Deep
reinforcement
learning
(DRL)
has
significantly
advanced
online
path
planning
for
unmanned
aerial
vehicles
(UAVs).
Nonetheless,
DRL-based
methods
often
encounter
reduced
performance
when
dealing
with
unfamiliar
scenarios.
This
decline
is
mainly
caused
by
the
presence
of
non-causal
and
domain-specific
elements
within
visual
representations,
which
negatively
impact
policies.
Present
techniques
generally
depend
on
predefined
augmentation
or
regularization
intended
to
direct
model
toward
identifying
causal
domain-invariant
components,
thereby
enhancing
model’s
ability
generalize.
However,
these
manually
crafted
approaches
are
intrinsically
constrained
in
their
coverage
do
not
consider
entire
spectrum
possible
scenarios,
resulting
less
effective
new
environments.
Unlike
prior
studies,
this
work
prioritizes
representation
presents
a
novel
method
disentanglement.
The
approach
successfully
distinguishes
between
data.
By
concentrating
aspects
during
policy
phase,
factors
minimized,
improving
generalizability
DRL
models.
Experimental
results
demonstrate
that
our
technique
achieves
reliable
navigation
collision
avoidance
unseen
surpassing
state-of-the-art
deep
algorithms.
Agriculture,
Год журнала:
2024,
Номер
14(8), С. 1294 - 1294
Опубликована: Авг. 5, 2024
With
the
global
population
growth
and
increasing
food
demand,
development
of
precision
agriculture
has
become
particularly
critical.
In
agriculture,
accurately
identifying
areas
nitrogen
stress
in
crops
planning
precise
fertilization
paths
are
crucial.
However,
traditional
coverage
path-planning
(CPP)
typically
considers
only
single-area
tasks
overlooks
multi-area
CPP.
To
address
this
problem,
study
proposed
a
Regional
Framework
for
Coverage
Path-Planning
Precision
Fertilization
(RFCPPF)
crop
protection
UAVs
tasks.
This
framework
includes
three
modules:
spatial
distribution
extraction,
environmental
map
construction,
path-planning.
Firstly,
Sentinel-2
remote-sensing
images
processed
using
Google
Earth
Engine
(GEE)
platform,
Green
Normalized
Difference
Vegetation
Index
(GNDVI)
is
calculated
to
extract
stress.
A
constructed
guide
multiple
UAV
agents.
Subsequently,
improvements
based
on
Double
Deep
Q
Network
(DDQN)
introduced,
incorporating
Long
Short-Term
Memory
(LSTM)
dueling
network
structures.
Additionally,
multi-objective
reward
function
state
action
selection
strategy
suitable
area
plant
operations
designed.
Simulation
experiments
verify
superiority
method
reducing
redundant
improving
efficiency.
The
improved
DDQN
achieved
an
overall
step
count
that
60.71%
MLP-DDQN
90.55%
Breadth-First
Search–Boustrophedon
Algorithm
(BFS-BA).
total
repeated
rate
was
reduced
by
7.06%
compared
8.82%
BFS-BA.
Drones,
Год журнала:
2025,
Номер
9(3), С. 203 - 203
Опубликована: Март 12, 2025
Obstacle
avoidance
is
crucial
for
the
successful
completion
of
UAV
missions.
Static
and
dynamic
obstacles,
such
as
trees,
buildings,
flying
birds,
or
other
UAVs,
can
threaten
these
As
a
result,
safe
path
planning
essential,
particularly
missions
involving
multiple
UAVs.
Collision-free
paths
be
designed
in
either
2D
3D
environments,
depending
on
scenario.
This
study
provides
an
overview
recent
advancements
obstacle
These
methods
are
compared
based
various
criteria,
including
techniques,
types,
environment
explored,
sensor
equipment,
map
statuses.
Additionally,
this
paper
includes
process
addressing
detection
reviews
evolution
(ODA)
techniques
UAVs
over
past
decade.
Drones,
Год журнала:
2024,
Номер
8(5), С. 173 - 173
Опубликована: Апрель 27, 2024
Unmanned
aerial
vehicles
(UAVs)
provide
benefits
through
eco-friendliness,
cost-effectiveness,
and
reduction
of
human
risk.
Deep
reinforcement
learning
(DRL)
is
widely
used
for
autonomous
UAV
navigation;
however,
current
techniques
often
oversimplify
the
environment
or
impose
movement
restrictions.
Additionally,
most
vision-based
systems
lack
precise
depth
perception,
while
range
finders
a
limited
environmental
overview,
LiDAR
energy-intensive.
To
address
these
challenges,
this
paper
proposes
VizNav,
modular
DRL-based
framework
navigation
in
dynamic
3D
environments
without
imposing
conventional
mobility
constraints.
VizNav
incorporates
Twin
Delayed
Deterministic
Policy
Gradient
(TD3)
algorithm
with
Prioritized
Experience
Replay
Importance
Sampling
(PER)
to
improve
performance
continuous
action
spaces
mitigate
overestimations.
employs
map
images
(DMIs)
enhance
visual
by
accurately
estimating
objects’
information,
thereby
improving
obstacle
avoidance.
Empirical
results
show
that
leveraging
TD3,
improves
navigation,
inclusion
PER
DMI
further
boosts
performance.
Furthermore,
deployment
across
various
experimental
settings
confirms
its
flexibility
adaptability.
The
framework’s
architecture
separates
agent’s
from
training
process,
facilitating
integration
DRL
algorithms,
simulation
environments,
reward
functions.
This
modularity
creates
potential
influence
RL
systems,
including
robotics
control
vehicles.
