Adaptive sliding mode control for disturbed multirobot systems performing target tracking under continuously time-varying topologies

Abstract

This article studies coordination control of the multirobot system for target tracking problem with disturbances in time-varying communication environment. An adaptive sliding mode control is designed to complete the target tracking task with the effect of disturbances and continuously time-varying topologies. This means that the communication between the robots is not switching among constant topologies but changes continuously along with time. The continuously time-varying topology is constructed by a set of fixed Laplacian matrices and variable scheduling functions, which is known as polytopic model structure. The designed adaptive sliding mode control can effectively lower the influence of disturbance and improve tracking performance even under the continuously time-varying topology. Furthermore, numerical simulations are given to demonstrate the effectiveness of the obtained results.

Keywords

Multirobot systems with disturbance time-varying topology adaptive sliding mode control target tracking strategy

Introduction

Collaborative distributed control of multirobot systems, such as swarm, formation, and collective missions, have raised widespread attention in the past decade.^1

–7 Consensus is the basic theory in multirobot coordination control, which is often done by designing a distributed consensus protocol to make a set of robots to agree on a common state.^8

–12 Target tracking is a typical application of multirobot systems, in which the robots’ states reach consensus with the target.^4
–6

Sliding mode control is known as a powerful tool for handling system operation under uncertain conditions. The corresponding process is divided into two steps. First of all, to achieve the required specifications, a sliding manifold is designed. Secondly, a discontinuous control law is designed so as to make the system trajectory reach the manifold in fixed time and stay close to the equilibrium point near the sliding mode manifold. Due to the effectiveness and simplicity of sliding mode control in practical application, it has been widely used in many fields and applications.^13

–17

Since disturbance is inevitable in most cases, some scholars have designed sliding-mode controller of multirobot system under disturbance.^18
–20 In the case of modeling and input uncertainty, faced with consistency tracking control problem of second-order multirobot system, a new nonsingular fast-sliding model surface is designed, in which the surface has a bounded convergence time without considering the initial state.¹⁸ A sliding mode nonsingular terminal distributed protocol is proposed to accomplish consensus tracking problem of second-order multirobot systems in fixed time.¹⁹ The authors propose a finite-time robust event trigger control strategy based on integral sliding mode with the purpose of solving the fast distributed coordination problem of multirobot systems with limited disturbances.²⁰ According to the definition of the new measurement error, the fast approach law is adopted to enhance the convergence speed of the algorithm, and the system triggering conditions suitable for model uncertainty and disturbance are derived.

Due to the limitation of communication technique, the topology of the system may be time varying.^21

–24 Recently, researchers have proposed different methods to deal with control problems with switching network structure. For example, stabilization of Boolean control networks which are switching among different Boolean networks is investigated.²² Several new multiple Lyapunov functions are constructed to solve the multirobot tracking control system with switching topologies.²³ There also have been some results on the adaptive sliding mode control of the system with switching topologies. Specifically, under switching topologies, an adaptive sliding mode control approach for multirobot systems is presented, in which the model parameters of high-order robots remain unknown.²⁵ The results are used to a set of flexible joint multiple single-link manipulators. An adaptive fixed-time distributed consensus protocol with switching topologies for consensus tracking of uncertain high-order multirobot systems is investigated.²⁶ The protocol is designed by dynamic surface control technique and backstepping method, making sure that the tracking errors converge although the communication topologies between robots change dynamically as time goes by.

The aforementioned results in the literature are based on fixed or switching topologies, as shown in Figure 1, namely, the communication topology between robots switches among several constant topologies. It is constant during time intervals. However, in real applications, it is possible that the communication topologies are changing continuously along with time, as shown in Figure 2. Therefore, in this article, an adaptive sliding mode control for multirobot systems performing target tracking task with disturbances and continuously time-varying topologies is studied. The continuously time-varying topology is constructed using a set of constant Laplacian matrices and variable scheduling function. This is known as a polytopic model structure. Simulation results are presented to verify the effectiveness of the proposed distributed adaptive control scheme.

