Third International Conference on Intelligent Computing and Human-Computer Interaction (ICHCI 2022),
Journal Year:
2023,
Volume and Issue:
unknown
Published: Jan. 13, 2023
Various
emergencies
occur
frequently,
posing
threats
and
challenges
to
people’s
lives
social
security.
In
consequence,
the
evacuation
of
multi-Agent
has
become
a
significant
part
emergency
response
process.
However,
few
existing
works
only
focus
on
small
number
agents,
which
does
not
consider
problem
cooperation
caused
by
increase
agents
impact
emergencies.
Therefore,
framework
for
event-driven
is
proposed
in
this
paper,
includes
three
parts:
event
collection,
sending,
task
execution.
During
execution,
are
divided
into
groups
select
leader
group,
while
other
group
move
with
leader.
Then,
reinforcement
learning
algorithm
Space
Multi-Agent
Deep
Deterministic
Policy
Gradient
(SMADDPG),
used
path
planning.
addition,
state,
action
reward
based
Markov
game
designed,
an
environment
presented
as
scenario.
The
experiment
results
show
that
method
can
shorten
length
path,
improve
interoperability
between
when
occur,
provide
decision-making
reference
departments
formulate
plans.
Chinese Journal of Aeronautics,
Journal Year:
2023,
Volume and Issue:
37(3), P. 237 - 257
Published: Oct. 16, 2023
In
some
military
application
scenarios,
Unmanned
Aerial
Vehicles
(UAVs)
need
to
perform
missions
with
the
assistance
of
on-board
cameras
when
radar
is
not
available
and
communication
interrupted,
which
brings
challenges
for
UAV
autonomous
navigation
collision
avoidance.
this
paper,
an
improved
deep-reinforcement-learning
algorithm,
Deep
Q-Network
a
Faster
R-CNN
model
Data
Deposit
Mechanism
(FRDDM-DQN),
proposed.
A
(FR)
introduced
optimized
obtain
ability
extract
obstacle
information
from
images,
new
replay
memory
(DDM)
designed
train
agent
better
performance.
During
training,
two-part
training
approach
used
reduce
time
spent
on
as
well
retraining
scenario
changes.
order
verify
performance
proposed
method,
series
experiments,
including
test
typical
episodes
conducted
in
3D
simulation
environment.
Experimental
results
show
that
trained
by
FRDDM-DQN
has
navigate
autonomously
avoid
collisions,
performs
compared
FR-DQN,
FR-DDQN,
FR-Dueling
DQN,
YOLO-based
YDDM-DQN,
original
FR
output-based
FR-ODQN.
Frontiers in Neuroinformatics,
Journal Year:
2023,
Volume and Issue:
17
Published: Jan. 23, 2023
Aiming
at
the
poor
robustness
and
adaptability
of
traditional
control
methods
for
different
situations,
deep
deterministic
policy
gradient
(DDPG)
algorithm
is
improved
by
designing
a
hybrid
function
that
includes
rewards
superimposed
on
each
other.
In
addition,
experience
replay
mechanism
DDPG
also
combining
priority
sampling
uniform
to
accelerate
DDPG's
convergence.
Finally,
it
verified
in
simulation
environment
can
achieve
accurate
robot
arm
motion.
The
experimental
results
show
converge
shorter
time,
average
success
rate
robotic
end-reaching
task
as
high
91.27%.
Compared
with
original
algorithm,
has
more
robust
environmental
adaptability.
IEEE Sensors Journal,
Journal Year:
2023,
Volume and Issue:
23(8), P. 8923 - 8931
Published: March 15, 2023
Crowd
congestion
is
an
important
factor
affecting
evacuation
efficiency,
and
reasonable
regulation
of
crowd
in
the
process
one
way
to
improve
efficiency.
Since
traditional
simulation
methods
are
based
on
hypothetical
scenarios
rules,
simulations
lack
realism.
To
solve
this
problem
reduce
scale
during
evacuation,
article
proposes
a
control
method
sensors
knowledge
graph.
First,
we
model
scenario
by
extracting
information
from
real
scenes
video
realism
simulation.
Second,
construct
graph
(CCKG)
represent
scenes,
which
improves
model's
ability
characterize
information.
Then,
gravity
field
generated
graph,
guided
for
evacuation.
Finally,
use
constructed
CCKG
predict
next
moment
regulate
route
real-time
congestion.
The
experimental
results
show
that
using
extract
data
can
At
same
time,
introduction
characterization
prediction
effect
model,
circumvent
large-scale
congestion,
efficiency