Skip to Content
Applied SciencesApplied Sciences
  • Article
  • Open Access

23 December 2024

20 Pages

Distributed Formation Control for Underactuated, Unmanned Surface Vehicles with Uncertainties and Disturbances

,
,
,
and
1
School of Marine Science and Technology, Northwestern Polytechnical University, Xi’an 710072, China
2
Marine Design & Research Institute of China, Shanghai 200011, China
3
National Key Laboratory of Autonomous Marine Vehicle Technology, Harbin Engineering University, Harbin 150001, China
*
Author to whom correspondence should be addressed.

Abstract

This paper investigates the distributed formation control problem of underactuated unmanned surface vehicles (UUSVs) with uncertainties and disturbances and proposes a novel distributed formation controller. The proposed controller redefines the dynamic and kinematic models for each UUSV, which reduces the complexity of the underactuated controller design. Dynamic surface control (DSC) is employed to eliminate the repeated derivatives of the virtual control law, which is crucial for the generation of real-time control signals. The proposed controller integrates neural network approximation with MLP-based adaptive laws to enhance the model’s resistance to disturbances. Then, an auxiliary adaptive law is designed for each UUSV to obtain a continuous controller under the compensation of approximate errors and disturbances. The results demonstrate that the controller achieves the desired goals for the formation control, and all control signals are guaranteed to be semi-global uniformly ultimately bounded (SGUUB). The final simulation results thoroughly prove the effectiveness of the theoretical results.

1. Introduction

With the development of ocean engineering technology and the expansion of mission requirements in both military and civilian fields, the control of unmanned surface vehicle (USV) formation has received widespread attention in recent years. The cooperation of multiple USVs offers several advantages, including greater flexibility, adaptability, and higher performance. For instance, in maritime search and rescue operations, multiple USVs can achieve extensive coverage of the target area, completing the entire mission in less time. Other applications include marine mineral exploration, territorial sea monitoring, the escorting of vehicles, fire support, and anti-submarine operations [1,2,3,4,5,6,7,8].
For multi-vehicle formation cooperative systems, Li et al. [9] provided a novel adaptive fuzzy output regulation method to address the multi-vehicle formation problem. Firstly, a distributed observer was employed to obtain the feedforward information of the reference system, and, then, within the backstepping framework, a distributed adaptive fuzzy control law for USV formation was designed. Based on neurodynamical optimization and fuzzy approximation techniques, Peng et al. [10] have not only achieved formation keeping for multi-vehicle formation but also enabled flexible formation maneuvers. Peng et al. [11] designed an adaptive dynamic surface multi-vehicle formation control scheme for USV clusters with uncertain dynamics. Subsequently, they developed a neural network formation controller for USV clusters based on graph theory formation structures, which can handle two types of cooperative formation problems for USVs with uncertain dynamics, both in leaderless and leader-based scenarios [12]. He and Zhao et al. [6,13] studied the control problem of a group of fully actuated USVs with a decentralized leader–follower structure. He et al. [14] improved the formation control design of a multi-vehicle radial basis function (RBF) neural network by introducing a robust feedback term in the feedback control to compensate for the neural network approximation error and external disturbance, thus obtaining the asymptotic tracking performance of USV formation. Chen et al. [15] proposed an adaptive cooperative control law for USVs based on the adaptive backstepping method to estimate the reference orbiting velocity in bidirectional communication topology, thus eliminating the limitation that every vehicle must obtain the reference trajectory in the existing literature. Aiming at the parametric uncertainty of the USV model, a new adaptive fault-tolerant formation control scheme for USVs with a leader–follower structure was designed, which is suitable for the control of UUSV clusters with a partially known input gain function [16]. Riahifard et al. [17] provided a multi-vehicle adaptive formation control, which can solve the leader–follower formation problem of UUSV clusters with uncertain dynamics and input constraints. Qu et al. [18] proposed a nonlinear positioning control for the UUSV, which resists unknown disturbance through adaptive heading control. Karnani et al. [19] considered UUSV trajectory tracking, provided an optimal control strategy for the vehicle in the presence of external disturbance, and proposed a controller-based model reference adaptive control scheme with an integrator, ensuring the stability of closed-loop systems. Mu et al. [20] considered the trajectory tracking control problem of UUSVs affected by uncertain dynamics and input saturation. This paper uses first-order and second-order sliding mode surfaces to design an underactuated trajectory tracking controller. The auxiliary system solves the potential input saturation problem, uses the MLP method of the neural network to further reduce the computational burden of the controller, and further enhances the robustness of the trajectory tracking system by introducing an adaptive robust term.
On this basis, we further studied the distributed formation control problem of UUSVs with uncertainties and disturbances, in which UUSVs can communicate with each other under directed interaction topology. For each UUSV, a new distributed neural network formation controller with two adaptive laws is proposed. First, a virtual control law is established based on static consensus protocol and directed graph theory. Then, the formation controller is derived using DSC, and, then, the adaptive law based on MLP is used to deal with uncertain dynamics. When compensating for approximation errors and external disturbances, auxiliary control laws and corresponding adaptive laws are added to obtain a continuous controller.
The main contributions of this paper are summarized as follows:
(1)
For the underactuated problem, the dynamic and kinematic models of the formation were transformed, redefined, and decomposed into the attitude subsystem and position subsystem, thereby achieving independent control of each degree of freedom of UUSVs.
(2)
For the complex differential explosion problem caused by the virtual control law in the calculation process, DSC technology was adopted in the controller design to avoid the repeated differentiation of the virtual control law and reduce the computational complexity. At the same time, combining RBF neural network approximation with MLP-based adaptive law enhances the model’s anti-disturbance ability.
The rest of this paper is organized as follows: Section 2 introduces the basic knowledge of graph theory and RBF neural networks. Section 3 introduces the design process of the cooperative formation controller. Section 4 conducts simulation. Section 5 summarizes this paper.
Notations: R N represents the N -dimensional real number space, given the vector x = x 1 , … , x N T , x 1 represents the 1-norm of x , x 2 represents the 2-norm of x , and x ∞ represents the ∞ -norm of x . Define tanh x = tanh x 1 , ⋯ , tanh x N T . Define vector h = h 1 T , … , h N T T , where h i ,  i = 1 , … , N is also a vector. Define h ∘ h = h 1 T h 1 , … , h N T h N T . I is a column vector with all elements being 1. A T represents the transposition of matrix A . λ A represents the eigenvalues of matrix A , A = diag A 11 , … , A N N represents the extraction of diagonal elements in matrix A . A ⊗ B represents the Kronecker product of matrices A and B .