Figure 1.

Illustration of switching topologies.

Figure 2.

Illustration of continuously time-varying topologies.

Preliminaries and problem description

Consider that there are n robots deployed in the three-dimensional space aiming to track any target in this space. The dynamics of the i’th tracking robot with disturbances is described by

\begin{array}{l} {\dot{x}}_{i} (t) = v_{i} (t) \\ {\dot{v}}_{i} (t) = f (x_{i} (t), v_{i} (t)) + u_{i} (t) + d_{i} (t) \end{array}

where $x_{i} (t) = (x_{i 1} (t), x_{i 2} (t), x_{i 3} (t {))}^{T} \in ℝ^{3}, v_{i} (t) = (v_{i 1} (t), v_{i 2} (t), v_{i 3} (t {))}^{T} \in ℝ^{3}$ , in which $x_{i 1} (t), x_{i 2} (t), x_{i 3} (t)$ and $v_{i 1} (t), v_{i 2} (t), v_{i 3} (t)$ are position and velocity states of the ith tracking robot on x-axis, y-axis, and z-axis, respectively. $f (x_{i} (t), v_{i} (t))$ denotes a nonlinear function. $u_{i} (t) \in ℝ^{3}$ is the controller to be designed. $d_{i} (t)$ describes the disturbances implied on robot i and satisfies the constraint that $d_{i} (t) \leq d_{i}^{*}$ , where $d_{i}^{*}$ is an unknown constant.

In a target tracking system, the robots are designated to track a target, whose dynamics is assumed to be as follows

\begin{array}{l} {\dot{x}}_{t} (t) = v_{t} (t) \\ {\dot{v}}_{t} (t) = f (x_{t} (t), v_{t} (t)) \end{array}

where $x_{t} (t) = (x_{t 1} (t), x_{t 2} (t), x_{t 3} (t {))}^{T} \in ℝ^{3}$ , in which $x_{t 1} (t), x_{t 2} (t), x_{t 3} (t)$ are the position states of the target robot on x-axis, y-axis, and z-axis, respectively. $v_{t} (t) = (v_{t 1} (t), v_{t 2} (t), v_{t 3} (t {))}^{T} \in ℝ^{3}$ , in which $v_{t 1} (t), v_{t 2} (t), v_{t 3} (t)$ are the velocity states of the target robot on x-axis, y-axis, and z-axis, respectively. $f (x_{t} (t), v_{t} (t))$ describes the nonlinear force of the target.

In this article, the nonlinear function $f (\cdot)$ satisfies Assumption 1.

Assumption 1

The nonlinear function $f (\cdot)$ satisfies the following inequality (3)

| f (x_{i}, x_{i}) - f (x_{t}, v_{t}) | \leq l_{x} | x_{i} - x_{t} | + l_{v} | v_{i} - v_{t} |

where l_x and l_v are positive constants.

We can model the topology of robots and the target as a dynamic graph, with each node representing a robot. Consider the multirobot target tracking system with n robots, $G ≜ {V, E, A}$ is a weighted graph, $V (G) = {1, 2, \dots, n}$ is a set of nodes, $E (G) \subseteq V \times V$ is a set of finite edges and $A = [a_{i j}] \in ℝ^{n \times n}$ is a weighted adjacency matrix with non-negative elements. The neighbors of robot i are represented by a set $V_{i} = {j | j \in V, (j, i) \in E}$ . If robot $j \notin V_{i}$ , which implies robot j is not able to get information about robot i. In this case, $a_{i j} = 0,$ else $a_{i j} > 0$ . Considering the following representation $D ≜ diag {\sum_{j \in V_{1}} a_{1 j}, \sum_{j \in V_{2}} a_{2 j}, \dots, \sum_{j \in V_{n}} a_{n j}},$ then the corresponding Laplacian matrix $L = [l_{i j}] \in ℝ^{n \times n}$ is $L ≜ D - A$ . b_i is used to describe the connection status between robot i and the target. If $b_{i} > 0$ , we say that the ith robot can obtain the target’s information. In other ways, $b_{i} = 0$ . The matrix that describes the connection status of the tracking robots and the target is denoted as $B ≜ diag {b_{1}, b_{2}, \dots, b_{n}}$ .

