Poster Teaser Session 1

jinksimaginaryΤεχνίτη Νοημοσύνη και Ρομποτική

7 Νοε 2013 (πριν από 3 χρόνια και 11 μήνες)

73 εμφανίσεις

12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Poster
Teaser Session 1
Wednesday, 31/08/2011
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
A Navigation System for a High
-
Speed Professional Cleaning Robot
Gorka Azkune, Mikel Astiz, Urko Esnaola, Unai Antero, Jose Vicente Sogorb, Antonio Alonso
Robotize a cleaning machine
Navigation system: teach&play approach with new motion
-
planning
algorithm due to special operational and platform requirements
Experiments to measure general
performance for cleaning tasks and to
measure the new motion planning
algorithm performance
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Long
-
term Experiment Using an Adaptive Appearance
-
based Map for
Visual Navigation by Mobile Robots
Feras Dayoub, Grzegorz Cielniak and Tom Duckett
{fdayoub,gcielniak,tduckett}@lincoln.ac.uk
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
An Implementation of a Biologically Inspired
Model of Head Direction Cells on a Robot

The influence of Visual, Vestibular
and
Kinesthetic
senses on Head
Direction cells.

Eight conditions are considered:

The model is tuned in order to
replicate biological observations in
seven of the conditions.

A conflict situation is finally presented.

Real robot data sets to train and
test the model.
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
CrunchBot

a mobile whiskered platform
Fox, Evans, Lepora, Pearson, Ham & Prescott
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Tactile Afferent Simulation from pressure Arrays
Rosana
Matuk
Herrera
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
The Interaction Between Vortices
and a Biomimetic Flexible Fin
T=345
T=695
T=0
T=190
T=525

Part of FILOSE EU project

aim to design and
develop robotic fish with artificial lateral line

Focus is on features available
to sense in the flow

The effect of a fin on the wake
behind a cylinder investigated

speed, strength and
trajectory of vortices
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Cooperative Box Pushing
Inspired by Human
Behaviour

Robots push a box that may
occlude the target

Human
-
inspired pushing strategy

Bullet Physics &
Enki
simulation

Performance increases with
number of robots (up to 64)

Preliminary test with
e
-
pucks
Jianing
Chen, Roderich Gross
The University of Sheffield
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Enhancing Self
-
similar Patterns
in Partially Connected Swarms
Controlling an ensemble of autonomous vehicles and make
them spontaneously arrange in self
-
similar patterns
0
20
40
60
80
100
120
0
20
40
60
80
100
120
nz = 1340
0
5
10
15
20
25
0
5
10
15
20
25
nz = 128
0
1
2
3
4
5
6
0
1
2
3
4
5
6
nz = 20
exp exp
ij ij
ij ij i j
i
j j
ij
ij ij ij ij
x x
Ca Cr x x
U
x la la lr lr
x
 
   


 
   
 
 
   

   
 
 
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Stephen Foster, Roderich Gross
The University of Sheffield

Spatial segregation

Brazil nut effect

NetLogo
particle
simulation

Nested structures

Symmetric and
asymmetric structures
Forming Nested 3D Structures
Based on the Brazil Nut Effect
Univ
.
Amsterdam
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Evolving Modularity in Robot Behaviour Using
Gene Expression Programming
Table 1: Comparison of average evaluations
(best result in bold)
Room
1
2
3
4
5
GP
3000
90000
20050
50000
14000
ugGEP
500
30616
19871
19425
15825
mgGEP
500
27611
21652
18809
19885
Fig 3: Comparison of ugGEP and mgGEP performances
across rooms 2
-
5
Experimental Results
Multigenic
GEP
Gene
Expression
Programming
is
a
relatively
recent
genotype/phenotype
evolutionary
algorithm
developed
in
2001
(Ferreira,
2001
)
.
GEP
individuals
can
be
formed
using
one
gene
(ugGEP)
or
they
can
be
implemented
using
multiple
genes
which
are
linker
together
using
a
special
function
called
a
linker
.
This
structure
is
referred
to
as
multigenic
GEP
(mgGEP)
.
Using
this
mgGEP,
behaviours
can
be
evolved
separately
in
each
gene(see
Figure
1
)
;
the
linker
can
then
be
used
to
decide
which
gene
controls
the
motor
output
at
any
given
time
.
In
essence
mgGEP
has
the
ability
to
evolve
modularity
akin
to
subsumption
or
layered
architecture
.
Fig 2 : rooms used in the experiment
Fig 1: Behaviour subdivision using mgGEP
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Instance
-
Based Reinforcement Learning Technique with a
Meta
-
Learning Mechanism for Robust Multi
-
Robot Systems
Toshiyuki Yasuda
,
Motohiro
Wada and Kazuhiro
Ohkura
(Hiroshima University, JAPAN)
Paper ID: 031
We have developed a RL technique called
Bayesian
-
discrimination
-
function
-
based
RL (BRL)

Continuous state space of BRL
Metaparameters are updated
This paper introduced an extended BRL by
providing a meta
-
learning mechanism that
adaptively modifies its learning parameters on the
basis of
stability of its own actions
.
The results of the physical experiments demonstrated
that our proposed method would improve the
robustness of an MRS.
denotes the
information entropy
of firing rules
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Toward an Ecological Approach to Interface Design for
Teaching Robots
Guillaume Doisy, Joachim Meyer, Yael Edan
EU FP7
-
INTRO project
Ben
-
Gurion University of the Neguev, Beer
-
Sheva, Israel
doisyg@bgu.ac.il

How should the human robot interface support online robot learning
for unknown new tasks in real world applications?

