up with the target vehiclethrough acceleration or deceleration control over a period of

Driver Horizon Judgment Acceleratorbrake pedal

Vehicle Environmental perception Learning Vehicle dynamics models

Engine characteristic

Braking system

A human-like platform for intelligent vehicle

Figure 1 A human-like platform for intelligence vehicle

2 Advances in Civil Engineering

time Finally the host vehicle will stop behind the targetvehicle Figure 2 shows the experimental scenario

As shown in Figure 3 we conduct further analysis ofvehicle velocity and acceleration in the stop-and-go scenarioIn the case of low speed the vehicle stop-and-go charac-teristics controlled by different drivers have some similar-ities One similarity is that the trend of the acceleration curvewill increase along with the velocity and then graduallyreduce to half of the peak value From a practical point ofview the drivers in the experiment with different startcharacteristics usually choose to step down the acceleratorpedal (en they slowly release the accelerator pedal in theprocess of starting (e other similarity is the time ofbraking Although different drivers have varying brakingdecelerations a humanized braking method is reflected in areasonable double brake Moreover the trend of accelerationchange is highly similar to the same velocity adjustment andthe trend of deceleration with a large increase or decreasecan be classified as a double brake

3 Learning Model for Vehicle AutomaticStarting Control

When the vehicle deals with repetitive tasks in an actualenvironment people expect to use a certain amount ofhuman manipulation samples to match the automaticcontrol system with the behavior of the drivers To imitatehuman learning abilities and their self-regulation functionthe IL algorithm can improve the control target throughiterative correction which is suitable for controlled objectswith repetitive motion properties (e IL algorithm canrealize the control of strong coupled nonlinear dynamicsystems with high uncertainty in a limited time Moreoverthis algorithm does not depend on precise mathematicalmodels [15 16] (erefore IL has been widely used inautomation controls such as industrial robots digitalcontrol machines and machine manufacturing

However in the practical environment the driving stylesof drivers are better described through behaviors but hard todescribe through theoretical analysis (erefore in hu-manized featuresrsquo extraction we will transform the practical

problem into a function approximation problem based onexperimental data (us the first step is to design a functionto make it as close as possible to the trend of the accelerationcurve

According to previous studies two methods are usuallyused for solving function approximation problems namelyinterpolation and fitting If the experimental data are dis-crete then the interpolation is suitable When consideringthe continuity of the acceleration curve the fitting is suitableinstead (e three main methods of fitting that are usedwidely are algebraic polynomials rational fractional func-tions and triangular polynomials Typical samples for thefitting methods above are polynomial fitting Gaussian fit-ting and fast Fourier transform (FFT) Moreover weconstruct a function similar to the trend of the accelerationcurve in the experiment which will increase along with thevelocity and then gradually reduce to half of the peak valueIf we choose the polynomial power function for fitting thenestimating the error is difficult although easy to operate andthe amount of calculation is small When the samplingpoints are rare the fitting effect is poor as shown inFigure 4(a) Otherwise if we choose the Gaussian functionfor fitting then the positioning accuracy for edge detectionof data will be low and the fitting will be sensitive to noisealthough calculating the integral is convenient as shown inFigure 4(b)

Figure 4 shows that underfitting occurs in polynomialfitting and overfitting occurs in Gaussian fitting From theresults above these two methods for acceleration curvefitting can meet the approximation standard but each hasshortcomings (erefore FFT fitting is made by analyzingpolynomial and Gaussian fittingmethods to better reflect thehumanization of acceleration in the vehicle stop-and-goscenario FFT is not a change of discrete Fourier transform(DFT) but is a fast algorithm which can reduce the numberof DFT operations Among them any periodic function canbe converted into the sum of several trigonometric functionsby dividing the periodic function to be converted f(t) into atrigonometric function of innumerable sin and cos as shownin the following equation

f(t) 1T

1113946t0+T

t0

f(t)dt +2T

1113946t0+T

t0

f(t)sin(ωt)dt sin(ωt) +2T

1113946t0+T

t0

f(t)cos(ωt)dt cos(ωt)1113890 1113891+

+2T

1113946t0+T

t0

f(t)sin(nωt)dt sin(nωt) +2T

1113946t0+T

t0

f(t)cos(nωt)dt cos(nωt)1113890 1113891+

(1)

Furthermore we use the toolbox ofMATLAB to conductthe acceleration fitting based on FFT to the experimentaldata for vehicle starting and Figure 5 shows the result

(e FFT fitting curve on level 4 can effectively express thefluctuation trend of the experimental data through comparativeanalysis If we regard the FFT fitting curve on level 4 as the targetacceleration that the system needs to follow then the second stepwould be the vehicle automatic starting based on the IL algorithm

Owing to the influence of several factors such as externalair resistance road friction resistance and internal trans-mission the longitudinal acceleration of the vehicle and thetorque of the engine have a nonlinear relationship Fur-thermore the control principle of the automatic vehicle startcontrol algorithm is based on the engine torque control totrack the target acceleration Hence to express the nonlinearrelationship this study uses the PD-type close-loop IL

