Universität Bremen

Self-Localization in Large-Scale Environments for theBremen Autonomous Wheelchair

Axel Lankenau, Thomas Röfer, Bernd Krieg-Brückner

Bremen Institute of Safe Systems

Center for Computing Technology (TZI)

Universität Bremen

Germany

Self-Localization in Large-Scale Environments 2

Universität Bremen

Outline of the Talk

� Motivation� Route Generalization

� Route Graph

� Probabilistic Approach to Self-Localization

� Experiment in Large-Scale Environment

� Related Work� Summary & Outlook

…

Self-Localization in Large-Scale Environments 3

Universität Bremen

Working Group “Cognitive Robotics”

Navigation Assistant

Spatial Cognition Safe Robotics Driving Assistant

SLAM

Reality

FAULT TREE ENVIRONMENT+EUC

SAFETYMECHANISMS

ArtificialEvents Relevant

Events

Arti ficialEvents

Release/Block

Safety module failure Collis ion

Communicationfailure

System safety vio lated

ActiveCollis ion

PassiveCollision

Reality

FAULT TREE ENVIRONMENT+EUC

SAFETYMECHANISMS

ArtificialEvents Relevant

Events

Arti ficialEvents

Release/Block

Safety module failure Collis ion

Communicationfailure

System safety vio lated

ActiveCollis ion

PassiveCollision

SLAMRoboCupRoboCup

Self-Localization in Large-Scale Environments 4

Universität Bremen

Navigation Assistant

Marauder‘s Map

Self-Localization in Large-Scale Environments 5

Universität Bremen

Navigation Assistant – Rolland

user interface visual instructions

Turn right at the next possibilityacoustic instructions

Self-Localization in Large-Scale Environments 6

Universität Bremen

224cm, 75°, 799cm, -83°, 880cm, -87°, 260cm

Generalization of Locomotion

Self-Localization in Large-Scale Environments 7

Universität Bremen

Modeling the Environment

A

E

BDC

F

I H

GJunction „DCI“

Self-Localization in Large-Scale Environments 8

Universität Bremen

Junctions

� Angle between incoming and outgoing segment

A B

C

1,5 m

1 m

Crossing

-90°1,5 m

A B

A B

A B

90°

0°

-90°

1 m

1 m

1 m

B

C

Junction

� Length of outgoing segment

� List of incoming segments

Self-Localization in Large-Scale Environments 9

Universität Bremen

Inductive Approach

� Idea: Assigning route corners to junctions

� Two-step assignment� Corner matches a junction

� The rest of the generalized route matches up to the junction

A

E

BDC

F

I H

G

routegraph

generalization ofdriven route

A

E

BDC

F

I H

G

routegraph

Self-Localization in Large-Scale Environments 10

Universität Bremen

Probabilities from Similarities

0

0.2

0.4

0.6

0.8

1

0 0.5 1 1.5 2 2.5 3

angle difference (radians)

probability

corridor length(cm) route segment length

(cm)

probability

Self-Localization in Large-Scale Environments 11

Universität Bremen

rout

e gr

aph

rout

e

Matching Corners

� Differentiation between the probabilities that� the corner previously generalized really exists, …� Angle of corner is similar to angle of the junction in the route graph

� ... the corner has been detected erroneously, ...� Angle of corner is similar to 0°

� ... a corner has been overlooked, ...� Angle of the junction in the route graph is similar to 0°

� ... it has been turned around at the corner previously generalized � Angle of corner is similar to 180°

Self-Localization in Large-Scale Environments 12

Universität Bremen

Propagation of Probabilities

=t

⋅tt+1

Self-Localization in Large-Scale Environments 13

Universität Bremen

Determining the Candidate Junction

Self-Localization in Large-Scale Environments 14

Universität Bremen

� Building: MZH

Indoor and Outdoor Navigation

Self-Localization in Large-Scale Environments 15

Universität Bremen

� Building: NW 2

Indoor and Outdoor Navigation

Self-Localization in Large-Scale Environments 16

Universität Bremen

� Buildings: IW + BIBA

Indoor and Outdoor Navigation

Self-Localization in Large-Scale Environments 17

Universität Bremen

� Overall length: 2176 m

� Building: MZH

Indoor and Outdoor Navigation

Self-Localization in Large-Scale Environments 18

Universität Bremen

Odometry Data

MZH

NW2 FZB

IW

Self-Localization in Large-Scale Environments 19

Universität Bremen

Route Graph

Self-Localization in Large-Scale Environments 20

Universität Bremen

Results I

Believed in previous corner

Not believed in previous corner

Intensity encodes confidence

Speeded up by factor 70

Most probable position

Self-Localization in Large-Scale Environments 21

Universität Bremen

Reference: Laser Scan Map

t

o p

n

q r

m l

k

d

c

PM

aWX

ZT

RL

US

V

h

J

D

f

i

e

Y

g

b

C

G

Q

I

ON

KH

s

F E

j

A

B

Self-Localization in Large-Scale Environments 22

Universität Bremen

Results II

Self-Localization in Large-Scale Environments 23

Universität Bremen

Related Work

Self-Localization in Large-Scale Environments 24

Universität Bremen

Summary & Outlook

�Probabilistic self-localization based on locomotion

� Minimal sensor equipment

� Suitable for large-scale environments

� Efficient

�Future Work

� Simultaneous localization and mapping (SLAM)

� Integration of additional sensor measurements