3. Controller Design

The adaptive controller design of multi-UUSVs is divided into two parts. First, the attitude subsystem is used to design the control law τ i σ for each UUSV, and, then, the position subsystem is used to design the control law τ i X for each UUSV. The specific design process is below.

3.1. Attitude Controller Design

For the attitude subsystem, define the error variables z i 1 , z i 2 , z i 3 as follows:
z i 1 = θ i − θ j z i 2 = α ¯ i 1 − α i 1 z i 3 = r i − α ¯ i 1
where α i 1 is a virtual control law. α ¯ i 1 is a filter control law. α ¯ i 1 is obtained by α i 1 through a first-order filter. Time constant γ i is given by the corresponding diagonal element and is a diagonal positive definite matrix.
γ i α ¯ ˙ i 1 + α ¯ i 1 = α i 1
The derivative of the error variable z i 1 , z i 2 , z i 3 can be obtained as follows:
z ˙ i 1 = z i 2 + z i 3 + α i 1 − r ˙ j z ˙ i 2 = − γ i − 1 z i 2 − α ˙ i 1 z ˙ i 3 = 1 m i 33 τ i σ + f ¯ i σ + γ i − 1 z i 2
In this paper, DSC is adopted to deal with the complex differential problem in the calculation process of virtual control law α i 1 , which further improves the practicability of the controller.
Firstly, a Lyapunov function is selected as follows:
V 1 = 1 2 z 1 T z 1
The derivative of V 1 with respect to time is
V ˙ 1 = z 1 T z 2 + z 3 + α 1 − r ˙ j
Construct the virtual control law as
α i 1 = − K i 1 z i 1 − K i 0 tanh N z i 1 ε
where K i 1 , K i 0 ∈ R 3 × 3 is a diagonal matrix. ε is a variable positive constant. Definition z 1 = z 11 , … , z N 1 T , z 2 = z 12 , … , z N 2 T , z 3 = z 13 , … , z N 3 T , K 0 = diag K i 0 , K 1 = diag K i 1 , and γ = diag γ i , α 1 = α i 1 , … , α N 1 T .
Therefore, the virtual control law α i 1 can be written in the following form:
α 1 = − K 1 z 1 − K 0 tanh N z 1 ε
Substituting (21) into (19),
V ˙ 1 = z 1 T z 2 + z 3 − z 1 T K 1 z 1 − K 0 z 1 T tanh N z 1 ε − z 1 T r ˙ j
According to Lemma 3, it can be obtained as follows:
V ˙ 1 ≤ z 1 T z 2 + z 3 − z 1 T K 1 z 1 − K 0 z 1 T tanh N z 1 ε + z 1 ∞ N ρ M ≤ z 1 T z 2 + z 3 − z 1 T K 1 z 1 − K 0 z 1 T tanh N z 1 ε + N ρ M k p ε + N 2 ρ M z 1 T tan h ( N z 1 ε )
Let λ K 0 min > N 2 ρ M , and we can further conclude that N 2 ρ M z 1 T tan h ( N z 1 ε ) − K 0 z 1 T tanh N z 1 ε < 0 . Then, the derivative of V 1 satisfies
V ˙ 1 ≤ − z 1 T K 1 z 1 + z 1 T z 2 + z 1 T z 3 + N ρ M k p ε
Next, the attitude control law τ i σ of each UUSV is designed. A Lyapunov function is selected as follows:
V 2 = V 1 + 1 2 z 2 T z 2 + 1 2 z 3 T z 3
The derivative of V 2 with respect to time is
V ˙ 2 = V ˙ 1 + z 2 T z ˙ 2 + z 3 T z ˙ 3
According to (16), z ˙ 2 can be represented as follows:
z ˙ 2 = − γ − 1 z 2 + K 1 z ˙ 1 + K 0 1 − tanh T N z 1 ε tanh N z 1 ε N z ˙ 1 ε
Let B = K 1 z ˙ 1 + K 0 1 − tanh T N z 1 ε tanh N z 1 ε N z ˙ 1 ε , B consist of bounded functions, so there is a positive constant B ¯ and B 2 ≤ B ¯ .
Equation (25) can be written as
V ˙ 2 = V ˙ 1 + z 2 T − γ − 1 z 2 + B + z 3 T z ˙ 3 = − z 1 T K 1 z 1 − z 2 T γ − 1 z 2 + z 1 T z 2 + z 2 T B + z 3 T z 1 + z ˙ 3 + N ρ M k p ε
According to Lemma 4, the following inequalities can be obtained:
z 1 T z 2 ≤ 1 2 ( ε z 1 T z 1 + 1 ε z 2 T z 2 )
z 2 T B ≤ 1 2 ( 1 ε z 2 T B T B z 2 + ε )
V ˙ 2 can be written as
V ˙ 2 ≤ − K 1 − 1 2 ε I z 1 T z 1 − z 2 T γ − 1 − 1 2 ε 1 + B T B I z 2 + 0.5 + N ρ M k p ε + z 3 T z 1 + z ˙ 3
With B 1 = z 1 + z ˙ 3 , f i ( · ) = f ¯ i σ + γ i − 1 z i 2 , the following equation can be obtained:
z 3 T B 1 = ∑ i = 1 N z i 3 T z i 1 + 1 m i 33 τ i σ + f i ( · )
f ¯ i σ is difficult to be accurate. Thus, the RBF neural network is used to process it.
f i ( · ) = W i ∘ h i + ε i
where W i = W i 1 T , W i 2 T , W i 3 T T , h i = h i 1 T , h i 2 T , h i 3 T T , and ε i ≤ ε N , i = 1 , … , N , ε N > 0 .
V ˙ 2 can be written as
V ˙ 2 ≤ − K 1 − 1 2 ε z 1 T z 1 − z 2 T γ − 1 − 1 2 ε 1 + B T B I z 2 + 0.5 + N ρ M k p ε + ∑ i = 1 N z i 3 T z i 1 + 1 m i 33 τ i σ + W i ∘ h i + ε i
The attitude control law τ i σ of each UUSV is designed as follows:
τ i σ = m i 33 − K i 2 z i 3 − N ϕ ^ i 2 ε z i 3 ∘ ( h i ∘ h i ) − z i 1 − N b ^ d i 2 ε z i 3
In τ i σ , K i 2 is a diagonal matrix, where the diagonal element is a positive constant, and the other elements are 0.
