Reference: [SAE, 2014]€¦ · Industrial researcher, PhD Peter Nilsson (Volvo Global Trucks...

1

ProfessorinVehicleDynamics,PhDBengtJacobson(ChalmersUniversityofTechnology;Sweden)Industrialresearcher,PhDPeterNilsson(VolvoGlobalTrucksTechnology;Sweden)Seniorresearcher,PhDMatsJonasson(ChalmersUniversityofTechnology;Sweden)

NeedsforPhysicalModelsandRelatedMethodsforDevelopmentof

AutomatedRoadVehicles

Automated DrivingSAEJ3016

Reference: [Matthijs Klomp, et al, 2019]

Reference: [SAE, 2014]

Jubilee Symposium 2019: Future Directions of System Modeling and Simulation

2

“FunctionArchitecture”forvehiclemotion&energy

VehicleMotionManagement

MotionSupportDeviceManagement

VehicleEnvironment

TrafficSituationManagement

RouteManagement

HumanMachineInterface

Req

uest

s

Esti

mat

es,

Conf

iden

ce

Stat

us,

Capa

bilit

ies

accelerations

environm

entobservations

wheeltorques,axlesteeringangles

velocities

from

hum

andriverdevices

Reference: [Nilsson, 2017]

TrafficSituation

MotionSupportDevices

VehicleMotion

physicalmodels



VehicleEnvironment


RouteManagement


data‐drivenmodels

Modelsforvehiclemotionandenergycontroldesign

• drivers• micro traffic

• estimators

sub‐system/actuatormodels

motionrelativelane&traffic

motion&individualtyreforces

fuel / SOC

velocity&energy • macro traffic


3



VehicleEnvironment


RouteManagement


Nextspeakers

Traffic Situation Management,Dynamically Feasible Trajectories,

Peter Nilsson, Volvo Trucks


4

Vehicle Longitudinal and Lateral Control

Motion Planning (Trajectory planning)Behaviour planning (Tactical decision)

Transitions between

driving modes

Robust control

Thrust and comfort

Predictions of surrounding

traffic and VRUs

Situation assessment

Consistent and predictable behaviour

Sensor imperfections and occlusions

Computational efficient methods

Examples of challenges for TSM

Functional safety

Predictions of surrounding

traffic and VRUsCollision free trajectoriesComfortable

and predictable trajectories

Dynamically feasible

trajectories

Trajectory planning“Trajectoryplanningisageneralizationofpathplanning,involvedwithplanningthestateevolutionintimewhilesatisfyinggivenconstraintsonthestatesandactuation”

Commonlyusedmethods:

• Numerical optimization (e.g. MPC)• Graph search (e.g. A*)• Neural network (e.g. Nvidia PilotNet)• ...

Trajectoryplanningexample:left curve, tractor semi-trailer


5

Heavy duty combination vehicles

Exampleofmotionconstraints:

• Position of first unit• Position of trailer units (off-tracking)• Roll-over threshold (rearward amplification)• ...

Trajectory planning modelling

Exampleofmodelling:

• One-track models : 𝑥 𝑓 𝑥, 𝑢, 𝑤• Possible states for A-double

• 1st unit (tractor) : 𝑣 , 𝑣 , 𝜓• 2nd unit (trailer) :∆𝜓 , Δ𝜓• 3rd unit (dolly) :∆𝜓 , Δ𝜓• 4th unit (trailer) :∆𝜓 , Δ𝜓


6

Vehicle variants and trajectory planning challenges

Challenge:

Trajectory planning methodology needs to scalable and robust with respect to variant combinatorics

Vehiclevariantcombinatorics:

• Powertrain : 10^2 variants• Chassis : 10^3 variants• Vehicle load 7 - 120t (incl. different heights to CoG)• Vehicle units : 1-4

Trajectoryplanningexample:Roundabout, tractor semi-trailer

Vehicle Motion Management, Road friction estimation,

Mats Jonasson


7

Challenges for VMM


slip (%)

force (N)

low friction

~ 5%

𝑓 𝜇𝑓

Most driving take place here, not possible to distinguish between low or high friction

To estimate frictionthe tyre must at least be excited to the nonlinear region at “the bend”

ABS activation, friction can be found 𝜇

Definitions:0 𝜇 0.4Low friction0.4 𝜇 0.7Mid friction0.7 𝜇High friction

high friction

Road condition – road frictionMore than 10% of all accidents occur because of slippery conditions*

In the US: yearly approx 500 000 accidents of which 1800 are deadly*

* Reference: [IVSS Road Friction Estimation Part II]* Reference: [ US Department of Transportation – Federal Highway Administration** Reference: [Wallman. Tema vintermodell – olycksrisker vid olika vinterväglag]


8

Confusion matrix of road friction


Assumed friction

True friction

High(dry asphalt)

Low(snow)

Low (snow) High (dry asphalt)

• False slippery warnings• AD Vehicle will drive

unacceptably slow (not transport efficient)

Vehicle speed can be adapted to friction

Vehicle speed can be adapted to friction

• AD Vehicle will drive too fast (not safe)

• High frequency of accidents

Methods for road friction estimationOptical measurement device Model-based estimator Machine learning estimator

• Contactless• Requires a map from

texture to friction

• Use the tyre as the sensor

• Requires knowledgeabout tyre physics

• Use features withoutknowledge of physics

• Requires training


9

State-of-the art model-based estimator

Kinetic and kinematic

models

Wheel speeds,Inertial Meas. Syst.Steering angle

Pre-processing Friction estimator

Tyre forces

Tyre slip

�̂�

Features and correlation to friction

Features 1...86

Temperature, GPS, vehicle speed, surface and road type are important features for friction estimation

Surface & road type are not available in the sensor suite -> important to use a new sensor e.g. a cameraTemperature

GPSwheel speeds

Surface, Road Type

Correlation to true friction

* Reference[Roychowdhury, et al, 2018]


10

Challenges road friction estimation

• General: • Difficult to identify friction for normal driving (low friction utilization)

• Model-based: • Model uncertainties for different tyres - the physics is hard to model• The pre-processing is not accurate enough

• Machine learning: • Generalizability of machine learning algorithms to various situations• Generalizability would require large testing• Training of machine learning algorithms require ground truth – road friction is hard

to measure

Reference [Jonasson, et al] 2018

Motion Devices, Virtual Verification, Wheel Model,

Bengt Jacobson


11



VehicleEnvironment


RouteManagement


ModelsforVirtualVerification

For Virtual Verification:

• Higher accurate and larger validityrange than for control design.

