2018 Winter Simulation Conference (WSC),
Год журнала:
2023,
Номер
unknown, С. 2710 - 2721
Опубликована: Дек. 10, 2023
This
study
introduces
a
deep
learning-based
method
for
indoor
3D
object
detection
and
localization
in
healthcare
facilities.
incorporates
spatial
channel
attention
mechanisms
into
the
YOLOv5
architecture,
ensuring
balance
between
accuracy
computational
efficiency.
The
network
achieves
an
AP50
of
67.6%,
mAP
46.7%,
real-time
rate
with
FPS
67.
Moreover,
proposes
novel
mechanism
estimating
coordinates
detected
objects
projecting
them
onto
maps,
average
error
0.24
m
0.28
x
y
directions,
respectively.
After
being
tested
validated
real-world
data
from
university
campus,
proposed
shows
promise
improving
disinfection
efficiency
facilities
by
enabling
robot
navigation.
Integrated Computer-Aided Engineering,
Год журнала:
2023,
Номер
30(3), С. 293 - 309
Опубликована: Март 21, 2023
Video
feeds
from
traffic
cameras
can
be
useful
for
many
purposes,
the
most
critical
of
which
are
related
to
monitoring
road
safety.
Vehicle
trajectory
is
a
key
element
in
dangerous
behavior
and
accidents.
In
this
respect,
it
crucial
detect
those
anomalous
vehicle
trajectories,
that
is,
trajectories
depart
usual
paths.
work,
model
proposed
automatically
address
by
using
video
sequences
cameras.
The
proposal
detects
vehicles
frame
frame,
tracks
their
across
frames,
estimates
velocity
vectors,
compares
them
vectors
other
spatially
adjacent
trajectories.
From
comparison
very
different
(anomalous)
neighboring
detected.
practical
terms,
strategy
wrong-way
Some
components
off-the-shelf,
such
as
detection
provided
recent
deep
learning
approaches;
however,
several
options
considered
analyzed
tracking.
performance
system
has
been
tested
with
wide
range
real
synthetic
videos.
Journal of Field Robotics,
Год журнала:
2025,
Номер
unknown
Опубликована: Май 8, 2025
ABSTRACT
The
use
of
ultraviolet
(UV‐C)
disinfection
robots
has
become
increasingly
popular
in
diverse
settings,
including
hospitals,
schools,
public
transportation,
and
high‐traffic
areas,
especially
following
the
COVID‐19
pandemic.
These
offer
potential
to
enhance
efficiency
reduce
human
exposure
microorganisms.
However,
application
UV‐C
light
for
is
not
without
challenges.
challenges
include
need
precise
environmental
mapping,
accurate
dose
delivery,
mitigation
safety
risks
associated
with
humans
animals.
This
systematic
review
aims
examine
current
development
robots,
identify
key
technological
challenges,
explore
methods
used
ensure
effective
safe
disinfection.
An
automated
search
was
conducted
Scopus,
IEEE
Xplore,
ACM
Digital
Library,
SpringerLink
studies
published
up
July
2023,
followed
by
snowballing
gather
additional
relevant
works.
A
total
96
were
reviewed.
majority
these
either
did
address
correct
UVGI
or
lacked
appropriate
delivery.
Additionally,
positioning
lamps
often
done
subjectively,
most
incorporate
any
measures
prevent
accidents
related
exposure.
Based
on
this
analysis,
a
new
classification
proposed,
highlighting
advancements
readiness
levels.
Despite
progress
made
field,
significant
remain
developing
that
deliver
doses
while
ensuring
operational
efficiency.
emphasizes
further
research
gaps,
particularly
concerning
navigation
algorithms,
accuracy,
measures.
Computer-Aided Civil and Infrastructure Engineering,
Год журнала:
2024,
Номер
unknown
Опубликована: Дек. 13, 2024
Abstract
Construction
robots
are
challenging
the
paradigm
of
labor‐intensive
construction
tasks.
Imitation
learning
(IL)
offers
a
promising
approach,
enabling
to
mimic
expert
actions.
However,
obtaining
high‐quality
demonstrations
is
major
bottleneck
in
this
process
as
teleoperated
robot
motions
may
not
align
with
optimal
kinematic
behavior.
In
paper,
two
innovations
have
been
proposed.
First,
traditional
control
using
controllers
has
replaced
vision‐based
hand
gesture
for
intuitive
demonstration
collection.
Second,
novel
method
that
integrates
both
and
simple
environmental
rewards
proposed
strike
balance
between
imitation
exploration.
To
achieve
goal,
two‐step
training
first
step,
an
collection
platform
virtual
reality
utilized.
algorithm
used
train
policy
Experimental
results
demonstrate
combining
IL
can
significantly
accelerate
training,
even
limited
data.
Integrated Computer-Aided Engineering,
Год журнала:
2023,
Номер
31(2), С. 139 - 156
Опубликована: Дек. 19, 2023
In
this
paper,
we
propose
a
novel
Reinforcement
Learning
(RL)
algorithm
for
robotic
motion
control,
that
is,
constrained
Deep
Deterministic
Policy
Gradient
(DDPG)
deviation
learning
strategy
to
assist
biped
robots
in
walking
safely
and
accurately.
