IEEE Transactions on Haptics,
Год журнала:
2023,
Номер
17(3), С. 396 - 404
Опубликована: Дек. 25, 2023
Haptic
temporal
signal
recognition
plays
an
important
supporting
role
in
robot
perception.
This
paper
investigates
how
to
improve
classification
performance
on
multiple
types
of
haptic
datasets
using
a
Transformer
model
structure.
By
analyzing
the
feature
representation
signals,
Transformer-based
two-tower
structural
model,
called
Touchformer,
is
proposed
extract
and
spatial
features
separately
integrate
them
self-attention
mechanism
for
classification.
To
address
characteristics
small
sample
datasets,
data
augmentation
employed
stability
dataset.
Adaptations
overall
architecture
training
optimization
procedures
are
made
robustness
model.
Experimental
comparisons
three
publicly
available
demonstrate
that
Touchformer
significantly
outperforms
benchmark
indicating
our
approach's
effectiveness
providing
new
solution
Micromachines,
Год журнала:
2024,
Номер
15(12), С. 1513 - 1513
Опубликована: Дек. 20, 2024
Robotic
devices
with
integrated
tactile
sensors
can
accurately
perceive
the
contact
force,
pressure,
sliding,
and
other
information,
they
have
been
widely
used
in
various
fields,
including
human–robot
interaction,
dexterous
manipulation,
object
recognition.
To
address
challenges
associated
initial
value
drift,
to
improve
durability
accuracy
of
detection
for
a
robotic
hand,
this
study,
flexible
sensor
is
designed
high
repeatability
by
introducing
supporting
layer
pre-separation.
The
proposed
has
range
0–5
N
resolution
0.2
N,
error
as
relatively
small
1.5%.
In
addition,
response
time
under
loading
unloading
conditions
are
80
ms
160
ms,
respectively.
Moreover,
three-dimensional
force
decoupling
method
developed
distributing
units
on
non-coplanar
fingertip.
Finally,
using
backpropagation
neural
network,
classification
recognition
processes
nine
types
objects
different
shapes
categories
realized,
achieving
an
higher
than
95%.
results
show
that
sensing
system
could
be
beneficial
delicate
manipulation
hands.
Interdisciplinary materials,
Год журнала:
2024,
Номер
unknown
Опубликована: Дек. 30, 2024
ABSTRACT
Humanoid
robots
have
garnered
substantial
attention
recently
in
both
academia
and
industry.
These
are
becoming
increasingly
sophisticated
intelligent,
as
seen
health
care,
education,
customer
service,
logistics,
security,
space
exploration,
so
forth.
Central
to
these
technological
advancements
is
tactile
perception,
a
crucial
modality
through
which
humanoid
exchange
information
with
their
external
environment,
thereby
facilitating
human‐like
behaviors
such
object
recognition
dexterous
manipulation.
Texture
perception
particularly
vital
for
tasks,
the
surface
morphology
of
objects
significantly
influences
manipulation
abilities.
This
review
addresses
recent
progress
sensing
machine
learning
texture
robots.
We
first
examine
design
working
principles
sensors
employed
differentiating
between
touch‐based
sliding‐based
approaches.
Subsequently,
we
delve
into
algorithms
implemented
using
sensors.
Finally,
discuss
challenges
future
opportunities
this
evolving
field.
aims
provide
insights
state‐of‐the‐art
developments
foster
robotics.
Advanced Theory and Simulations,
Год журнала:
2023,
Номер
6(12)
Опубликована: Сен. 13, 2023
Abstract
Piezoelectric
tiles
harvest
mechanical
vibrations
and
convert
them
into
electrical
energy,
making
an
attractive
energy‐harvesting
technology.
However,
their
performance
is
heavily
influenced
by
the
terrain
where
they
are
installed.
Traditional
experimental
methods
for
predicting
on
different
terrains
time‐consuming,
so
a
computational
approach
necessary
to
improve
efficiency.
To
address
this,
machine
learning‐based
proposed
using
Artificial
Neural
network
(ANN)
Deep
neural
(DNN)
with
Tanh
activation
function
predict
piezoelectric
tile
in
diverse
terrains.
The
models
trained
dataset
consisting
of
four
terrains,
including
Flat
(FT)
Hilly
Terrain
(HT)
1,
2,
3
road
angles
0,
3,
6,
10
degrees.
A
finite
element
model
also
established
optimize
estimate
suitable
parameter
range
prevent
damage
during
experiments.
results
indicate
that
DNN
performs
better
than
ANN
model,
achieving
high
accuracy
These
findings
suggest
learning
can
provide
time
cost‐effective
way
varied
thereby
facilitating
betters
installation
maintenance
decisions
systems.
IEEE Transactions on Haptics,
Год журнала:
2023,
Номер
17(3), С. 396 - 404
Опубликована: Дек. 25, 2023
Haptic
temporal
signal
recognition
plays
an
important
supporting
role
in
robot
perception.
This
paper
investigates
how
to
improve
classification
performance
on
multiple
types
of
haptic
datasets
using
a
Transformer
model
structure.
By
analyzing
the
feature
representation
signals,
Transformer-based
two-tower
structural
model,
called
Touchformer,
is
proposed
extract
and
spatial
features
separately
integrate
them
self-attention
mechanism
for
classification.
To
address
characteristics
small
sample
datasets,
data
augmentation
employed
stability
dataset.
Adaptations
overall
architecture
training
optimization
procedures
are
made
robustness
model.
Experimental
comparisons
three
publicly
available
demonstrate
that
Touchformer
significantly
outperforms
benchmark
indicating
our
approach's
effectiveness
providing
new
solution