Adaptive control law ϕ ^ ˙ i and auxiliary adaptive law b ^ ˙ d i based on MLP are as follows:
ϕ ^ ˙ i = Γ N 2 ε ∑ j = 1 3 z i 3 , j 2 h i j 2 2 − K ϕ i [ ϕ ^ i − ϕ ^ i ( 0 ) ]
b ^ ˙ d i = Γ b N 2 ε z i 3 2 2 − K b i [ b ^ d i − b ^ d i ( 0 ) ]
where ϕ i = W i j 2 2 , ϕ i > 0 . ϕ ^ i is the estimation of ϕ . The initial value of ϕ ^ i is ϕ ^ i 0 . ϕ ˜ i = ϕ ^ i − ϕ i is the estimation error. Γ > 0 is the adaptive gain. K ϕ i > 0 is the feedback gain. b ^ d i is the estimation of b d i = ε N + b d . b ˜ d i = b ^ d i − b d i is the estimation error. Γ b is the positive adaptive gain. K b i is the positive feedback gain.
V ˙ 2 can be written as
V ˙ 2 ≤ − K 1 − 1 2 ε I z 1 T z 1 − z 2 T γ − 1 − 1 2 ε 1 + B T B I z 2 − z 3 T K 2 z 3 + 0.5 + N ρ M k p ε + ∑ i = 1 N z i 3 T W i ∘ h i − N 2 ε ϕ ^ i z i 3 ∘ h i ∘ h i − N 2 ε b ^ d i z i 3 + ε i
where K 2 = diag K i 2 . For the convenience of subsequent discussion, let K ϕ = diag K ϕ i , K b = diag K b i .
Theorem 1.
Under Assumptions 1 and 2, a formation composed of  N  UUSVs (10) is considered in a directed graph G . The proposed attitude control law (35), MLP-based adaptive control law (36), and auxiliary control law (37) for each UUSV guarantee that the formation system is SGUUB by selecting positive controller parameters K 0 ,  K 1 ,  K 2 ,  K ϕ ,  K b ,  Γ ,  Γ b ,  γ ,  ε  under bounded initial conditions, where  K 0 ,  K 1 ,  γ  satisfy the following conditions:
λ K 0 min > N 2 ρ M , λ min K 1 > 1 2 ε , λ max γ < 2 ε 1 + B ¯ 2
Let  ϕ = ϕ 1 , … , ϕ N T , ϕ ^ = ϕ ^ 1 , … , ϕ ^ N T , ϕ ^ 0 = ϕ ^ 1 0 , … , ϕ ^ N 0 T , ϕ ˜ = ϕ ˜ 1 , … , ϕ ˜ N T , b ¯ d = b d 1 , … , b d N T , b ¯ ^ d = b ^ d 1 , … , b ^ d N T , b ¯ ˜ d = b ˜ d 1 , … , b ˜ d N T , and b ¯ ^ d 0 = b ^ d 1 0 , … , b ^ d N 0 T .
Consider the candidate Lyapunov function V 3 as follows:
V 3 = V 2 + 1 2 Γ ϕ ˜ T ϕ ˜ + 1 2 Γ b b ¯ ˜ d T b ¯ ˜ d
The derivative of V 3 with respect to time is
V ˙ 3 = B 2 + 0.5 + N ρ M k p ε + ∑ i = 1 N z i 3 T W i ∘ h i − N 2 ε ϕ ^ i z i 3 T z i 3 ∘ h i ∘ h i + ϕ ˜ i N 2 ε ∑ j = 1 3 z i 3 , j 2 h i j 2 2 + z i 3 T ε i − N 2 ε b ^ d i z i 3 + N 2 ε b ˜ d i z i 3 2 2 − ∑ i = 1 N K ϕ i ϕ ˜ i ϕ ^ i − ϕ ^ i ( 0 ) + K b i b ˜ d i b ^ d i − b ^ d i ( 0 )
Let B 2 = − K 1 − 1 2 ε I z 1 T z 1 − z 2 T γ − 1 − 1 2 ε 1 + B T B I z 2 − z 3 T K 2 z 3 .
According to Lemma 4, we have
z i 3 , j W i j T h i j ≤ N 2 ε ϕ i z i 3 , j 2 h i j T h i j + ε 2 N
It can then be concluded that
z i 3 T W i ∘ h i = ∑ j = 1 3 z i 3 , j W i j T h i j ≤ N 2 ε ϕ i ∑ j = 1 3 z i 3 , j 2 h i j T h i j + ε 2 N
N 2 ε ϕ ^ i z i 3 T z i 3 ∘ h i ∘ h i − N 2 ε ϕ ˜ i ∑ j = 1 3 z i 3 , j 2 h i j 2 2 = N 2 ε ϕ i ∑ j = 1 3 z i 3 , j 2 h i j T h i j
z i 3 T ε i − N 2 ε b ^ d i z i 3 ≤ z i 3 2 b d i − N 2 ε b ^ d i z i 3 2 2
z i 3 T ε i − N 2 ε b ^ d i z i 3 ≤ N 2 ε z i 3 2 2 + ε 2 N b d i − N 2 ε b ^ d i z i 3 2 2 = − N 2 ε b ˜ d i z i 3 2 2 + ε 2 N b d i
V ˙ 3 can be written as
V ˙ 3 ≤ B 2 + 1 2 + N ρ M k p + 1 2 ε N + b d ε − ∑ i = 1 N K ϕ i ϕ ˜ i ϕ ^ i − ϕ ^ i ( 0 ) + K b i b ˜ d i b ^ d i − b ^ d i ( 0 )
According to the following two inequalities,
ϕ ˜ i ϕ ^ i − ϕ ^ i ( 0 ) ≥ 1 2 ϕ ˜ i 2 − 1 2 ϕ i − ϕ ^ i ( 0 ) 2
b ˜ d i b ^ d i − b ^ d i ( 0 ) ≥ 1 2 b ˜ d i 2 − 1 2 b d i − b ^ d i ( 0 ) 2
it can further be obtained as
V ˙ 3 ≤ B 2 + 1 2 + N ρ M k p + 1 2 ε N + b d ε − 1 2 ∑ i = 1 N K ϕ i ϕ ˜ i 2 + K b i b ˜ d i 2 + 1 2 ∑ i = 1 N K ϕ i ϕ i − ϕ ^ i ( 0 ) 2 + K b i b d i − b ^ d i ( 0 ) 2 = − K 1 − 1 2 ε I z 1 T z 1 − z 2 T γ − 1 − 1 2 ε 1 + B T B I z 2 − z 3 T K 2 z 3 − 1 2 ϕ ˜ T K ϕ ϕ ˜ − 1 2 b ¯ ˜ d T K b b ¯ ˜ d + 1 2 ϕ − ϕ ^ ( 0 ) T K ϕ ϕ − ϕ ^ ( 0 ) + 1 2 b ¯ d − b ¯ ^ d ( 0 ) T K b b ¯ d − b ¯ ^ d ( 0 ) + 1 2 + N ρ M k p + 1 2 ε N + b d ε
If
β = 1 2 ϕ − ϕ ^ ( 0 ) T K ϕ ϕ − ϕ ^ ( 0 ) + 1 2 b ¯ d − b ¯ ^ d ( 0 ) T K b b ¯ d − b ¯ ^ d ( 0 ) + 1 2 + N ρ M k p + 1 2 ε N + b d ε
we have
V ˙ 3 ≤ − λ min K 1 − 1 2 ε z 1 T z 1 − λ min γ − 1 − 1 2 ε 1 + B ¯ 2 z 2 T z 2 − λ min K 2 z 3 T z 3 − 1 2 λ min K ϕ ϕ ˜ T ϕ ˜ − 1 2 λ min K b b ¯ ˜ d T b ¯ ˜ d + β
If
λ K 0 min > N 2 ρ M , λ min K 1 > 1 2 ε , λ max γ < 2 ε 1 + B ¯ 2
V ˙ 3 can be written as
V ˙ 3 ≤ − α V 3 + β
where
α = min 2 λ min K 1 − ε , 2 λ min γ − 1 − 1 ε 1 + B ¯ 2 , 2 λ min K 2 , Γ λ min K ϕ , Γ b λ min K b
According to Lemma 5, we can conclude that all attitude subsystems are SGUUB.

