Electronics,
Journal Year:
2024,
Volume and Issue:
13(10), P. 1819 - 1819
Published: May 8, 2024
This
research
explores
the
use
of
Q-Learning
for
real-time
swarm
(Q-RTS)
multi-agent
reinforcement
learning
(MARL)
algorithm
robotic
applications.
study
investigates
efficacy
Q-RTS
in
reducing
convergence
time
to
a
satisfactory
movement
policy
through
successful
implementation
four
and
eight
trained
agents.
has
been
shown
significantly
reduce
search
terms
training
iterations,
from
almost
million
iterations
with
one
agent
650,000
agents
500,000
The
scalability
was
addressed
by
testing
it
on
several
agents’
configurations.
A
central
focus
placed
design
sophisticated
reward
function,
considering
various
postures
their
critical
role
optimizing
Q-learning
algorithm.
Additionally,
this
delved
into
robustness
agents,
revealing
ability
adapt
dynamic
environmental
changes.
findings
have
broad
implications
improving
efficiency
adaptability
systems
applications
such
as
IoT
embedded
systems.
tested
implemented
using
Georgia
Tech
Robotarium
platform,
showing
its
feasibility
above-mentioned
Applied Artificial Intelligence,
Journal Year:
2024,
Volume and Issue:
38(1)
Published: Aug. 5, 2024
In
recent
years,
the
integration
of
intelligent
industrial
process
monitoring,
quality
prediction,
and
predictive
maintenance
solutions
has
garnered
significant
attention,
driven
by
rapid
advancements
in
digitalization,
data
analytics,
machine
learning.
As
traditional
production
systems
evolve
into
self-aware
self-learning
configurations,
capable
autonomously
adapting
to
dynamic
environmental
conditions,
significance
reinforcement
learning
becomes
increasingly
apparent.
This
paper
provides
an
overview
developments
applications
manufacturing
industry.
Various
sectors
within
manufacturing,
including
robot
automation,
welding
processes,
semiconductor
industry,
injection
molding,
metal
forming,
milling
power
are
explored
for
instances
application.
The
analysis
focuses
on
application
types,
problem
modeling,
training
algorithms,
validation
methods,
deployment
statuses.
Key
benefits
these
identified.
Particular
emphasis
is
placed
elucidating
primary
obstacles
impeding
adoption
implementation
technology
settings,
such
as
model
complexity,
accessibility
simulation
environments,
safety
constraints,
interpretability.
concludes
proposing
potential
alternatives
avenues
future
research
address
challenges,
improving
sample
efficiency
bridging
simulation-to-reality
gap.
IEEE Robotics and Automation Letters,
Journal Year:
2023,
Volume and Issue:
8(8), P. 4418 - 4425
Published: June 5, 2023
Active
tracking
of
space
noncooperative
object
that
merely
relies
on
vision
camera
is
greatly
significant
for
autonomous
rendezvous
and
debris
removal.
Considering
its
Partial
Observable
Markov
Decision
Process
(POMDP)
property,
this
letter
proposes
a
novel
deep
recurrent
neural
network
architecture,
named
as
attention
module
based
active
visual
(RAMAVT),
incorporating
Multi-Head
Attention
(MHA)
Squeeze-and-Excitation
(SE)
layer
remarkably
improve
the
representative
ability
with
almost
no
extra
computational
cost.
It
has
been
successfully
applied
to
value-based
policy
gradient-based
reinforcement
learning
algorithm,
learned
drive
chasing
spacecraft
follow
arbitrary
high-frequency
near-optimal
velocity
control
commands.
Extensive
experiments
robustness
evaluations
implemented
non-cooperative
(SNCOAT)
benchmark
show
betterment
our
method
compared
other
state-of-the-art
trackers.
In
addition,
we
make
further
ablation
study
interpretability
research
RAMAVT
which
validity
rationality
have
demonstrated.
Micromachines,
Journal Year:
2024,
Volume and Issue:
15(1), P. 112 - 112
Published: Jan. 9, 2024
Microrobotics
has
opened
new
horizons
for
various
applications,
especially
in
medicine.
However,
it
also
witnessed
challenges
achieving
maximum
optimal
performance.
One
key
challenge
is
the
intelligent,
autonomous,
and
precise
navigation
control
of
microrobots
fluid
environments.
The
intelligence
autonomy
microrobot
control,
without
need
prior
knowledge
entire
system,
can
offer
significant
opportunities
scenarios
where
their
models
are
unavailable.
In
this
study,
two
systems
based
on
model-free
deep
reinforcement
learning
were
implemented
to
movement
a
disk-shaped
magnetic
real-world
environment.
training
results
an
off-policy
SAC
algorithm
on-policy
TRPO
revealed
that
successfully
learned
path
reach
random
target
positions.
During
training,
exhibited
higher
sample
efficiency
greater
stability.
showed
100%
97.5%
success
rates
reaching
targets
evaluation
phase,
respectively.
These
findings
basic
insights
into
intelligent
autonomous
advance
capabilities
applications.
Electronics,
Journal Year:
2024,
Volume and Issue:
13(10), P. 1819 - 1819
Published: May 8, 2024
This
research
explores
the
use
of
Q-Learning
for
real-time
swarm
(Q-RTS)
multi-agent
reinforcement
learning
(MARL)
algorithm
robotic
applications.
study
investigates
efficacy
Q-RTS
in
reducing
convergence
time
to
a
satisfactory
movement
policy
through
successful
implementation
four
and
eight
trained
agents.
has
been
shown
significantly
reduce
search
terms
training
iterations,
from
almost
million
iterations
with
one
agent
650,000
agents
500,000
The
scalability
was
addressed
by
testing
it
on
several
agents’
configurations.
A
central
focus
placed
design
sophisticated
reward
function,
considering
various
postures
their
critical
role
optimizing
Q-learning
algorithm.
Additionally,
this
delved
into
robustness
agents,
revealing
ability
adapt
dynamic
environmental
changes.
findings
have
broad
implications
improving
efficiency
adaptability
systems
applications
such
as
IoT
embedded
systems.
tested
implemented
using
Georgia
Tech
Robotarium
platform,
showing
its
feasibility
above-mentioned