Advances in Civil Engineering 3

control method to convert the engine torque controlproblem into an engine throttle opening problem (e ILalgorithm can be described as shown in the followingequation

pk(t) pkminus1(t) + Lpek(t) + Ldek(t)

ek(t) aexpected(t) minus areal(t)

⎧⎨

⎩ (2)

where pk(t) represents the throttle percentage aexpecred(t)represents the expected acceleration areal(t) represents thereal acceleration Lp and Ld represent the learning gainfactors and ek(t) represents the error

As mentioned before in this study a typical fittingfunction is extracted from the experimental test data to be

used as the target acceleration for tracking the target of ILcontrol as shown in Figures 6 and 7

In Figure 6 the error gradually decreases with the in-crease of the iteration number and when the iterationreaches the eighth time the error is close to 0 (erefore inFigure 7 the acceleration following the curve from the firstiteration to the eighth iteration and the throttle change curveis provided Figure 7(a) shows that with the increase ofiteration time the actual acceleration of the vehicle drawscloser to the target acceleration and converges to the ex-pected trajectory In addition Figure 7(b) shows that withthe increase of the number of iterations the control quantitygradually converges to the expected control quantity andconverges to the expected trajectory

Start control and speed up Change Brake control and slow down

Figure 2 (e experimental scenario

Time (s)

Velo

city

(km

h)

An aggressive driverA normal driverA steady driver

40353025201510

50

0 10 20 30 40 50 60 70 80 90 100 110 120 140130 150 160 170 180 190 200

(a)


Acce

lera

tion

(ms

2 )

435

325

2

ndash2

15

ndash15

ndash3ndash25

1

ndash1

05

ndash050

Time (s)0 10 20 30 40 50 60 70 80 90 100 110 120 140130 150 160 170 180 190 200

(b)

Figure 3 (e experimental results for three drivers in the vehicle stop-and-go scenario


Acce

lera

tion

(ms

2 )

Time (s)

Experimental data curvePolynomial fitting curve

Underfitting

30

25

20

15

10

05

00 25 50 75

(a)

Experimental data curveGaussian fitting curve

Overfitting

Acce

lera

tion

(ms

2 )

30

25

20

15

10

05

0

Time (s)0 25 50 75

(b)

Figure 4 Acceleration fitting based on different fitting methods (a) Polynomial fitting (b) Gaussian fitting

Acce

lera

tion

(ms

2 )

Experimental dataLevel 1 FFT fitting Level 2 FFT fitting

Level 3 FFT fittingLevel 4 FFT fittingLevel 5 FFT fitting

Time (s)0 25 50 75

30

25

20

15

10

05

0

Figure 5 Acceleration fitting based on multiple FFT method

e time for iterative learning

Erro

r

0 1 2 3 4 5 6 7 8 9 10

25

2

15

1

05

0

Figure 6 (e error varies with the time for iteration learning


4 Learning Model for Vehicle AutomaticBraking Control

(e vehicle automatic braking control is an active method toapply braking force through the electronic stability programbefore the collision happens Safety should be prioritized toensure that the host vehicle can slow down when the targetvehicle decelerates or suddenly brakes when car followingwhich is different from the starting control According toprevious studies [17ndash19] the cause of the collision is relatedto different driversrsquo braking time braking behavior andbraking force (e design of the conventional vehicle au-tomatic braking control method only considers the rigidbody kinematics characteristics and collision theory(erefore the driver and the passenger will feel an evidentsense of frustration and tension due to the lack of under-standing of the driverrsquos braking characteristics

At present classification judgment is a popular methodwidely used in the vehicle longitudinal automatic brakingcontrol (is method is based on the comprehensive in-formation of vehicle and traffic conditions As the dangercomes the vehicle will slow down through active inter-vention However in this study we aim to consider theuncertainty of the target vehicle and propose an anthro-pomorphic automatic braking control method based on theDTTC model (us through the analysis of Figure 3 thehumanized design problem is transformed into the inter-pretation of brake commonality

As previously mentioned the basic requirement of thevehicle automatic braking control is to achieve driving

safety (e conventional calculation method for TTC isshown in the following equation

TTC Drel

Vrel (3)

where Drel and Vrel represent the relative distance and ve-locity between the vehicles respectively However thisdefinition of TTC does not consider the speed change of theego vehicle and target vehicle during acceleration and de-celeration (erefore as shown in Figure 8 DTTC is adefinition of a collision which considers the host vehiclersquosvelocity vego and deceleration aego and the target vehiclersquosvelocity vtar and deceleration atar

Based on the kinematic relation of the vehicle the valueof DTTC can be obtained as shown in the followingequation

Dego Vegot +12

aegot2

Dtar Vtart +12

atart2

Dego Drel + Dtar

Vrelt +12

atart2

+ Drel 0

(4)

Moreover through the further discussion on vrel and arelthe DTTC can be described as shown in the followingequation

e 5th iteratione 7th iteratione 8th iteration

FFT fitting functione 1st iteratione 2nd iteratione 3rd iteration

Acce

lera

tion

(ms

2 )

Time (s)0 25 50 75

3

25

2

15

1

05

0

(a)

rottl

e deg

ree (

times100

)

03

025

035

02

05

045

04

015

01

005

035 40 45 50 55 60 65

Time (s)