3.2. Position Controller Design

For the position subsystem, define error variables z i 4 , z i 5 , z i 6 as follows:
z i 4 = ∑ j = 1 N a i j P i − Δ i − P j − Δ j + d i P i − Δ i − P r z i 5 = α ¯ i 2 − α i 2 z i 6 = U i − α ¯ i 2
where α i 2 is a virtual control law. α ¯ i 2 is a filter control law. α ¯ i 2 is obtained by α i 2 through a first-order filter. χ i is given by the corresponding diagonal element and is a diagonal positive definite matrix.
χ i α ¯ ˙ i 2 + α ¯ i 2 = α i 2
The derivative of the error variable z i 4 , z i 5 , z i 6 can be obtained by
z ˙ i 4 = ∑ j = 1 N a i j R i θ i z i 5 + z i 6 + α i 2 − U r − z j 5 + z j 6 + α j 2 − U r + d i R i θ i z i 5 + z i 6 + α i 2 − U r z ˙ i 5 = − χ i − 1 z i 5 − α ˙ i 2 z ˙ i 6 = 1 m i 11 τ i X + f ¯ i X + χ i − 1 z i 5
where z ˙ 4 is expressed by the Kronecker product as z ˙ 4 = R T L ¯ ⊗ I 1 z 5 + z 6 + α 2 − U r , R = diag R i θ i .
Consider the candidate Lyapunov function V 4 as follows:
V 4 = 1 2 z 4 T z 4
The derivative of V 4 with respect to time is
V ˙ 4 = z 4 T R T L ¯ ⊗ I z 5 + z 6 + α 2 − U r
Construct the virtual control law as
α i 2 = − M i 1 z i 4 − M i 0 tanh N z i 4 δ
where M i 1 , M i 0 ∈ R 3 × 3 is a diagonal matrix. δ is a variable positive constant. Definition z 4 = z 14 , … , z N 4 T , z 5 = z 15 , … , z N 5 T , z 6 = z 16 , … , z N 6 T , M 0 = diag M i 0 , M 1 = diag M i 1 , and χ = diag χ i , α 2 = α i 2 , … , α N 2 T .
Therefore, the virtual control law α i 2 can be written in the following form:
α 2 = − M 1 z 4 − M 0 tanh N z 4 δ
Substituting (61) into (59), we have
V ˙ 4 = z 4 T R T L ¯ ⊗ I z 5 + z 6 − z 4 T R T L ¯ ⊗ I M 1 z 4 − M 0 z 4 T R T L ¯ ⊗ I tanh N z 4 δ − z 4 T R T L ¯ ⊗ I 1 U r
According to Lemma 3, it can be obtained as
V ˙ 4 ≤ z 4 T R T L ¯ ⊗ I z 5 + z 6 − z 4 T R T L ¯ ⊗ I M 1 z 4 + z 4 ∞ R T L ¯ ⊗ I 1 N ρ U − M 0 z 4 T R T L ¯ ⊗ I tanh N z 4 δ ≤ z 4 T R T L ¯ ⊗ I z 5 + z 6 − z 4 T R T L ¯ ⊗ I M 1 z 4 + N ρ U k p 1 δ R T L ¯ ⊗ I 1 + N 2 ρ U z 4 T R T L ¯ ⊗ I 1 tan h ( N z 4 δ ) − M 0 z 4 T R T L ¯ ⊗ I tanh N z 4 δ
Let λ M 0 min > N 2 ρ U , and we can further conclude that N 2 ρ U z 4 T R T L ¯ ⊗ I 1 tan h ( N z 4 δ ) − M 0 z 4 T R T L ¯ ⊗ I tanh N z 4 δ < 0 .
Further, we have
V ˙ 4 ≤ − z 4 T R T L ¯ ⊗ I M 1 z 4 + z 4 T R T L ¯ ⊗ I z 5 + z 6 + N ρ U k p 1 δ R T L ¯ ⊗ I 1
Let s = R T L ¯ ⊗ I . We have
V ˙ 4 ≤ − z 4 T s M 1 z 4 + z 4 T s z 5 + z 6 + N ρ U k p 1 δ s
Next, the attitude control law τ i X of each UUSV is designed. And a Lyapunov function is selected as follows:
V 5 = V 4 + 1 2 z 5 T z 5 + 1 2 z 6 T z 6
The derivative of V 5 with respect to time is
V ˙ 5 = V ˙ 4 + z 5 T z ˙ 5 + z 6 T z ˙ 6
where z ˙ 5 can be represented as follows:
z ˙ 5 = − χ − 1 z 5 + M 1 z ˙ 4 + M 0 1 − tanh T N z 4 δ tanh N z 4 δ N z ˙ 4 δ
Let C = M 1 z ˙ 4 + M 0 1 − tanh T N z 4 δ tanh N z 4 δ N z ˙ 4 δ . And there exists a positive constant C ¯ that satisfies C 2 ≤ C ¯ .