The main purpose is to design an adaptive distributed tracking controller that enables the robots to successfully track the target with continuously time-varying topologies. In a continuously time-varying communication topology, the connection between robot j and robot i changes along with time, then, $a_{i j}$ and b_i are denoted as $a_{i j} (t)$ and $b_{i} (t)$ in this article. Thereafter, the Laplacian matrix then becomes a time-varying matrix $L (t)$ . Similarly, the matrix $B$ representing the connection status between robots and the target turns to be $B (t)$ . It should be noted that a dynamic graph is called undirected if $a_{i j} (t) = a_{j i} (t)$ , otherwise, it is called directed. In this article, we consider the continuously time-varying communication topology to be undirected.

Polytopic model of Laplacian matrix

In this article, the continuously time-varying topology is established as a polytopic model. Denote $L (t) = L (t) + B (t)$ as the relative Laplacian matrix of the multirobot system performing target tracking task. The polytopic structure describes the overall time-varying Laplacian matrices as following: $L (t) = \sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k}$ , where ${\hat{L}}_{k} \in ℝ^{n \times n}$ are constant matrices, and N is a finite positive constant. The corresponding scheduling functions are $α_{k} (t) > 0$ and $\sum_{k = 1}^{N} α_{k} (t) = 1$ and the derivative of $α_{k} (t)$ satisfies ${\dot{α}}_{k} (t) \leq α_{d k}$ , in which $α_{d k}$ are known constants.

Remark 1

Using the proposed polytopic model to describe the continuously time-varying topology has certain advantages. Currently, the results on multirobot systems with continuously time-varying topology in the literature are usually based on the eigenvalues of the corresponding time-varying Laplacian matrix. It requires the eigenvalues of Laplacian matrix to be positive, while the eigenvalues of the derivative of Laplacian matrix to be negative. The proposed polytopic model avoids such conservative constraints on the time-varying topology. However, this structure also suffers some limitations. For example, the time-varying Laplacian matrix is not always easy to be decoupled as appropriate scheduling functions and constant Laplacian matrices.

Definition 1

We can say that robot i (1) successfully tracks the target (2) if the following equation exists

lim_{t \to \infty} (∥ x_{i} (t) - x_{t} (t) ∥ + ∥ v_{i} (t) - v_{t} (t) ∥) = 0, i \in V

According to Definition 1, successful tracking is achieved if the tracking errors tend to be zero. Therefore, we reorganize the target tracking multirobot system into an overall tracking error system as follows.

Define the tracking position and velocity errors of the ith robot and the target as $e_{x i} (t) = x_{i} (t) - x_{t} (t), e_{v i} (t) = v_{i} (t) - v_{t} (t)$ , respectively. Then, the whole error of ith robot with its neighbors and the target is

\begin{array}{l} ε_{x i} (t) = & \sum_{j = 1}^{n} a_{i j} (t) (e_{x i} (t) - e_{x j} (t)) + b_{i} (t) e_{x i} (t) \\ ε_{v i} (t) = & \sum_{j = 1}^{n} a_{i j} (t) (e_{v i} (t) - e_{v j} (t)) + b_{i} (t) e_{v i} (t) \end{array}

Based on the characteristics of Laplacian matrix $L (t)$ , the tracking error can be reorganized in a compact form (6)

\begin{array}{l} E_{x} (t) = [\begin{matrix} ε_{x 1} (t) \\ ε_{x 2} (t) \\ \dots \\ ε_{x n} (t) \end{matrix}] = L (t) [\begin{matrix} e_{x 1} (t) \\ e_{x 2} (t) \\ \dots \\ e_{x n} (t) \end{matrix}] = L (t) e_{x} (t) \\ E_{v} (t) = [\begin{matrix} ε_{v 1} (t) \\ ε_{v 2} (t) \\ \dots \\ ε_{v n} (t) \end{matrix}] = L (t) [\begin{matrix} e_{v 1} (t) \\ e_{v 2} (t) \\ \dots \\ e_{v n} (t) \end{matrix}] = L (t) e_{v} (t) \end{array}