Expected valuee 1st iteratione 2nd iteratione 3rd iteration

(b)

Figure 7 (e results of the vehicle automatically start based on the IL control method (a) Acceleration following with multiple iterations(b) (rottle change with acceleration following


DTTC

infin Vrel lt 0 and arel lt 0

minusVrel( 1113857 minus

V2rel minus 2arelDrel( 1113857

1113969

arel Vrel lt 0 or arel lt 0

⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(5)

An ideal braking model based on the minimum safedistance should be considered to make the decelerationprocess of vehicle automatic braking control reflect thecharacteristics of the driverrsquos behavior In Figure 8 on theone hand the ideal braking model provides an expected

deceleration in the process of slowing down which can bedefined as areq represents the deceleration avoiding a col-lision On the other hand the host vehicle should maintain aminimum safe distance from the target vehicle after brakingwhich can be defined as Dlowast as shown in Figure 9 Fur-thermore the following equation provides the calculation ofareqwhere Drel represents the current distance Dtar repre-sents the estimated distance and Dego represents the dis-tance the host vehicle traveled

Dego Vegot +12areqt

2

Dego + Dlowast

Drel + Dtar

areq

0 Vrel gt 0 and atar gt 0

V2ego

V2taratar1113872 1113873 minus 2lowast Drel minus D

lowast( 1113857

Vrel gt 0 and atar lt 0 Vrel lt 0 and atar lt 0 and tego gt ttar

atar minusV

2rel

2lowast Drel minus Dlowast

( 1113857 Vrel lt 0 and atar gt 0 Vrel lt 0 and atar lt 0 and tego lt ttar

⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(6)

5 Simulation and Experimental Tests

A simulation test based on CARSIM and a real vehicle testunder the urban overpass driving condition are provided toverify the validity and feasibility of the proposed method Inthis part we first establish a vehicle longitudinal automaticcontrol model in MATLABSIMULINK (is model con-tains car sensor decision-making and execution modulesas shown in Figure 10

(e car module is connected to CARSIM whichprovides some vehicle environmental and road pa-rameters (e sensor module is connected betweenCARSIM and SIMULINK which provides informationfor the host and target vehicles (e decision-makingmodule is designed on MATLABSIMULINK whichprovides the key strategies and important logics for the

vehicle autonomous stop-and-go task (e executionmodule is a computational relationship between vehicledynamics and the braking system (en we conducted asimulation test for the vehicle autonomous stop-and-goscenario through CARSIMSIMULINK as shown inFigures 11 and 12

To further verify the proposed method a real vehicle testis conducted (e experimental vehicle is HAVAL H7equipped with a millimeter wave radar and dSPACEAutoBox(e former is used to obtain the information of thetarget vehicle whereas the latter is used to obtain the in-formation of the host vehicle Moreover we download thecontrol algorithm to dSPACE AutoBox instead of theoriginal controller of the vehicle(erefore a real vehicle testis performed under the urban overpass driving condition asshown in Figures 13 and 14

Distance traveled by the ego vehicleDego

vego aego

Estimated rangeDtar

e current distanceDrel

vtar atar

Figure 8 (e condition of the DTTC process


Slow down with the target vehicle Stop with the target vehicle Start with the target vehicle

Brake Drive

Figure 11 (e vehicle autonomous stop-and-go task in CARSIM

Time for simulation

Car module

160

Velocity

Acceleration

Distance

Relative velocity

Vehicle automaticstop-and-go

control methods

Sensor moduleDecision-making

module

Expectedacceleration

Execute module

Car_signal

Time (sec)Clock

Front_sensor

Vx

Vr

Distance

Axa_des Out3

Figure 10 (e vehicle longitudinal automatic control model based on MATLABSIMULINK

Distance traveled by the ego vehicle

Dego

vego aego

Estimated rangeDtar


vtar atar

Safe distanceDlowast

Figure 9 (e ideal braking model based on DTTC

e host vehiclee target vehicle

5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Velo

city

(km

h)

(a)

Figure 12 Continued


Distance between two vehiclesAn ideal safe distance (Dlowast = 35m)

5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Dist

ance

(m)

(b)

Figure 12 (e simulation results for vehicle automatic stop-and-go control (a) (e simulation results for velocity in vehicle stop-and-goscenario (b) (e simulation results for distance in vehicle stop-and-go scenario

Slow down with the target vehicle

Brake

Stop with the target vehicle Start with the target vehicle

Drive

Figure 13 (e real vehicle automatic stop-and-go task under the urban overpass driving condition

30

25

20

15

10

5

20 4 6 8 10 12 14 16 18 20Time (s)

Vel

ocity

(km

h)

(a)


20 4 6 8 10 12 14 16 18 20Time (s)

14

12

10

8

6

4

2

Dist

ance

(m)

(b)

Figure 14 (e real vehicle automatic stop-and-go task under the urban overpass driving condition (a) (e host vehiclersquos velocity (b) (edistance between two vehicles