Then, (67) can be written as
V ˙ 5 = − z 4 T s M 1 z 4 + s z 4 T z 5 + z 6 T s z 4 + z ˙ 6 + z 5 T − χ − 1 z 5 + C + N ρ U k p 1 δ s
According to Lemma 4, the following inequalities can be obtained:
z 4 T z 5 ≤ 1 2 ( δ z 4 T z 4 + 1 δ z 5 T z 5 )
z 5 T C ≤ 1 2 1 δ z 5 T C T C z 5 + δ
V ˙ 5 can be written as
V ˙ 5 ≤ − s M 1 − 1 2 s δ z 4 T z 4 − z 5 T χ − 1 + 1 2 δ 1 + C T C I z 5 + 1 2 + N ρ U k p 1 s δ + z 6 T s z 4 + z ˙ 6
With C 1 = s z 4 + z ˙ 6 , F i X ( · ) = f ¯ i X + χ i − 1 z i 5 , the following equation can be obtained:
z 6 T C 1 = ∑ i = 1 N z i 6 T s z i 4 + 1 m i 11 τ i X + F i X ( · )
f ¯ i X is difficult to be accurate. Thus, the RBF neural network is used to process it.
F i X ( · ) = W i X ∘ h i X + δ i
where W i = W i 4 T , W i 5 T , W i 6 T T , h i = h i 4 T , h i 5 T , h i 6 T T , and δ i ≤ δ N , i = 1 , … , N , δ N > 0 .
V ˙ 5 can be written as
V ˙ 5 ≤ − s M 1 − 1 2 s δ z 4 T z 4 − z 5 T χ − 1 + 1 2 δ 1 + C T C I z 5 + 1 2 + N ρ U k p 1 s δ + ∑ i = 1 N z i 6 T s z i 4 + 1 m i 11 τ i X + W i X ∘ h i X + δ i
The attitude control law τ i X of each UUSV is designed as follows:
τ i X = m i 11 − M i 2 z i 6 − N ϕ ^ i X 2 δ z i 6 ∘ ( h i X ∘ h i X ) − s z i 4 − N b ^ d i X 2 δ z i 6
where M i 2 is a diagonal matrix.
Adaptive control law ϕ ^ ˙ i X and auxiliary adaptive law b ^ ˙ d i X based on MLP are as follows:
ϕ ^ ˙ i X = Γ X N 2 δ ∑ j = 1 3 z i 6 , j 2 h i j X 2 2 − M ϕ i X [ ϕ ^ i X − ϕ ^ i X ( 0 ) ]
b ^ ˙ d i X = Γ b X N 2 δ z i 6 2 2 − M b i X [ b ^ d i X − b ^ d i X ( 0 ) ]
where ϕ i X = W i j 2 2 , ϕ i X > 0 . ϕ ^ i X is the estimation of ϕ X . ϕ ^ i X 0 is the initial value of ϕ ^ i X . ϕ ˜ i X = ϕ ^ i X − ϕ i X is the estimation error. Γ X > 0 is the adaptive gain. M ϕ i X > 0 is the feedback gain. b ^ d i X is the estimation of b d i X = δ N + b d X . b ˜ d i X = b ^ d i X − b d i X is the estimation error. Γ b X is the positive adaptive gain. M b i X is the positive feedback gain.
V ˙ 5 can be written as
V ˙ 5 ≤ − s M 1 − 1 2 s δ z 4 T z 4 − z 5 T χ − 1 + 1 2 δ 1 + C T C I z 5 − z 6 T M 2 z 6 + 1 2 + N ρ U k p 1 s δ + ∑ i = 1 N z i 6 T W i X ∘ h i X − N ϕ ^ i X 2 δ z i 6 ∘ ( h i X ∘ h i X ) − N b ^ d i X 2 δ z i 6 + δ i
where M 2 = diag M i 2 . For the convenience of subsequent discussion, let M ϕ X = diag M ϕ i X , M b X = diag M b i X .
Theorem 2.
Under assumptions 1 and 2, a formation composed of  N  UUSVs (11) is considered in a directed graph  G . The proposed position control law (76), the MLP-based adaptive control law (77), and the auxiliary control law (78) for each UUSV guarantee that the formation system is SGUUB by selecting positive controller parameters M 0 ,  M 1 ,  M 2 ,  M ϕ X ,  M b X ,  Γ X ,  Γ b X ,  χ ,  δ  under bounded initial conditions, where  M 0 ,  M 1 ,  χ  satisfy the following conditions:
λ M 0 min > N 2 ρ U , λ min s M 1 > 1 2 s δ , λ max χ < 2 δ 1 + C ¯ 2