• But onlysimulate‐able, no need for linearized, inversion, etc.

MechatronicSensor

Environment (other vehicles, lane edges, …)

Mecha‐tronicActuator

wheels

absolute position

relative position

suspension

body

Realworld

Theoreticalworld

Computation/Simulation

Explicitformmodelling

Mathematicalmodelling

Interpret results, judge model validity

Physicalmodelling

Final Design

Formulate engineering

task (problem)

Initial Design

Re-DesignEvaluate

requirement fulfilment

Real-world testing

OKNOK

…oneviewofmodelbasedengineering

Drawing

DAE(Modelica)

ODE


12

Wheelmodelasexample

𝟏 𝟑 𝟒 𝟐 𝟑 ⋅ 𝟐 𝟐𝟔 wheels

104 tonnes, 33 m

𝑇 𝑇

𝜔

𝑇𝐹

𝐹𝐹

𝑣𝑣

Wheelmodelusecases

𝑇 𝑇

𝜔

𝑇𝐹

𝐹𝐹

𝑣

𝑣

Control Longitudinal vehicle translation Control Longitudinal wheel rotation


13

Wheelmodel,Mechanicalchallenges

𝐹 𝐶 ⋅ 𝑠 ;

𝑠𝑅 ⋅ 𝜔 𝑣

𝑅 ⋅ 𝜔;

𝐹 , 𝐹 min 𝐶 ⋅ 𝑠 , 𝜇 ⋅ 𝐹 ⋅ sin 𝜃 , cos 𝜃 ;

𝑠𝑅 ⋅ 𝜔 𝑣 𝑣

𝑅 ⋅ 𝜔; 𝜃 𝑎𝑟𝑐𝑡𝑎𝑛2 𝑣 , 𝑅 ⋅ 𝜔 𝑣 ;

𝐽 ⋅ 𝜔 𝑇 𝐹 ⋅ 𝑅 𝑇 ;𝑇 sign 𝜔 ⋅ 𝑇 𝑅𝑅𝐶 ⋅ 𝑅 ⋅ 𝐹 ;

If vehicle standstill and two or more wheels locked:

Statically underdetermined

ContinuouslyRenewedFrictionSurfaces

RelativeVelocityDirection

DryFrictioninBrake

RollingResistance

Multiplewheels

Wheelmodelinitsmodelcontext



VehicleEnvironment


RouteManagement


MechatronicSensor

Environment (other

vehicles, lane edges, …)

Mecha‐tronicActuator

wheels

absolute position

relative position

suspension

body


14

Conclusions

Automated driving needs modelling in many aspects:

• TSM and VMM needs Physical modelling for “Control/algorithmdesign”.

• “Virtualverification” drives Physical modelling, incl. exchange of models between organisation.

𝑇 𝑇

𝜔

𝑇𝐹

𝐹𝐹

𝑣𝑣

Youhaveseen:


15

ReferencesMatthijs Klomp, et al, Trendsinvehiclemotioncontrolforautomateddrivingonpublicroads, 2019. https://www.tandfonline.com/doi/full/10.1080/00423114.2019.1610182

Nilsson, Peter, TrafficSituationManagementforDrivingAutomationofArticulatedHeavyRoadTransports‐ Fromdriverbehaviourtowardshighwayautopilot, PhD thesis, Chalmers, 2017. https://research.chalmers.se/publication/251872

Weitao Chen et al, IntegrationandAnalysisofEPASandChassisSysteminFMI‐basedco‐simulation,2019. http://www.ep.liu.se/ecp/article.asp?issue=157%26article=74

SAE, TaxonomyandDefinitionsforTermsRelatedtoOn‐RoadMotorVehicleAutomatedDrivingSystems,Standard J3016, 2014.https://saemobilus.sae.org/content/j3016_201401

IVSS Road Friction Estimation Part II Report, http://www.ivss.se

US Department of Transportation – Federal Highway Administration

C.-G Wallman. Tema vintermodell – olycksrisker vid olika vinterväglag. Technical Report N60-2001, VTI, 2001.

S. Roychowdhury, M. Zhao, A. Wallin, N. Ohlsson, M. Jonasson, ‘Machine learning models for road surface and friction estimation using front-camera images’, International Joint Conference on Neural Networks (IJCNN 2018), Rio, Brazil, 2018.

M. Jonasson, N. Olsson, S. R. Chowdhury, S. Muppirisetty, Z. Minming, AutomatedRoadFrictionEstimationusingCar‐sensorSuite:MachineLearningApproach, Autonomous Vehicle Software Symposium, Stuttgart, Germany, 2018

Thanks for your attention


Reference: [SAE, 2014]€¦ · Industrial researcher, PhD Peter Nilsson (Volvo Global Trucks...

Documents

Transcript of Reference: [SAE, 2014]€¦ · Industrial researcher, PhD Peter Nilsson (Volvo Global Trucks...