6 Results and Comparative Analysis

During the simulation as shown in Figure 12 the drivingcondition is set to realize the vehiclersquos frequent stop-and-gotask(e results show that the host vehicle can catch up withthe target vehicle and keep an ideal distance when carfollowing Moreover as shown in Figure 14 the experi-mental condition is random under urban overpass (eresults show that the host vehicle can achieve an autono-mous stop-and-go control based on the proposed methodwhen the target vehicle is accelerating or deceleratingHowever to further illustrate humanization a comparativeanalysis is conducted as shown in Figure 15

(e trend of acceleration is similar to the trend inFigure 7(a) Moreover the trend of throttle degree is alsosimilar to the trend in Figure 7(b) (erefore on the onehand the sense of starting in the real vehiclersquos experimentaltest will be similar to the real drivers On the other handwith the evident change in double deceleration the behaviorof braking will also be similar to the real drivers Further-more the ideal distance in the simulation is similar to thetest and then the effectiveness of the proposed method isverified on the premise of ensuring safety

7 Conclusions and Future Work

In this article we propose an automatic stop-and-go controlmethod based on a learning model for vehicles First the realdriversrsquo starting and braking behaviors are obtained throughthe real vehiclersquos experimental test (e results show that thecommon characteristics of human drivers with differentdriving styles are their humanized acceleration and decel-eration (en according to the variation trend of acceler-ation the vehicle automatic starting control strategy isdesigned based on FFTand the IL algorithm Next based onthe DTTC model the vehicle automatic braking controlstrategy is designed by further analyzing the common hu-manized characteristics Finally the validity and feasibility

are proved through the simulation and real vehicle testsFurthermore compared with the initial experiment themethod proposed can provide automatic stop-and-gocontrol in car-following and improve the sense of humanityin the vehicle stop-and-go task

However the humanized learning control method hasstill some limitations (1) considering the limited sample sizethe humanization of the algorithm is slightly inadequate (2)A switch logic should be designed between drive and brakecontrols (3) (e proposed method does not consider someextreme conditions such as emergency braking behavior

Future work can increase the sample size of the learningmodel In addition the switch strategy can be improved tosolve the fluctuation problem and some extreme conditionsmay be tested in the simulation

Data Availability

(e data used in the paper are obtained through actualexperiments rather than using the established experimentaldata Among them some data or curves were derived fromprevious research results which have been presented at aconference in 2019 Chinese Automation Congress (CAC)httpsieeexploreieeeorgdocument8996633)

Disclosure

(is study is a continuation of the previous work which hasbeen presented at a conference in 2019 Chinese AutomationCongress (CAC httpsieeexploreieeeorgdocument8996633)

Conflicts of Interest

(e authors declare no potential conflicts of interest withrespect to the research authorship andor publication ofthis article

3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )

Figure 15 (e acceleration and throttle degree for the real vehicle experimental test


Acknowledgments

(is work was supported by National Science Foundation ofChina (Grants 51775236 51675224 and U1564214) Na-tional Key Research and Development Program (Grant2017YFB0102600) and Science and Technology Develop-ment Project of Jilin province (20200501012GX) (e au-thors gratefully acknowledge Zhenhai Gao Fei Gao TianyaoZhang Ji Di and Siyan Chen of Jilin University for theirhelpful discussion

References

[1] Z Li Research on Full Speed Adaptive Cruise Control System ofPire Electric Vehicle Harbin Institute of Technology HarbinChina 2017

[2] L Yishi Z ZheW Liqiang and H Zongqi ldquoDesign of vehiclefull-speed adaptive cruise control with stop and gordquo Ma-chinery Design and Manufacture vol 5 pp 174ndash177 2017

[3] V Paul N Karl and A Bartono Stop and Go Cruise ControlFISITA World Automatic Congress Seoul Korea 2000

[4] J Junchen ldquoImproved analysis of the car-following modelbased on ACC systemrdquo Science Technology and Engineeringvol 11 no 26 pp 6396ndash6400 2011

[5] Z Jahandideh B Mirbaha and A A Rassafi ldquoModelling therisk intensity of crossing Pedestrians in intersections based onselected critical time to collision a case study of Qazvin cityrdquoTransportation Research Board 96th Annual Meeting vol 12017

[6] G Zhenhai ldquoVehicle virtual following collision avoidancedriver braking time modelrdquo Journal of Jilin University (En-gineering Edition) vol 44 no 5 pp 1233ndash1239 2014

[7] T Wu ldquoA Study on Forward Collision Prevention SystemConsidering the Characteristics of the Driverrsquos CollisionAvoidance Behavior Jilin University Changchun China2014

[8] C Lu J Gong C Lv X Chen D Cao and Y Chen ldquoApersonalized behavior learning system for human-like lon-gitudinal speed control of autonomous vehiclesrdquo Sensors(Basel Switzerland) vol 19 no 17 pp 3671ndash3690 2019

[9] Y Xing C Lv and D Cao ldquoPersonalized vehicle trajectoryprediction based on joint time series modeling for connectedvehiclesrdquo IEEE Transactions on Vehicular Technology vol 69no 2 pp 1341ndash1352 2019

[10] Y Xing C Lv H Wang et al ldquoDriver lane change intentioninference for intelligent vehicles framework survey andchallengesrdquo IEEE Transactions on Vehicular Technologyvol 68 no 5 pp 4377ndash4390 2019