The
previous
research
on
topic
highlighted
the
limitations
controller’s
ability
accurately
track
foot
placement
discrete
terrains
lack
of
consideration
safety
concerns.
study,
address
these
challenges
by
focusing
ensuring
overall
system’s
safety.
To
begin
with,
tackle
inverse
kinematics
problem
introducing
constraints
damping
least
squares
method.
This
enhancement
not
only
addresses
singularity
issues
but
also
guarantees
safe
ranges
joint
angles,
thus
stability
reliability
system.
Based
this,
adoption
DDPG
method
correct
controller
deviations.
DDPG,
incorporate
constraint
layer
into
Actor
network,
incorporating
deviations
as
state
inputs.
By
conducting
offline
training
within
range
it
serves
corrector.
Lastly,
validate
effectiveness
our
proposed
approach
dynamic
simulations
using
CRANE
robot.
Through
comprehensive
assessments,
including
analysis,
evaluation,
experiments
terrains,
demonstrate
superiority
practicality
enhancing
performance
while
Overall,
contributes
advancement
robot
locomotion
addressing
gait
optimisation
from
multiple
perspectives,
handling,
constraints,
learning.
Concrete
cracking
in
bridges
significantly
endangers
their
safety
and
integrity.
Traditional
crack
detection
methods,
reliant
on
human
visual
inspection,
are
labor-intensive
prone
to
errors.
This
paper
introduces
a
unique
framework
for
bridge
integration
with
building
information
models
(BIM),
trialed
423-ft
Atlanta,
Georgia.
The
comprises
two
main
stages:
(1)
creating
BIM
model
using
drone-captured
images
structure
from
motion
(SFM)
photogrammetry,
(2)
utilizing
deep
learning-based
encoder-decoder
network
segment
cracks
orthomosaic
superimpose
these
segmented
onto
the
model.
suggested
method
showed
robust
performance,
achieving
mean
intersection
over
union
(mIoU)
of
0.787,
precision
0.751,
recall
0.742.
These
results
underline
potential
proposed
improve
efficiency
inspection
processes.
Authorea (Authorea),
Год журнала:
2024,
Номер
unknown
Опубликована: Июнь 6, 2024
Several
room
disinfection
robots
using
Ultraviolet
C
(UV-C)
light
have
emerged
recently,
especially
with
the
COVID-19
outbreak.
This
systematic
review
aims
to
identify
current
development
status
of
Germicidal
Irradiation
(UVGI)
robots,
their
limitations,
and
technologies
they
use.
An
automated
search
was
performed
on
Scopus,
ACM
Digital
Library,
IEEE
Xplore,
SpringerLink
platforms
for
papers
up
July
2023,
followed
by
a
snowballing
search.
Were
found
96
studies,
which
majority
were
not
concerned
dose
UVGI
applied
or
did
implement
any
technique
deliver
appropriate
dose;
positioning
lamps
carried
out
subjectively;
most
works
prevent
accidents
UVGI.
From
analysis
these
it
possible
propose
novel
classification
different
types
based
technological
readiness
levels.
The
data
shows
that
despite
recent
advances,
is
still
early,
many
advances
be
made.
2018 Winter Simulation Conference (WSC),
Год журнала:
2022,
Номер
unknown, С. 2440 - 2450
Опубликована: Дек. 11, 2022
Accurate
mapping
of
urban
subsurface
is
essential
for
managing
underground
infrastructure
and
preventing
excavation
accidents.
Ground-penetrating
radar
(GPR)
a
non-destructive
test
method
that
has
been
used
extensively
to
locate
utilities.
However,
existing
approaches
are
not
able
retrieve
detailed
utility
information
(e.g.,
material
dimensions)
from
GPR
scans.
This
research
aims
automatically
detect
characterize
buried
utilities
with
location,
dimension,
by
processing
To
achieve
this
aim,
inverting
data
based
on
deep
learning
developed
directly
reconstruct
the
permittivity
maps
cross-sectional
profiles
structure
corresponding
A
large
number
synthetic
scans
ground-truth
labels
were
generated
train
inversion
network.
The
experiment
results
indicated
proposed
achieved
Mean
Absolute
Error
0.53,
Structural
Similarity
Index
Measure
0.91,
an
$R^{2}$
0.96.
2018 Winter Simulation Conference (WSC),
Год журнала:
2023,
Номер
unknown, С. 2710 - 2721
Опубликована: Дек. 10, 2023
This
study
introduces
a
deep
learning-based
method
for
indoor
3D
object
detection
and
localization
in
healthcare
facilities.
incorporates
spatial
channel
attention
mechanisms
into
the
YOLOv5
architecture,
ensuring
balance
between
accuracy
computational
efficiency.
The
network
achieves
an
AP50
of
67.6%,
mAP
46.7%,
real-time
rate
with
FPS
67.
Moreover,
proposes
novel
mechanism
estimating
coordinates
detected
objects
projecting
them
onto
maps,
average
error
0.24
m
0.28
x
y
directions,
respectively.
After
being
tested
validated
real-world
data
from
university
campus,
proposed
shows
promise
improving
disinfection
efficiency
facilities
by
enabling
robot
navigation.