Limited possibility to test the learned behaviour

Difficulty for the user to understand the learning process

Risk of mismatch between user expectation and robot behaviour

Solution inspired by Gibson’s theory of visual perception:

Instead of trying to make the user develop an internal model of the robot learning
process, the interface should explicitly communicate
learning affordance
,
corresponding to possible learning or learned actions linked to environmental
constraints.
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
On the Analysis of Parameter Convergence for Temporal Difference
Learning of an Exemplar Balance Problem
Onder
Tutsoy
and Martin Brown
Despite
RL
is
being
accepted
as
a
complex
learning
algorithm,
a
simple
exemplar
test
problem
leads
to
analysing
its
properties
.
Finite
horizon
closed
form
solution
of
the
value
function
by
using
a
state
based
reward
for
a
first
order
unstable
plant
is
derived
.






*
0.1*
0 0
0
1 1
k
t t
k k
t
V x e x x x


   


0
0.5
1
1.5
2
-2.5
-2
-1.5
-1
-0.5
0
0.5
w
1
w
2
k=0
*
k=k
k=1
-
2
.
5
-
2
-
1
.
5
-
1
-
0
.
5
0
0
.
5
-
0
.
2
-
0
.
1
0
0
.
1
0
.
2
0
.
3
w
2
w
3
*
k -1
*
k
The
bias
and
linear
parameters
are
learnt
during
the
whole
batch
whereas
the
parameter
with
the
higher
order
is
learnt
just
at
the
end
of
the
horizon
.


0.1
0.1 0.1
1 1

1 1 1 1
k
e
V x
e e
 
   
 
 
 
 
 
   
 
 
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
+
-
JOIN
CONTROLLERS
JOIN SENSORS
τ
SINGLE RA
TE
CONTROLLER
A
T
LOW RA
TE
SINGLE RA
TE
CONTROLLER
A
T
HIGH RA
TE
T
DR-HOH
DR-VISUAL
SER
VOING CONTROLLER
IMAGE
ACQUISITION
N
v
T
IMAGE
PROCESSING
VISION SYSTEM
DR-ESTIMA
T
OR
ROBOT CONTROL
UNIT
CONTROL
SYSTEM
DESIRED
OBJECT
VIEW
MEASURED
OBJECT
VIEW
OBJECT
N
v
T
N
c
T
N
c
T
N
c
T
V
ISUAL
S
ERVOING
M
ULTI
-
RATE
U
SING
D
UAL
R
ATE
H
IGH
O
RDER
H
OLDS
0
5
10
15
20
25
30
-2
-1
0
1
2
Time (sec)
System output
CLOSED LOOP SINWAVE RESPONSE G(s)=1/s
2


Dual-Rate Approach
Single-Rate High freq.
Single-Rate Low freq.
SR
-
IBVS Controller:
SR Control
;
control
signal estimation
;
error
signal estimation
.
PID
Controller
:
SR
control
;
control
signal estimation
;
error
signal estimation
;
eDR
-
PID
controller
.

Multi
-
rate
estimation
and
control
based
in
DR
-
HOHs
is
introduced
to
improve
the
behavior
of
non
-
linear
systems
.

The
inclusion
of
DR
-
HOHs
provides
additional
degrees
of
freedom
changing
the
system
behavior
.

DR
-
HOHs
allow
separate
Controllers
depending
on
their
intrinsic
dynamics
.

To
validate
the
multi
-
rate
scheme
implementing
a
Visual
S
ervoing
(MR
-
VS)
application
for
an
industrial
robot
arm
.
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
S
MOOTH
K
INEMATIC
C
ONTROLLER
vs.
P
URE
-
P
URSUIT
F
OR
N
ON
-
H
OLONOMIC
V
EHICLES
DCC
PP
Settling
time (t
e
)


Overshoot
(
d
)

Robustness

Comfort
/Safety


Smooth kinematic controller is
introduced, mathematically
and algorithmically.

Double Continuous
-
Curvature
path (DCC) takes
into
account
Kinematic constraints.

To improve the performance a multi
-
rate control loop is made with a curvature control from a
planned curvature profile.