[11] C Lv X Hu A Sangiovanni-Vincentelli Y LiC M Martinez and D Cao ldquoDriving-style-based codesignoptimization of an automated electric vehicle a cyber-physicalsystem approachrdquo IEEE Transactions on Industrial Electronicsvol 66 pp 2965ndash2975 2018

[12] N Kuge T Yamamura O Shimoyama and A Liu ldquoA driverbehavior recognition method based on a driver modelframeworkrdquo SAE Technical Paper vol 109 no 6 pp 469ndash4762000

[13] M Quintero J Lopez and A C C Pinilla ldquoDriver behaviorclassification model based on an intelligent driving diagnosissystemrdquo in Proceedings of the 2012 15th International IEEEConference on Intelligent Transportation Systems pp 894ndash899Anchorage Alaska USA September 2012

[14] G Zhenhai S Tianjun and H Lei ldquoA multi-mode controlstrategy for EV based on typical situationrdquo SAE TechnicalPaper vol 1 pp 1ndash8 2017

[15] H-S Ahn Y Chen and K L Moore ldquoIterative learningcontrol brief survey and categorizationrdquo IEEE Transactionson Systems Man and Cybernetics Part C (Applications andReviews) vol 37 no 6 pp 1099ndash1121 2007

[16] R Chi D Wang Z Hou and S Jin ldquoData-driven optimalterminal iterative learning controlrdquo Journal of Process Controlvol 22 no 10 pp 2026ndash2037 2012

[17] M M Minderhoud and P H L Bovy ldquoExtended time-to-collision measures for road traffic safety assessmentrdquo Ac-cident Analysis amp Prevention vol 33 no 1 pp 89ndash97 2001

[18] W Wachenfeld P Junietz R Wenzel and H Winner ldquo(eworst-time-to-collision metric for situation identificationrdquo inProceedings of the Intelligent Vehicles Symposium (IVS)pp 356ndash362 Gotenburg Sweden September 2016

[19] L Zhang H Ding J Shi et al ldquoAn adaptive backsteppingsliding mode controller to improve vehicle maneuverabilityand stability via torque vectoring controlrdquo IEEE Transactionson Vehicular Technology vol 69 no 3 pp 2598ndash2612 2020


summarized as model-based control For example Junchen[4] established a car-following model based on driversrsquobehavior and dynamic vehicle characteristics Howevermodel-based control methods are usually regarded ascomplicated and hard-to-debug systems In the abovemethods while also considering safety the calculation oftime to collision (TTC) plays an important role in theprocess of braking As a quintessential example Jahandidehet al [5] found that pedestrians prefer to cross the road at theintersection point at the average TTC of approximately 62 sand 54 of men and 39 of women face TTC risk for lessthan 3 s Zhenhai andWu [6 7] established models based onTTC for warning and braking time prediction on the basis ofdriversrsquo habitual braking and braking force characteristicsHowever the conventional TTC algorithm cannot meet thechange of acceleration Compared with the conventionalmethods some new trends toward learning control exist Forexample Lu et al [8] designed a learning module based onneural reinforcement learning before the PID control (ismodel can learn from human drivers online based on the on-board sensing information and realize human-like longi-tudinal speed control by learning from the demonstrationparadigm Xing et al [9] provided a personalized leadingvehicle trajectory prediction method based on joint timeseries modeling (is method can generate three differentdriving styles for the leading vehicle through the JTSMalgorithm and achieve better results Lv et al [10 11]proposed a CPS-based framework for codesign optimizationof an automated electric vehicle with different driving styleswhich used unsupervised machine learning to develop thedriving style recognition algorithm

By analyzing the above results this study finds thatconventional methods are applicable but humanization isnot considered due to its inherent logic strategies(erefore based on previous studies we propose a hu-manized vehicle automatic stop-and-go control methodbased on the iterative learning (IL) algorithm and thedynamic time to collision (DTTC) model (is study aimsto develop a humanized learning control model based on ILand DTTC while focusing on the stop-and-go task forautonomous driving (e main contributions of this studyare as follows (1) the driversrsquo control behavior

characteristics are extracted through the real vehicle testswith different types of drivers (2) An automatic drivecontrol method based on the IL algorithm is proposed tolearn the driver behavior and realize the human-likecontrol Based on the IL algorithm complex dynamiccalculations are not required and the control method canbe learned directly from the acceleration curve equation(3) An automatic brake control method based on the DTTCmodel is proposed in this study to learn the driver behaviorand realize the human-like control Based on the DTTCmodel on the one hand the safety of the braking process isimproved by the consideration of dynamic deceleration onthe other hand different types of braking can be learneddirectly from the real vehicle tests (4) Automatic stop-and-go control is implemented by combining the automaticdrive control and automatic brake control methods (evalidity and feasibility of the proposed method are verifiedthrough simulation and real tests

(e remainder of this paper is organized as follows thesecond part is ldquoextraction of driversrsquo control behaviorcharacteristicsrdquo (e third part is the ldquolearning model forvehicle automatic starting controlrdquo (e fourth part is theldquolearning model for vehicle automatic braking controlrdquo (efifth part is ldquosimulation and experimental testsrdquo (e sixthpart is ldquoresults and comparative analysisrdquo (e final part isldquoconclusions and future workrdquo

