|
4. Local path planning with unknown and dynamic obstacles
In dynamic and uncertain underwater environments, it is difficult
or even impossible to obtain the information of various obstacles
before path planning. Obviously, unknown and dynamic obstacles will
pose big safety issues for AUVs to perform underwater tasks. In this
case, there is usually a planned path from a global path planner and
AUV still needs a local path planner to avoid these unknown and
dynamic obstacles. In recent years, many path planning methods have
been proposed for unknown and dynamic obstacle avoidance, like
Rapidly-exploring Random Trees, Artificial Potential Field, Fuzzy Logic
Algorithm, Neural Network, Reinforcement Learning and even Deep
Reinforcement Learning. Different form other robot platforms on land
or in the air, these methods mostly use sonar sensors (
Nussbaum et al.
,
1996
;
Ghatak et al.
,
2006
;
Zou et al.
,
2007
) to obtain the real-time
distances and angles of surrounding obstacles, and use local dynamic
path planning technology to complete obstacle avoidance tasks, which
greatly improves the autonomy of AUV.
4.1. Rapidly-exploring Random Trees (RRT)
The fundamental idea behind the RRT algorithm is to scatter some
points randomly in the search space and then connect them into a
motion path for robot via calculation. The mechanism of a basic RRT
algorithm is shown in
Fig. 7
. In
Fig. 7
,
𝑥
𝑖𝑛𝑖𝑡
is the starting node and
𝑥
𝑟𝑎𝑛𝑑
is the target node. Through collision detection of random sampling
points in the state space, the nearest
𝑥
𝑛𝑒𝑎𝑟
node to the target node can
be found, which extends the
𝑥
𝑛𝑒𝑤
node to the open undetected area.
The RRT algorithm considers both the algebraic constraints caused by
obstacles and the differential constraints caused by dynamics of AUV
in Section
2 Tan et al.
(
2004
), and can effectively find a collision free
path from the starting node to the target node.
Tan et al.
(
2005
) combined RRT with a hybrid modeling technique –
Maneuver Automaton (MA) – to capture the key dynamics of a nonlin-
ear autonomous underwater vehicle (AUV) such as rudder deflection
and rudder rate. In the path planning with obstacles, their proposed
RRT algorithm increases the generation of sub nodes, which allows
AUV to complete the task safely with greater probability. To reduce
unnecessary space exploration,
Hernández et al.
(
2011
) proposed the
homotopy rapidly-exploring random trees (HRRT) algorithm, HRRT can
prevent the tree from growing out of the space region of homotopy class
and allow AUV to move towards the target point in different ways while
avoiding obstacles.
Hernández et al.
(
2015
) applied the transition
rapidly-exploring random trees (
Jaillet et al.
,
2010
) algorithm to AUV
path planning. In their method, the distance between seabed and AUV
is considered in the costmap of RRT algorithm. Their experimental
results show that AUV can complete corresponding underwater explo-
ration tasks near the seabed.
Li et al.
(
2017
) proposed a liveness-based
RRT (Li-RRT) by adding a liveness index to describe the effectiveness
of each node in the tree. In Li-RRT, a collision detection index, collision
degree, offspring collision and adjacent collision that will greatly affect
the expansion ability are also defined. Their numerical simulation
experimental results show that AUV can avoid collision with the coast
in underwater environments.
Yu et al.
(
2017
) proposed a smooth-RRT
algorithm for AUV motion planning in accordance with the steering
characteristics of AUV. Compared with the classical RRT, a convergence
factor and angle factor are introduced which can reduce the generation
of unnecessary nodes. In addition, the hydrodynamic angle is consid-
ered when AUV calculates the motion angle, and the greedy strategy
is used to make the path smoother. Simulated experimental results
show that AUV can reduce the number of turns in presence of many
obstacles.
Taheri et al.
(
2019
) proposed a closed-loop rapidly-exploring
random tree (CL-RRT) algorithm to solve the kinematic constraints
caused by obstacles and the characteristics of AUV. In CL-RRT, three
fuzzy proportional derivative controllers and AUV models in Section
2
were used to evaluate whether the range and vertices of the search tree
satisfy the nonholonomic dynamic constraints of AUV. It was verified
on xPC Target generator that CL-RRT could find collision-free paths in
3D space with clutter obstacles, but the experiment did not involve an
actual AUV.
𝑅𝑅𝑇
∗
is one of the sampling-based algorithms proposed
in recent years (
Karaman and Frazzoli
,
2011
). It not only ensures the
probability completeness, but also solves the problem that the RRT
algorithm might get a suboptimal solution. In addition,
𝑅𝑅𝑇
∗
changes
the expansion mode of nodes and selects the node with the lowest cost
in the extended domain as the parent node with a cost function.
Fu
et al.
(
2019
) improved the
𝑅𝑅𝑇
∗
algorithm using heuristic ellipsoid
subset sampling for centralized search. Their experimental results with
terrain obstacles and floating objects show that AUV can approach the
optimal path faster.
Do'stlaringiz bilan baham: | 