Benchmark evaluation is done to show
better performance than Pure
-
Pursuit methods (PP).
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Optimal path planning for
Nonholonomic
robotic systems
Nonholonomic
kinematics
can be represented by
a differential equation on a
Matrix Lie group SE(2)
.
Minimising a
cost function
of the form:
Simple Example Problem (Planar case): Motion planning for a wheeled robot
Wheeled Robot
Kinematic model
Mathematical representation
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Airborne ultrasonic position and velocity measurement
using two cycles of linear
-
period
-
modulated signal.
Shinya Saito
1
, Minoru Kuribayashi Kurosawa
1
, Yuichiro Orino
1
, and Shinnosuke Hirata
2
(
1
Tokyo Institute of Technology;
2
The University of Electro
-
communications)
Paper ID : 015
Recursive cross
-
correlation operation
of single
-
bit signals.
Experimental result
2D position and velocity
vector.
Two microphones
L
inear
-
P
eriod
-
M
odulated
(Chirp wave)
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
OCCUPANCY GRID
-
BASED SLAM USING A MOBILE ROBOT
WITH A RING OF EIGHT SONAR TRANSDUCERS
Lynxmotion
4WD platform
Sabertooth
LISY300AL Gyro
Senscomp6500
Arduino
Mega MCU
EasyRadio
Transceiver
George
Terzakis
,
Sanja
Dogramadzi
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Visual
-
Inertial Motion Priors for Robust Monocular SLAM
By Usman Qayyum
{usman.qayyum@anu.edu.au}
and
Jonghyuk
Kim
{jon.kim@anu.edu.au}
,
Figure 1
: SLAM results on ground vehicle
dataset
1
-
Localization
2
-
Environment mapping
1
-
SLAM
Problem

SLAM is a task of building:
Map { Environment around me ? } Location { Where am I ? }

Our work utilized visual
-
inertial motion priors to
built maps while keeping track of our location
with better Observability of map
-
scale.
2
-
Vision/Inertial
Approaches

GPS
availability
is
an
issue
in
cluttered,
indoor
or
space
exploration
.

Humans
use
visual
information
fused
with
vestibular
system
{inertial}
as
a
primary
sense
for
navigation

We focused our approach on Monocular Camera as they are more favoured, due to their availability, cost,
wide visual spectrum and less calibration complexity [as compare to stereo system] .
3
-
Visual
-
Inertial
Priors

Relative velocity information of
accelerometer is used to generate the translational magnitude
[body motion in metric space]

Angular orientation between the consecutive images is estimated by dense image correspondence
without calculating 3D point features.

The motion estimates from accelerometer are preferred over the visual only approaches to avoid the general motion model and l
ack
of
true scale map.
4
-
Drift
Free
Tracking

Visual
-
inertial motion priors are integrated to EKF
-
SLAM filter,
which maintains few 3D features
.

The
constraint imposed on camera location by using few 3D features
helps to minimize the position drift found initially in visual inertial
priors.
5
-
Summary

Visual
Inertial
priors
provides
robustness
to
EKF
-
SLAM
with
better
scale
Observability
of
map
{metric
space
manipulation}
and
filter
consistency
.
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Study of routing algorithms considering real time
restrictions using a connectivity function
Magali Arellano
-
Vázquez¹, Héctor Benítez
-
Pérez², Jorge L. Ortega
-
Arjona³
¹ Posgrado en ciencia e ingeniería de la computación, UNAM.
² Depto. de Ingeniería en sistemas computacionales y automatización. IIMAS
-
UNAM.
³ Facultad de Ciencias, UNAM
Goal
The
proposed
algorithm
does
not
attempt
to
reserve
a
channel
communication,
it
assures
that
the
message
arrives
to
its
destination
despite
frequent
changes
in
the
network’s
topology,
because
the
route
is
rebuilt
at
each
step
.
Overview
This
work
introduces
an
adaptive
routing
algorithm
based
on
a
connectivity
function
that
evaluates
the
state
of
the
node’s
task
scheduler
as
well
as
the
general
conditions
of
the
network
.
Connectivity Function
The
connectivity
function
assesses
the
status
of
the
current
node
and
the
connection
states
of
its
neighboring
nodes,
thus
obtaining
the
overall
state
of
the
system
through
local
data
Path Construction
1)
Calculating
node’s
availability
.
2)
Evaluating
connection
state
of
neighboring
nodes
.
3)
Calculating
the
connectivity
function
per
node
.
4)
Calculating
or
updating
the
adjacency
matrix
.
5)
Calculating
node
k
+
1
using
Floyd
-
Warshall’s
algorithm
.
12th Conference Towards Autonomous Robotic Systems
TAROS 2011
Using Sequences of Knots as a
Random
Search
C. A.
Pina
-
Garcia and
Dongbing Gu
A knot can be defined as
“a sequence of
moves creating
an
aesthetic structure or topology, where
its properties
are
preserved under continuous deformations
”.

Thus
, it is possible to emulate a
random search
behaviour,
using a set of steps that
represents a
knot.
However, a single knot is not
enough to
cover a specific
area, due to this lack of
coverage, we
suggest link
several
knots in
order
to increase
the searching
scope
.