2 Extraction of Driversrsquo ControlBehavior Characteristics

(e three types of typical drivers are examined based on theprevious classification to obtain the common characteristicsof different drivers with evident driving styles [12ndash14] Inthis study we choose three types namely aggressive drivernormal driver and steady driver Many factors causingdifferent characteristics should be neglected because thispart aims to find the humanized common characteristics ofhuman drivers compared with the conventional model-based design (erefore we assume that the target vehiclestops at the stop line 200m away from the host vehicle(enthe host vehicle starts to catch up with the target vehiclethrough acceleration or deceleration control over a period of

Driver Horizon Judgment Acceleratorbrake pedal

Vehicle Environmental perception Learning Vehicle dynamics models

Engine characteristic

Braking system

A human-like platform for intelligent vehicle

Figure 1 A human-like platform for intelligence vehicle










f(t) 1T

1113946t0+T

t0

f(t)dt +2T

1113946t0+T

t0


1113946t0+T

t0

f(t)cos(ωt)dt cos(ωt)1113890 1113891+

+2T

1113946t0+T

t0


1113946t0+T

t0


(1)








⎧⎨

⎩ (2)







Time (s)

Velo

city

(km

h)


40353025201510

50

0 10 20 30 40 50 60 70 80 90 100 110 120 140130 150 160 170 180 190 200

(a)


Acce

lera

tion

(ms

2 )

435

325

2

ndash2

15

ndash15

ndash3ndash25

1

ndash1

05

ndash050

Time (s)0 10 20 30 40 50 60 70 80 90 100 110 120 140130 150 160 170 180 190 200

(b)



Acce

lera

tion

(ms

2 )

Time (s)


Underfitting

30

25

20

15

10

05

00 25 50 75

(a)


Overfitting

Acce

lera

tion

(ms

2 )

30

25

20

15

10

05

0

Time (s)0 25 50 75

(b)


Acce

lera

tion

(ms

2 )



Time (s)0 25 50 75

30

25

20

15

10

05

0



Erro

r

0 1 2 3 4 5 6 7 8 9 10

25

2

15

1

05

0








TTC Drel

Vrel (3)



Dego Vegot +12

aegot2

Dtar Vtart +12

atart2

Dego Drel + Dtar

Vrelt +12

atart2

+ Drel 0

(4)




Acce

lera

tion

(ms

2 )

Time (s)0 25 50 75

3

25

2

15

1

05

0

(a)

rottl

e deg

ree (

times100

)

03

025

035

02

05

045

04

015

01

005

035 40 45 50 55 60 65

Time (s)



(b)



DTTC




1113969


⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(5)



Dego Vegot +12areqt

2

Dego + Dlowast

Drel + Dtar

areq


V2ego


lowast( 1113857


atar minusV

2rel



⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(6)







vego aego

Estimated rangeDtar


vtar atar




Brake Drive


Time for simulation

Car module

160

Velocity

Acceleration

Distance

Relative velocity


control methods


module


Execute module

Car_signal

Time (sec)Clock

Front_sensor

Vx

Vr

Distance

Axa_des Out3



Dego

vego aego

Estimated rangeDtar


vtar atar




5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Velo

city

(km

h)

(a)

Figure 12 Continued



5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Dist

ance

(m)

(b)



Brake


Drive


30

25

20

15

10

5

20 4 6 8 10 12 14 16 18 20Time (s)

Vel

ocity

(km

h)

(a)


20 4 6 8 10 12 14 16 18 20Time (s)

14

12

10

8

6

4

2

Dist

ance

(m)

(b)











Data Availability


Disclosure




3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )



Acknowledgments


References





























f(t) 1T

1113946t0+T

t0

f(t)dt +2T

1113946t0+T

t0


1113946t0+T

t0

f(t)cos(ωt)dt cos(ωt)1113890 1113891+

+2T

1113946t0+T

t0


1113946t0+T

t0


(1)








⎧⎨

⎩ (2)







Time (s)

Velo

city

(km

h)


40353025201510

50

0 10 20 30 40 50 60 70 80 90 100 110 120 140130 150 160 170 180 190 200

(a)


Acce

lera

tion

(ms

2 )

435

325

2

ndash2

15

ndash15

ndash3ndash25

1

ndash1

05

ndash050

Time (s)0 10 20 30 40 50 60 70 80 90 100 110 120 140130 150 160 170 180 190 200

(b)



Acce

lera

tion

(ms

2 )

Time (s)


Underfitting

30

25

20

15

10

05

00 25 50 75

(a)


Overfitting

Acce

lera

tion

(ms

2 )

30

25

20

15

10

05

0

Time (s)0 25 50 75

(b)


Acce

lera

tion

(ms

2 )



Time (s)0 25 50 75

30

25

20

15

10

05

0



Erro

r

0 1 2 3 4 5 6 7 8 9 10

25

2

15

1

05

0








TTC Drel

Vrel (3)



Dego Vegot +12

aegot2

Dtar Vtart +12

atart2

Dego Drel + Dtar

Vrelt +12

atart2

+ Drel 0

(4)




