San Francisco, California, USA, FW8, Room 8, 30th september …pnavhe11.irccyn.ec-nantes.fr/material/session5/Avanzini/... · 2015-11-05 · Philippe Martinet IROS11 International

PhilippeMartinet

IROS11 International workshop on Perception and Navigationfor Autonomous Vehicles in Human Environment

San Francisco, California, USA, FW8, Room 8, 30th september 2011Session: Mobile Robot modeling and control

LASMEA, UMR 6602 du CNRSClermont-Ferrand, France

[email protected]://wwwlasmea.univ-bpclermont.fr/rosace

http://wwwlasmea.univ-bpclermont.fr/Personnel/Philippe.Martinethttp://www.irccyn.ec-nantes.fr/~martinet

Avanzini Pierre, Thuilot Benoit, Martinet Philippe

A control strategy taking advantage of inter-vehiclecommunication for platooning navigation in urban environment

CityvipProject

IROS11 International workshop on Perception and Navigation for Autonomous Vehicles in Human Environment


PhilippeMartinet 2



Outline of the presentation

IntroductionMotivationPrevious contribution

Autonomous navigationTrajectory based strategyScale factor problemOnline estimation

Conclusion

Decentralized control strategyModelingControl

ResultsScale factor estimationExperimental vehiclesReal Experiment

PhilippeMartinet 3



Introduction Decentralized Control strategy

Autonomousnavigation

Results Conclusion

waste of time / activitieswaste of energy (fuel, gas, ...)

atmospheric pollution noise pollution

Motivation : Inner-cities congestion

PhilippeMartinet 4



Introduction Results Conclusion

Motivation: Urban Transportation Systems Concept

"they have to answer to various public needs"

Car sharing concept

Their attractiveness depends mainly on their flexibility

appears as a suitable answer in specific areas (inner-cities pedestrian zones, airport terminals, ...)

Such systems have been developed since the mid-90's :Praxitèle in France, CarLink in USA, Crayon in Japan…

Praxitèle

Lisélec

Crayon

Honda Singapour

CARLINK

Decentralized Control strategy


PhilippeMartinet 5




Motivation: Autonomous navigation

Car sharing concept demands for

automatic guidance capabilities

To transport passengers

in an entire automatic way

To bring back empty vehicles to stations for refilling and reuse

station #1

station #2

station #3

To deliver free vehicles for customer useAutonomous vehicle Platoon capability



PhilippeMartinet 6



Previous Work AGV & Platoon


04 05 06 07 08 09 10 1103

AG

VPL

AT

OO

N

02Fir

st 2 v

ehicl

esRTK-G

PS (IA

V04)

RTK-GPS

(ICRA05

,IROS0

5)

VISIO

N 3D (IR

OS05)

VISIO

N(IC

AR-CV08

)RTK-G

PS(IR

OS09)

VISIO

N (ove

r 1km

)

3D, T

opolo

gical

AutonomousLeader

Manually drivenLeader

RTK-GPS

Joini

ng, in

sertio

n

RTK-GPS

VISIO

N

(IROS1

0)



VISIO

N

(PNAVHE11

)

PhilippeMartinet 7



Previous Work AGV

Introduction

Automatic Guided Vehicles

Navigation using RTK-GPSClermont-Ferrand : LASMEA-GRAVIR

Navigation using vision onlyClermont-Ferrand : LASMEA-GRAVIR

BODEGA : Clermont-Fd 2005Mobivip : Clermont-Fd 2004

Results ConclusionDecentralized Control strategy


PhilippeMartinet 8



Previous Work AGV

Introduction


Platoon using RTK-GPSInsertion

Clermont-Ferrand : LASMEA-GRAVIR

Platoon using RTK-GPSJoining

Clermont-Ferrand LASMEA-GRAVIR

Mobivip : Clermont-Fd 2006Mobivip : Clermont-Fd 2006

Platooning



PhilippeMartinet 9



Previous Work AGV

Introduction


Manually driven Platoon using RTK-GPS

Clermont-Ferrand : LASMEA-GRAVIR

Platoon using VisionClermont-Ferrand LASMEA-GRAVIR

Cityvip : Clermont-Fd 2010Cityvip : Clermont-Fd 2009

Platooning



PhilippeMartinet 10




Modeling: Notations

F θ : angular deviationF y : lateral deviation

F δ : wheel angleF v : vehicle linear velocity

F l : wheel base length

F s : curvilinear coordinateF c(s) : curvature

M

C

Reference pathtangent

O

δ

y

θ XA

YA

M

v



PhilippeMartinet 11




Modeling: platoon F di : curvilinear distance



PhilippeMartinet 12




Modelling is derived under non-slipping assumptions (tricyle model)[Samson95, Daviet95, Thuilot04] → relies on a kinematic model

→ designed with respect to the reference path

Modeling: State model

di = si − si+1⎧⎨⎩ di = si − si+1di = vi

cos θi1−yic(si) − vi+1

cos θi+11−yi+1c(si+1)

Longitudinal Modelling

Control objectives

(si − si+1) to d

⎧⎪⎪⎨⎪⎪⎩si = vi

cos θi1−yic(si)

yi = vi sin θi˙θi = vi

³tan δil − c(si) cos θi

1−yi c(si)

