The rapid development of artificial intelligence and deep learning provides new
ideas and new methods for unmanned autonomous navigation. Yize Chen et al. propose a data-driven control method based on deep neural networks, which greatly
improves the control efficiency and accuracy of complex systems [8]. The NVIDIA
AutoPilot team use a deep neural network to map the raw images obtained from the
front camera to the instructions for car autopilot, which achieve good results in simulators and road tests [9].
As described above, the artificial neural network is accessed to the control loop, to
control the ship motion directly. We propose a prediction model based on deep learning
and artificial neural network, which predicts the waypoint behavior effect and provides
assistance and reference for the maneuvering decision of the USV, instead of traditional
method in which the neural network is a part of ship motion control system. According
to the waypoint behavior, the deep neural network model is established, combined with
the navigation data samples of different speeds, steering angles, and other parameters of
the USV in the simulated environment of Mission-Oriented Operating Suite-Interval
Programming (MOOS-IvP). The deep learning method is used to train and solve the
problem of multi-dimensional vector space decision-making involved in the trajectory
track control process of the vehicle, making decisions for its autonomous navigation.
2 Background
2.1 The MOOS-IvP Architecture
MOOS-IvP is an open-source project aiming to provide a general autonomy architecture for autonomous system research. MOOS is a suite of software modules that
coordinate software processes running on an autonomous platform. IvP is a technique
for solving multi-objective optimization problems.
The MOOS-IvP software design is based on three architecture philosophies [10],
(a) the backseat driver paradigm, (b) publish and subscribe autonomy middleware, and
(c) behavior-based autonomy.
2.2 Waypoint Behavior
In the 1990s, Rodney Brooks proposed an inclusive structure that simplified the control
of agents into several task-based behaviors. Based on this, the autonomous control of
USV can be transformed into waypoint behavior, patrol behavior, fixed-point behavior,
and trailing behavior. Among them, the waypoint behavior is the most basic, and many
other behaviors are improved on this basis [3].
(1) Definition
The waypoint behavior means that the USV moves along a series of specified points in
the X-Y plane, controlled by trajectory tracking.
As shown in Fig. 1a, the connection between each two adjacent waypoints is the
shortest track, also known as the planned route. The USV sails autonomously
according to the waypoint behavior, that is, sailing along the planned route. With the
592
W. Sun and X. Gao
ideas and new methods for unmanned autonomous navigation. Yize Chen et al. propose a data-driven control method based on deep neural networks, which greatly
improves the control efficiency and accuracy of complex systems [8]. The NVIDIA
AutoPilot team use a deep neural network to map the raw images obtained from the
front camera to the instructions for car autopilot, which achieve good results in simulators and road tests [9].
As described above, the artificial neural network is accessed to the control loop, to
control the ship motion directly. We propose a prediction model based on deep learning
and artificial neural network, which predicts the waypoint behavior effect and provides
assistance and reference for the maneuvering decision of the USV, instead of traditional
method in which the neural network is a part of ship motion control system. According
to the waypoint behavior, the deep neural network model is established, combined with
the navigation data samples of different speeds, steering angles, and other parameters of
the USV in the simulated environment of Mission-Oriented Operating Suite-Interval
Programming (MOOS-IvP). The deep learning method is used to train and solve the
problem of multi-dimensional vector space decision-making involved in the trajectory
track control process of the vehicle, making decisions for its autonomous navigation.
2 Background
2.1 The MOOS-IvP Architecture
MOOS-IvP is an open-source project aiming to provide a general autonomy architecture for autonomous system research. MOOS is a suite of software modules that
coordinate software processes running on an autonomous platform. IvP is a technique
for solving multi-objective optimization problems.
The MOOS-IvP software design is based on three architecture philosophies [10],
(a) the backseat driver paradigm, (b) publish and subscribe autonomy middleware, and
(c) behavior-based autonomy.
2.2 Waypoint Behavior
In the 1990s, Rodney Brooks proposed an inclusive structure that simplified the control
of agents into several task-based behaviors. Based on this, the autonomous control of
USV can be transformed into waypoint behavior, patrol behavior, fixed-point behavior,
and trailing behavior. Among them, the waypoint behavior is the most basic, and many
other behaviors are improved on this basis [3].
(1) Definition
The waypoint behavior means that the USV moves along a series of specified points in
the X-Y plane, controlled by trajectory tracking.
As shown in Fig. 1a, the connection between each two adjacent waypoints is the
shortest track, also known as the planned route. The USV sails autonomously
according to the waypoint behavior, that is, sailing along the planned route. With the
592
W. Sun and X. Gao