Let ϕ X = ϕ 1 X , … , ϕ N X T , ϕ ^ X = ϕ ^ 1 X , … , ϕ ^ N X T , ϕ ^ X 0 = ϕ ^ 1 X 0 , … , ϕ ^ N X 0 T , ϕ ˜ X = ϕ ˜ 1 X , … , ϕ ˜ N X T , b ¯ d X = b d 1 X , … , b d N X T , b ¯ ^ d X = b ^ d 1 X , … , b ^ d N X T , b ¯ ˜ d X = b ˜ d 1 X , … , b ˜ d N X T , and b ¯ ^ d X 0 = b ^ d 1 X 0 , … , b ^ d N X 0 T .
Consider the candidate Lyapunov function V 6 as follows:
V 6 = V 5 + 1 2 Γ X ϕ ˜ X T ϕ ˜ X + 1 2 Γ b X b ¯ ˜ d X T b ¯ ˜ d X
The derivative of V 6 with respect to time is
V ˙ 6 = C 2 + 1 2 + N ρ U k p 1 s δ + ∑ i = 1 N z i 6 T W i X ∘ h i X − N ϕ ^ i X 2 δ z i 6 T z i 6 ∘ ( h i X ∘ h i X ) + ϕ ˜ i X N 2 δ ∑ j = 1 3 z i 6 , j 2 h i j X 2 2 + z i 6 T δ i − N b ^ d i X 2 δ z i 6 + N 2 δ b ˜ d i X z i 6 2 2 − ∑ i = 1 N M ϕ i X ϕ ˜ i X [ ϕ ^ i X − ϕ ^ i X ( 0 ) ] + M b i X b ˜ d i X [ b ^ d i X − b ^ d i X ( 0 ) ]
Let C 2 = − s M 1 − 1 2 s δ z 4 T z 4 − z 5 T χ − 1 + 1 2 δ 1 + C T C I z 5 − z 6 T M 2 z 6 . According to Lemma 4, we have
z i 6 T W i X ∘ h i X ≤ N 2 δ ϕ i X ∑ j = 1 3 z i 6 , j 2 h i j X T h i j X + δ 2 N
N ϕ ^ i X 2 δ z i 6 T z i 6 ∘ ( h i X ∘ h i X ) − ϕ ˜ i X N 2 δ ∑ j = 1 3 z i 6 , j 2 h i j X 2 2 = N 2 δ ϕ i X ∑ j = 1 3 z i 6 , j 2 h i j X T h i j X
z i 6 T δ i − N b ^ d i X 2 δ z i 6 ≤ N 2 δ z i 6 2 2 + δ 2 N b d i X − N 2 δ b ^ d i X z i 6 2 2 = − N 2 δ b ˜ d i X z i 3 2 2 + δ 2 N b d i X
V ˙ 6 can be written as
V ˙ 6 ≤ C 2 + 1 2 + 1 2 δ N + b d X + N ρ U k p 1 s δ − ∑ i = 1 N M ϕ i X ϕ ˜ i X [ ϕ ^ i X − ϕ ^ i X ( 0 ) ] + M b i X b ˜ d i X [ b ^ d i X − b ^ d i X ( 0 ) ]
According to the following two inequalities,
ϕ ˜ i X [ ϕ ^ i X − ϕ ^ i X ( 0 ) ] ≥ 1 2 ϕ ˜ i X 2 − 1 2 ϕ i X − ϕ ^ i X ( 0 ) 2
b ˜ d i X [ b ^ d i X − b ^ d i X ( 0 ) ] ≥ 1 2 b ˜ d i X 2 − 1 2 b d i X − b ^ d i X ( 0 ) 2
Further, it can be obtained
V ˙ 6 ≤ C 2 + 1 2 + 1 2 δ N + b d X + N ρ U k p 1 s δ − 1 2 ∑ i = 1 N M ϕ i X ϕ ˜ i X 2 + M b i X b ˜ d i X 2 + 1 2 ∑ i = 1 N M ϕ i X ϕ i X − ϕ ^ i X ( 0 ) 2 + M b i X b d i X − b ^ d i X ( 0 ) 2 = − s M 1 − 1 2 s δ z 4 T z 4 − z 5 T χ − 1 + 1 2 δ 1 + C T C I z 5 − z 6 T M 2 z 6 − 1 2 ϕ ˜ X T M ϕ X ϕ ˜ X − 1 2 b ¯ ˜ d X T M b X b ¯ ˜ d X + 1 2 ϕ X − ϕ ^ X ( 0 ) T M ϕ X ϕ X − ϕ ^ X ( 0 ) + 1 2 b ¯ d X − b ¯ ^ d X ( 0 ) T M b X b ¯ d X − b ¯ ^ d X ( 0 ) + 1 2 + 1 2 δ N + b d X + N ρ U k p 1 s δ
If
β X = 1 2 ϕ X − ϕ ^ X ( 0 ) T M ϕ X ϕ X − ϕ ^ X ( 0 ) + 1 2 b ¯ d X − b ¯ ^ d X ( 0 ) T M b X b ¯ d X − b ¯ ^ d X ( 0 ) + 1 2 + 1 2 δ N + b d X + N ρ U k p 1 s δ
we have
V ˙ 6 ≤ − λ min s M 1 − 1 2 s δ z 4 T z 4 − λ min χ − 1 − 1 2 δ 1 + C ¯ 2 z 5 T z 5 − λ min M 2 z 6 T z 6 − 1 2 λ min M ϕ X ϕ ˜ X T ϕ ˜ X − 1 2 λ min M b X b ¯ ˜ d X T b ¯ ˜ d X + β X
If
λ M 0 min > N 2 ρ U , λ min s M 1 > 1 2 s δ , λ max χ < 2 δ 1 + C ¯ 2
V ˙ 6 can be written as
V ˙ 6 ≤ − α X V 6 + β X
where
α X = min 2 λ min s M 1 − s δ , 2 λ min χ − 1 − 1 δ 1 + C ¯ 2 , 2 λ min M 2 , Γ X λ min M ϕ X , Γ b X λ min M b X
According to Lemma 5, we can conclude that all position subsystems are SGUUB.