´Syst Ia

vi+1

δi+1

yi+1 and θi+1 to 0

Syst Ib

The vector (si, yi, θi) describes the state of the ith

vehicle



PhilippeMartinet 13




Chained Control vector M = (m1i,m2i)T = Υ(vi, δi)

Chained state vector

Control: Lateral control in curved path following [Thuilot04]

(∗)0 = d

ds(∗)

Syst Ia

A = (a1i, a2i, a3i)T = Θ(si, yi, θi)

Chained System [Samson95]Exact linearization

Chained model driven by s

→ linear system structure→ full lateral / longitudinal decoupling

m3i = −kd a3i − kp a2iPD control law

a002i + kd a02i + kp a2i = 0 (a3i, a2i)→ (0, 0)

³yi, θi

´→ (0, 0)

Syst IIa

(kd, kp) tuning specifies a settling distance

Vehicle trajectories are velocity independent

½a02i = a3ia03i = m3i

⎧⎨⎩ a1i = m1i

a2i = a3im1i

a3i = m2i

⎧⎨⎩a1i = sia2i = y1ia3i = tan θi (1− yi c(si))



PhilippeMartinet 14




Control: Longitudinal control in curved path followingdi = si − si+1 ei = di − d

ei = vicos θi

1 − yic(si)− vi+1

cos θi+11 − yi+1c(si+1)

Proportional control law

k > 0

Syst Ib

Syst IIb

Exact linearization

Kinematic model

vi+1 =1 − yi+1 c(si+1)

cos θi+1

Ãvi cos θi

1 − yi c(si)− ui+1

!

ui+1 = −k ei vi+1 =1 − yi+1 c(si+1)

cos θi+1

Ãvi cos θi

1 − yi c(si)+ kei

!ui+1 = ei auxiliary control law

Standard longitudinal control modeei = −k ei di → d

[Bom05]



PhilippeMartinet 15




Control: Longitudinal control in curved path following

GCS

MCS

LCS

Leader

Leader

Leader

LCS

GC

SM

CS

MCS : Mixte Control Strategy

xi+1 = e1i+1

xi+1 = σi+1e1i+1 + (1− σi+1)eii+1

xi+1 = eii+1 eii+1=si−si+1−d

e1i+1 = s1 − si+1 − i.d



PhilippeMartinet 16




Trajectory based strategy



Autonomous platoon in 3 steps:

-Learning step (manuallydriven) (offline)

-Trajectory (offline)

-Real time localization andcontrol using V2V communication (online)

PhilippeMartinet 17



Introduction Results ConclusionDecentralized Control strategy


Trajectory based strategy Manually driven Autonomous platoon(Platoon Leader:1st vehicle)

PhilippeMartinet 18





Trajectory based strategy Autonomous platoon(Platoon follower: 2nd vehicle)

First solution using

Laser rangefinderfor the onlinescale factorestimation

PhilippeMartinet 19





Second solution using

odometersfor the onlinescale factorestimation

Trajectory based strategy Autonomous platoon(Platoon follower: 2nd vehicle)

PhilippeMartinet 20




Scale factor estimation On-line local scale factor computation (odometer)

We design an observer to estimate

The vehicle state space model expressed in the virtual vision world



[ICARCV10]

PhilippeMartinet 21




Experimental results On-line scale factor estimation



PhilippeMartinet 22




Experimental vehicles

Thales Navigation ‘Zmax' RTK GPS receiver

positionvelocity

supplied at 10 Hz, with a 2cm accuracy

18 km.h-1

4 DCmotors

lead-acid batteries

Cycab vehicle Robosoft company

camera



PhilippeMartinet 23




Experimental vehicles : platoon architecture

(ith CycabState vector)

High levelComputer

Cycab Computer Cycab Computer

CAN CAN

WiFi Network

Wireless Communication

RTK-GPS

MPC MPC MPC MPC

Actuators Actuators

Cycab i+1Cycab i

RS-232

High levelComputer

RTK-GPS

RS-232

WiFiNetwork

UHF UHF

18 km.h-1

Cycab from Robosoft

IEEE-1394 IEEE-1394



PhilippeMartinet 24




Experimental results Manual guidance mode : Vehicle lateral deviations



PhilippeMartinet 25




Experimental results



Manual guidance mode : Vehicle inter-distance errors

PhilippeMartinet 26




Experimental results



Localisation by vision : Vehicle inter-distance errors

PhilippeMartinet 27




Conclusion

Autonomous navigation control strategyModeling, Control and Communication

Scale factor estimationObserver design : range finder based, odometer based

ValidationOn line scale factor estimationReal experimentation with 4 vehicles

PerspectivesLeader manually driven platoon using solely vision



PhilippeMartinet 28



Introduction Results

[email protected]

http://wwwlasmea.univ-bpclermont.fr/rosace

Thanks for your attention

Any questions

ConclusionDecentralized Control strategy


Documents

San Francisco, California, USA, FW8, Room 8, 30th september …pnavhe11.irccyn.ec-nantes.fr/material/session5/Avanzini/... · 2015-11-05 · Philippe Martinet IROS11 International