Then, the whole tracking system with disturbances under time-varying topology turns to be

\begin{array}{l} {\dot{E}}_{x} (t) = E_{v} (t) \\ {\dot{E}}_{v} (t) = \sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} [F (e_{x} (t), e_{v} (t)) + U (t) + D (t)] \end{array}

where $F (e_{x} (t), e_{v} (t)) = (f (x_{1} (t), v_{1} (t)) - f (x_{t} (t), v_{t} (t)), f (x_{2} (t), v_{2} (t)) - f (x_{t} (t), v_{t} (t)),..., f (x_{n} (t), v_{n} (t)) - f (x_{t} (t), v_{t} (t {)))}^{T}$ stands for the whole nonlinear autodynamic inconsistency of the multirobot target tracking system at time t. $D (t) = [d_{1} (t), d_{2} (t), \dots, d_{n} (t {)]}^{T}$ is the collective disturbances at time t.

Before addressing the main results, we introduce the following lemma.

Lemma 1

For vectors x,y with any suitable dimensions, there exists a symmetric positive definite matrix Z such that the following inequality holds²⁷

\pm 2 x^{T} y \leq x^{T} Z x + y^{T} Z^{- 1} y

Adaptive sliding mode controller design and stability analysis for disturbed systems with time-varying topology

In practical applications, disturbances are not avoidable since these robots move in an uncertain environment. To successfully perform target tracking for multirobot system (7) with disturbances, this section designs an adaptive sliding mode control method. Then, the stability is analyzed.

The sliding mode is defined as

S = C E_{x} (t) + E_{v} (t)

where $C = diag \{c_{1}, c_{2}, \dots, c_{n}\}$ is a constant control matrix. $S = {(s_{1}, s_{2}, \dots, s_{n})}^{T}$ is the collective sliding mode surface designed for the multirobot target tracking system.

For multirobot system with disturbances (7), an adaptive overall sliding mode control law is designed as

\begin{array}{l} U (t) = & {(\sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k})}^{- 1} \\ [- C E_{v} (t) - η S - β sign (S) - \frac{| S |}{S} \hat{δ}] \end{array}

where $\hat{δ} = {[{\hat{δ}}_{1}, {\hat{δ}}_{2}, \dots, {\hat{δ}}_{n}]}^{T}$ is the estimation of upper bound of disturbances $δ^{*} = \sum_{k = 1}^{N} λ_{max} \{{\hat{L}}_{k}\} {[d_{1}^{*}, d_{2}^{*}, \dots, d_{n}^{*}]}^{T}$ . The adaptive law is

\dot{\hat{δ}} = | S |

It should be noted that here we use $| S |$ , denoting that all the elements are absolute values of the elements in S.

The tracking control protocol for the ith tracking robot becomes

\begin{array}{l} u_{i} (t) = {[\sum_{j = 1, j \neq i}^{n} a_{i j} (t) + b_{i} (t)]}^{- 1} [\sum_{j = 1, j \neq i}^{n} (a_{i j} (t) u_{j} (t)) \\ - c_{i} ε_{v i} - η s_{i} - β sign (s_{i}) - \frac{| s_{i} |}{s_{i}} {\hat{δ}}_{i}] \end{array}

Under continuously time-varying topologies constructed by polytopic model structure, the following theorem provides a sufficient condition for disturbed multirobot systems realizing successful target tracking.