4. Simulation

This section intends to validate the effectiveness of the distributed formation control algorithm proposed in Section 3. For this purpose, numerical simulation is carried out for a formation mission scenario. The parameters are shown in Table 1 [24]. Consider a formation composed of six UUSVs. The initial state of the formation is set as shown in Table 2.
Table 1. Main model parameters of the UUSV.
Table 2. Initial states of the UUSV formation.
By using the proposed control law and adaptive law, the simulation results are depicted in Figure 1, Figure 2, Figure 3 and Figure 4. Figure 1 shows the trajectories of six vehicles on the plane, and it is obvious that the desired formation shape is finally obtained. Using the designed controller, all the closed-loop systems should be SGUUB.
Figure 1. Formation trajectory on plane.
Figure 2. Model error variables.
Figure 3. Virtual control law.
Figure 4. Control Input.
Figure 2 shows the error variables of each vehicle. It is obvious that the chattering of the curve is less, and the error variables of the model are SGUUB.
The virtual control laws α 1 and α 2 of each vehicle are shown in Figure 3. We can observe that the control of the attitude subsystem and the position subsystem are continuous and SGUUB.
The control inputs τ X and τ σ of each vehicle are shown in Figure 4. We can observe that the simulation curve of τ σ is relatively chattering-free, and all the controls of the position subsystem and attitude subsystem are continuous and SGUUB.
The above simulation results are consistent with Theorems 1 and 2.

5. Conclusions and Discussion

This paper presents a distributed adaptive formation control scheme for UUSV based on a directed graph. In the design process, considering the underactuation characteristics, we redefine the dynamic and kinematic models of the formation. The RBF neural network based on MLP is used for the online approximation of uncertain model parameters and external disturbances. DSC is used to solve the complex differential problem in the calculation process, which further improves the practicability of the controller. From the simulation results, it can be seen that the controller and two adaptive laws designed by this scheme successfully realize the effective control of the formation, achieve the desired goal, and effectively solve the underactuated control of UUSV formation and the uncertain problems of the dynamic model and quantitative parameters. In the future, we will carry out research on the event-triggered UUSV formation control technology.

Author Contributions

Conceptualization, W.H., L.Z. and X.C.; methodology, W.H. and Y.L.; software, W.H.; validation, Y.Z., L.Z. and Y.L.; formal analysis, Y.Z. and W.H.; investigation, L.Z.; resources, L.Z.; data curation, Y.Z.; writing—original draft preparation, W.H.; writing—review and editing, L.Z.; visualization, Y.Z.; supervision, Y.Z.; project administration, L.Z.; funding acquisition, L.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This research was co-funded by the Fundamental Research Funds for the Central Universities grant number 3072024XX0108 and the National Key Research and Development Program of China grant number 2021YFC2803400.

Institutional Review Board Statement

Not applicable.