Acce

lera

tion

(ms

2 )

Time (s)0 25 50 75

3

25

2

15

1

05

0

(a)

rottl

e deg

ree (

times100

)

03

025

035

02

05

045

04

015

01

005

035 40 45 50 55 60 65

Time (s)



(b)



DTTC




1113969


⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(5)



Dego Vegot +12areqt

2

Dego + Dlowast

Drel + Dtar

areq


V2ego


lowast( 1113857


atar minusV

2rel



⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(6)







vego aego

Estimated rangeDtar


vtar atar




Brake Drive


Time for simulation

Car module

160

Velocity

Acceleration

Distance

Relative velocity


control methods


module


Execute module

Car_signal

Time (sec)Clock

Front_sensor

Vx

Vr

Distance

Axa_des Out3



Dego

vego aego

Estimated rangeDtar


vtar atar




5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Velo

city

(km

h)

(a)

Figure 12 Continued



5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Dist

ance

(m)

(b)



Brake


Drive


30

25

20

15

10

5

20 4 6 8 10 12 14 16 18 20Time (s)

Vel

ocity

(km

h)

(a)


20 4 6 8 10 12 14 16 18 20Time (s)

14

12

10

8

6

4

2

Dist

ance

(m)

(b)











Data Availability


Disclosure




3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )



Acknowledgments


References
























⎧⎨

⎩ (2)







Time (s)

Velo

city

(km

h)


40353025201510

50

0 10 20 30 40 50 60 70 80 90 100 110 120 140130 150 160 170 180 190 200

(a)


Acce

lera

tion

(ms

2 )

435

325

2

ndash2

15

ndash15

ndash3ndash25

1

ndash1

05

ndash050

Time (s)0 10 20 30 40 50 60 70 80 90 100 110 120 140130 150 160 170 180 190 200

(b)



Acce

lera

tion

(ms

2 )

Time (s)


Underfitting

30

25

20

15

10

05

00 25 50 75

(a)


Overfitting

Acce

lera

tion

(ms

2 )

30

25

20

15

10

05

0

Time (s)0 25 50 75

(b)


Acce

lera

tion

(ms

2 )



Time (s)0 25 50 75

30

25

20

15

10

05

0



Erro

r

0 1 2 3 4 5 6 7 8 9 10

25

2

15

1

05

0








TTC Drel

Vrel (3)



Dego Vegot +12

aegot2

Dtar Vtart +12

atart2

Dego Drel + Dtar

Vrelt +12

atart2

+ Drel 0

(4)




Acce

lera

tion

(ms

2 )

Time (s)0 25 50 75

3

25

2

15

1

05

0

(a)

rottl

e deg

ree (

times100

)

03

025

035

02

05

045

04

015

01

005

035 40 45 50 55 60 65

Time (s)



(b)



DTTC




1113969


⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(5)



Dego Vegot +12areqt

2

Dego + Dlowast

Drel + Dtar

areq


V2ego


lowast( 1113857


atar minusV

2rel



⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(6)







vego aego

Estimated rangeDtar


vtar atar




Brake Drive


Time for simulation

Car module

160

Velocity

Acceleration

Distance

Relative velocity


control methods


module


Execute module

Car_signal

Time (sec)Clock

Front_sensor

Vx

Vr

Distance

Axa_des Out3



Dego

vego aego

Estimated rangeDtar


vtar atar




5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Velo

city

(km

h)

(a)

Figure 12 Continued



5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Dist

ance

(m)

(b)



Brake


Drive


30

25

20

15

10

5

20 4 6 8 10 12 14 16 18 20Time (s)

Vel

ocity

(km

h)

(a)


20 4 6 8 10 12 14 16 18 20Time (s)

14

12

10

8

6

4

2

Dist

ance

(m)

(b)











Data Availability


Disclosure




3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )



Acknowledgments


References





















Acce

lera

tion

(ms

2 )

Time (s)


Underfitting

30

25

20

15

10

05

00 25 50 75

(a)


Overfitting

Acce

lera

tion

(ms

2 )

30

25

20

15

10

05

0

Time (s)0 25 50 75

(b)


Acce

lera

tion

(ms

2 )



Time (s)0 25 50 75

30

25

20

15

10

05

0



Erro

r

0 1 2 3 4 5 6 7 8 9 10

25

2

15

1

05

0








TTC Drel

Vrel (3)



Dego Vegot +12

aegot2

Dtar Vtart +12

atart2

Dego Drel + Dtar

Vrelt +12

atart2

+ Drel 0

(4)




Acce

lera

tion

(ms

2 )

Time (s)0 25 50 75

3

25

2

15

1

05

0

(a)

rottl

e deg

ree (

times100

)

03

025

035

02

05

045

04

015

01

005

035 40 45 50 55 60 65

Time (s)



(b)



DTTC




1113969


⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(5)



Dego Vegot +12areqt

2

Dego + Dlowast

Drel + Dtar

areq


V2ego


lowast( 1113857


atar minusV

2rel



⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(6)







vego aego

Estimated rangeDtar


vtar atar




Brake Drive


Time for simulation

Car module

160

Velocity

Acceleration

Distance

Relative velocity


control methods


module


Execute module

Car_signal

Time (sec)Clock

Front_sensor

Vx

Vr

Distance

Axa_des Out3



Dego

vego aego

Estimated rangeDtar


vtar atar




5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Velo

city

(km

h)