Theorem 1

Consider the multirobot system (7) with disturbances and continuously time-varying topologies. The designed adaptive sliding mode controller (11) can solve the disturbed target tracking problem of (7) if there exists a symmetric positive definite matrix P_k with suitable scale satisfying the condition (12) with known positive constants $η, l_{x}, l_{v}$ , $α_{d 1}, α_{d 2}, \dots α_{d N}$ , designed control matrix C and positive definite matrix Z

Φ_{k} = [\begin{matrix} Φ_{k 11} & P_{k} - 2 η C^{T} + C^{T} Z^{- 1} + l_{x} l_{v} Z \\ * & - 2 η I + Z^{- 1} + l_{v}^{2} Z \end{matrix}] < 0

where $Φ_{k 11} = α_{d 1} P_{1} + α_{d 2} P_{2} + \dots + α_{d N} P_{N} - 2 C^{T} η C + C^{T} Z^{- 1} C + l_{x}^{2} Z,$ for all $k = 1, 2, \dots, N$ , namely the tracking robots (1) are able to successfully track the target (2).

Proof

The Lyapunov function is designed as

\begin{array}{l} V (t) = \sum_{k = 1}^{N} α_{k} (t) E_{x}^{T} (t) P_{k} E_{x} (t) + S^{T} S + {\tilde{δ}}^{T} \tilde{δ} \end{array}

where $\tilde{δ} = δ^{*} - \hat{δ}$ is the estimation error.

Along the trajectory of the system (7), the derivative of Lyapunov function (13) is

\begin{array}{l} \dot{V} (t) = V_{1} (t) + V_{2} (t) + V_{3} (t) \end{array}

with

\begin{array}{l} V_{1} (t) = \sum_{k = 1}^{N} {\dot{α}}_{k} (t) E_{x}^{T} (t) P_{k} E_{x} (t) \\ V_{2} (t) = \sum_{k = 1}^{N} α_{k} (t) [{\dot{E}}_{x}^{T} (t) P_{k} E_{x} (t) + E_{x}^{T} (t) P_{k} {\dot{E}}_{x} (t)] \\ V_{3} (t) = S^{T} \dot{S} + {\dot{S}}^{T} S + {\dot{\tilde{δ}}}^{T} \tilde{δ} + {\tilde{δ}}^{T} \dot{\tilde{δ}} \end{array}

In the light of the fact that ${\dot{α}}_{k} (t) \leq α_{d k}$ , one has

\begin{array}{l} V_{1} (t) \leq \sum_{k = 1}^{N} α_{d k} E_{x}^{T} (t) P_{k} E_{x} (t) \end{array}

According to system (7), $V_{2} (t)$ can be further derived as

\begin{array}{l} V_{2} (t) = \sum_{k = 1}^{N} α_{k} (t) [E_{v}^{T} (t) P_{k} E_{x} (t) + E_{x}^{T} (t) P_{k} E_{v} (t)] \end{array}

Considering the form of sliding mode S and the system (7), we have

\begin{array}{l} V_{3} (t) & = S^{T} [C E_{v} (t) + \sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} F (t, e_{x} (t), e_{v} (t)) \\ + \sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} (U (t) + D (t))] + [C E_{v} (t) \\ + \sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} F (t, e_{x} (t), e_{v} (t)) \\ + \sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} (U (t) + D (t {))]}^{T} S \\ + {[{\dot{δ}}^{*} - \dot{\hat{δ}}]}^{T} \tilde{δ} + {\tilde{δ}}^{T} [{\dot{δ}}^{*} - \dot{\hat{δ}}] \end{array}

Substituting the overall sliding mode control (9) into (18) and considering the fact that ${\dot{δ}}^{*} = 0$ , since $δ^{*}$ is constant, equation (19) is derived

\begin{array}{l} V_{3} (t) & = - 2 S^{T} η S + S^{T} \sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} F (t, e_{x} (t), e_{v} (t)) \\ + {[\sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} F (t, e_{x} (t), e_{v} (t))]}^{T} S \\ + S^{T} [\sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} D (t) - \hat{δ}] \\ + {[\sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} D (t) - \hat{δ}]}^{T} S \\ - S^{T} β sign (S) - {(sign (S))}^{T} β S - {\dot{\hat{δ}}}^{T} \tilde{δ} - {\tilde{δ}}^{T} \dot{\hat{δ}} \end{array}