Data Availability Statement

Data are unavailable due to privacy restrictions.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Zhang, L.; Zheng, Y.; Huang, B.; Su, Y. Finite-time trajectory tracking control for under-actuated unmanned surface vessels with saturation constraint. Ocean Eng. 2022, 249, 110745. [Google Scholar] [CrossRef] [Scilit]
  2. Zheng, Y.; Zhang, L. Internal model observer-based distributed moving-target-fencing control of multiple underactuated marine surface vehicles with intermittent communications. Ocean Eng. 2024, 300, 117383. [Google Scholar] [CrossRef] [Scilit]
  3. Zhang, L.; Zheng, Y.; Huang, Z.; Huang, B.; Su, Y. Distributed global output-feedback formation control without velocity measurement for multiple unmanned surface vehicles. ISA Trans. 2024, 147, 118–129. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  4. Yan, C.; Zhang, W.; Su, H.; Li, X. Adaptive bipartite time-varying output formation control for multiagent systems on signed directed graphs. IEEE Trans. Cybern. 2021, 52, 8987–9000. [Google Scholar] [CrossRef] [Scilit]
  5. Wu, W.; Peng, Z.; Wang, D.; Liu, L.; Han, Q.-L. Network-based line-of-sight path tracking of underactuated unmanned surface vehicles with experiment results. IEEE Trans. Cybern. 2021, 52, 10937–10947. [Google Scholar] [CrossRef] [Scilit]
  6. He, S.; Wang, M.; Dai, S.-L.; Luo, F. Leader–follower formation control of USVs with prescribed performance and collision avoidance. IEEE Trans. Indust. Informat. 2018, 15, 572–581. [Google Scholar] [CrossRef] [Scilit]
  7. Peng, Z.; Wang, D.; Li, T.; Han, M. Output-feedback cooperative formation maneuvering of autonomous surface vehicles with connectivity preservation and collision avoidance. IEEE Trans. Cybern. 2019, 50, 2527–2535. [Google Scholar] [CrossRef] [Scilit]
  8. Zheng, Y.; Zhang, L.; Huang, B.; Su, Y. Distributed secure formation control for autonomous surface vessels by performance adjustable event-triggered mechanism. Int. J. Robust Nonlinear Control 2023, 33, 8490–8507. [Google Scholar] [CrossRef] [Scilit]
  9. Li, S.; Er, M.J.; Wang, N.; Chien, C.-J.; Yang, Z. Adaptive fuzzy output regulation for formation control of unmanned surface vehicles. In Proceedings of the 2017 International Conference on Fuzzy Theory and Its Applications, New York, NY, USA, 12–15 November 2017; pp. 1–6. [Google Scholar]
  10. Peng, Z.; Wang, J.; Wang, D. Distributed maneuvering of autonomous surface vehicles based on neurodynamic optimization and fuzzy approximation. IEEE Trans. Control Syst. Technol. 2018, 26, 1083–1090. [Google Scholar] [CrossRef] [Scilit]
  11. Peng, Z.; Wang, D.; Chen, Z.; Hu, X.; Lan, W. Adaptive dynamic surface control for formations of autonomous surface vehicles with uncertain dynamics. IEEE Trans. Control Syst. Technol. 2012, 21, 513–520. [Google Scholar] [CrossRef] [Scilit]
  12. Peng, Z.; Wang, D.; Li, T.; Wu, Z. Leaderless and leader-follower cooperative control of multiple marine surface vehicles with unknown dynamics. Nonlinear Dyn. 2013, 74, 95–106. [Google Scholar] [CrossRef] [Scilit]
  13. Zhao, Y.; Guo, G.; Ding, L.; Xu, Y.; Xu, H. Collision/obstacle avoidance control of vessel formation. Inf. Control. 2012, 41, 425–432, 438. [Google Scholar]
  14. He, S.; Dai, S.-L.; Luo, F. Asymptotic trajectory tracking control with guaranteed transient behavior for MSV with uncertain dynamics and external disturbances. IEEE Trans. Ind. Electron. 2018, 66, 3712–3720. [Google Scholar] [CrossRef] [Scilit]
  15. Chen, Y.; Wei, P. Coordinated adaptive control for coordinated path-following surface vessels with a time-invariant orbital velocity. IEEE/CAA J. Autom. Sin. 2014, 1, 337–346. [Google Scholar] [CrossRef] [Scilit]
  16. Jin, X. Fault tolerant finite-time leader-follower formation control for autonomous surface vessels with LOS range and angle constraints. Automatica 2016, 68, 228–236. [Google Scholar] [CrossRef] [Scilit]
  17. Riahifard, A.; Rostami, S.M.H.; Wang, J.; Kim, H.-J. Adaptive leader-follower formation control of under-actuated surface vessels with model uncertainties and input constraints. Appl. Sci. 2019, 9, 3901. [Google Scholar] [CrossRef] [Scilit]
  18. Qu, Y.; Cai, L. Nonlinear positioning control for underactuated unmanned surface vehicles in the presence of environmental disturbances. IEEE/ASME Trans. Mechatron. 2022, 27, 5381–5391. [Google Scholar] [CrossRef] [Scilit]
  19. Karnani, C.; Raza, S.A.; Asif, M.; Ilyas, M. Adaptive control algorithm for trajectory tracking of underactuated unmanned surface vehicle (UUSV). J. Robot. 2023, 2023, 4820479. [Google Scholar] [CrossRef] [Scilit]
  20. Mu, D.; Wang, G.; Fan, Y. Trajectory tracking control for underactuated unmanned surface vehicle subject to uncertain dynamics and input saturation. Neural Comput. Appl. 2021, 33, 12777–12789. [Google Scholar] [CrossRef] [Scilit]
  21. Li, S.; Wang, X. Finite-time consensus and collision avoidance control algorithms for multiple AUVs. Automatica 2013, 49, 3359–3367. [Google Scholar] [CrossRef] [Scilit]
  22. Lombana, D.A.B.; Di Bernardo, M. Multiplex PI control for consensus in networks of heterogeneous linear agents. Automatica 2016, 67, 310–320. [Google Scholar] [CrossRef] [Scilit]
  23. Ge, S.S.; Wang, C. Adaptive neural control of uncertain MIMO nonlinear systems. IEEE Trans. Neural Netw. 2004, 15, 674–692. [Google Scholar] [CrossRef] [Scilit] [PubMed]
  24. Skjetne, R.; Fossen, T.I.; Kokotović, P.V. Adaptive maneuvering, with experiments, for a model ship in a marine control laboratory. Automatica 2005, 41, 289–298. [Google Scholar] [CrossRef] [Scilit]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Article Metrics

Citations

Article Access Statistics

Multiple requests from the same IP address are counted as one view.