(a)

Figure 12 Continued



5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Dist

ance

(m)

(b)



Brake


Drive


30

25

20

15

10

5

20 4 6 8 10 12 14 16 18 20Time (s)

Vel

ocity

(km

h)

(a)


20 4 6 8 10 12 14 16 18 20Time (s)

14

12

10

8

6

4

2

Dist

ance

(m)

(b)











Data Availability


Disclosure




3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )



Acknowledgments


References


























TTC Drel

Vrel (3)



Dego Vegot +12

aegot2

Dtar Vtart +12

atart2

Dego Drel + Dtar

Vrelt +12

atart2

+ Drel 0

(4)




Acce

lera

tion

(ms

2 )

Time (s)0 25 50 75

3

25

2

15

1

05

0

(a)

rottl

e deg

ree (

times100

)

03

025

035

02

05

045

04

015

01

005

035 40 45 50 55 60 65

Time (s)



(b)



DTTC




1113969


⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(5)



Dego Vegot +12areqt

2

Dego + Dlowast

Drel + Dtar

areq


V2ego


lowast( 1113857


atar minusV

2rel



⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(6)







vego aego

Estimated rangeDtar


vtar atar




Brake Drive


Time for simulation

Car module

160

Velocity

Acceleration

Distance

Relative velocity


control methods


module


Execute module

Car_signal

Time (sec)Clock

Front_sensor

Vx

Vr

Distance

Axa_des Out3



Dego

vego aego

Estimated rangeDtar


vtar atar




5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Velo

city

(km

h)

(a)

Figure 12 Continued



5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Dist

ance

(m)

(b)



Brake


Drive


30

25

20

15

10

5

20 4 6 8 10 12 14 16 18 20Time (s)

Vel

ocity

(km

h)

(a)


20 4 6 8 10 12 14 16 18 20Time (s)

14

12

10

8

6

4

2

Dist

ance

(m)

(b)











Data Availability


Disclosure




3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )



Acknowledgments


References





















DTTC




1113969


⎧⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎩

(5)



Dego Vegot +12areqt

2

Dego + Dlowast

Drel + Dtar

areq


V2ego


lowast( 1113857


atar minusV

2rel



⎧⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎨

⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎩

(6)







vego aego

Estimated rangeDtar


vtar atar




Brake Drive


Time for simulation

Car module

160

Velocity

Acceleration

Distance

Relative velocity


control methods


module


Execute module

Car_signal

Time (sec)Clock

Front_sensor

Vx

Vr

Distance

Axa_des Out3



Dego

vego aego

Estimated rangeDtar


vtar atar




5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Velo

city

(km

h)

(a)

Figure 12 Continued



5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Dist

ance

(m)

(b)



Brake


Drive


30

25

20

15

10

5

20 4 6 8 10 12 14 16 18 20Time (s)

Vel

ocity

(km

h)

(a)


20 4 6 8 10 12 14 16 18 20Time (s)

14

12

10

8

6

4

2

Dist

ance

(m)

(b)











Data Availability


Disclosure




3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )



Acknowledgments


References






















Brake Drive


Time for simulation

Car module

160

Velocity

Acceleration

Distance

Relative velocity


control methods


module


Execute module

Car_signal

Time (sec)Clock

Front_sensor

Vx

Vr

Distance

Axa_des Out3



Dego

vego aego

Estimated rangeDtar


vtar atar




5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Velo

city

(km

h)

(a)

Figure 12 Continued



5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Dist

ance

(m)

(b)



Brake


Drive


30

25

20

15

10

5

20 4 6 8 10 12 14 16 18 20Time (s)

Vel

ocity

(km

h)

(a)


20 4 6 8 10 12 14 16 18 20Time (s)

14

12

10

8

6

4

2

Dist

ance

(m)

(b)











Data Availability


Disclosure




3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )



Acknowledgments


References






















5 10 15 20 25 30 35 40 45 50 55 60 65 70Time (s)

75 80 85 90 95 100 105 110 115 120 125 130 135 140 145 150 155 160

5045

40

35

30

25

20

15

10

5

0

Dist

ance

(m)

(b)



Brake


Drive


30

25

20

15

10

5

20 4 6 8 10 12 14 16 18 20Time (s)

Vel

ocity

(km

h)

(a)


20 4 6 8 10 12 14 16 18 20Time (s)

14

12

10

8

6

4

2

Dist

ance

(m)

(b)











Data Availability


Disclosure




3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )



Acknowledgments


References





























Data Availability


Disclosure




3

2

1

0

ndash1

ndash2

ndash3

ndash40 2 4 6 8 10 12 14 16

10 12 14 16

18 20

05

04

03

02

01

ro

ttle d

egre

e (times1

00

)

Time (s)

Time (s)

Acce

lera

tion

(ms

2 )



Acknowledgments


References





















Acknowledgments


References





















intelligent vehicle automatic stop-and-go task based on

Documents