Recalling $δ^{*} = \sum_{k = 1}^{N} λ_{max} \{{\hat{L}}_{k}\} {[d_{1}^{*}, d_{2}^{*}, \dots, d_{n}^{*}]}^{T}$ and combining Lemma 1, it can be obtained that

\begin{array}{l} V_{3} (t) \leq - 2 S^{T} η S + S^{T} Z^{- 1} S + {[\sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} (l_{x} e_{x} (t) + l_{v} e_{v} (t))]}^{T} \\ Z [\sum_{k = 1}^{N} α_{k} (t) {\hat{L}}_{k} (l_{x} e_{x} (t) + l_{v} e_{v} (t))] \\ - S^{T} β sign (S) - {(sign (S))}^{T} β S \\ + | S |^{T} [δ^{*} - \hat{δ}] + {[δ^{*} - \hat{δ}]}^{T} | S | \\ - {\dot{\hat{δ}}}^{T} \tilde{δ} - {\tilde{δ}}^{T} \dot{\hat{δ}} \end{array}

According to the adaptive law (10) and using equation (6), it can be further derived that

\begin{array}{l} V_{3} (t) \leq & - 2 {[C E_{x} (t) + E_{v} (t)]}^{T} η [C E_{x} (t) + E_{v} (t)] \\ + {[C E_{x} (t) + E_{v} (t)]}^{T} Z^{- 1} [C E_{x} (t) + E_{v} (t)] \\ + {[l_{x} E_{x} (t) + l_{v} E_{v} (t)]}^{T} Z [l_{x} E_{x} (t) + l_{v} E_{v} (t)] \\ + | S |^{T} [δ^{*} - \hat{δ} - \tilde{δ}] + {[δ^{*} - \hat{δ} - \tilde{δ}]}^{T} | S | \\ - β S^{T} sign (S) - β {(sign (S))}^{T} S \end{array}

Given the above inequalities, we are able to organize the derivative of Lyapunov function (14) in a compact form

\begin{array}{l} \dot{V} (t) \leq & \sum_{k = 1}^{N} α_{k} (t) {[\begin{matrix} E_{x} (t) \\ E_{v} (t) \end{matrix}]}^{T} Φ_{k} [\begin{matrix} E_{x} (t) \\ E_{v} (t) \end{matrix}] \\ - β S^{T} sign (S) - β {(sign (S))}^{T} S \end{array}

It is obvious that inequality (12) gives rise to $\dot{V} (t) < 0$ , which suggests that the overall tracking error system (7) is stable. This implies that the position and velocity errors between the robots and the target tend to zero asymptotically. Then, according to Definition 1, the disturbed robots can track the target successfully with proposed adaptive control even under the effect of continuously time-varying topology. The proof is completed.

Numerical results

Numerical examples are conducted to demonstrate the foregoing research result. It is assumed that three robots track a moving target in a three-dimensional space. Numerical comparisons are made to verify the advantages of the designed adaptive control law for disturbed multirobot systems with continuously time-varying topologies.

Simulation of tracking system with time-varying topology and disturbances

The target is assumed to start moving at the origin $x_{t} (0) = [0, 0, 0]$ , with velocity $v_{t} (0) = [0.35, 0.8, 0.6]$ in the three-dimensional space. Three tracking robots are initially deployed at positions, $x_{1} (0) = [0.2, - 1.3, 1], x_{2} (0) = [0.5, 1, 0.7], x_{3} (0) = [0.8, 0, 1.2]$ , with initial velocities $v_{1} (0) = [- 0.8, - 0.2, 0.08], v_{2} (0) = [0.25, - 0.7, 0.2], v_{3} (0) = [- 0.5, 0.8, 0.1]$ , respectively. The following is a description of the nonlinear dynamics

\begin{array}{l} f (x (t), v (t)) = 1.5 | x (t) | - 0.26 v (t) - 1 \end{array}

which satisfies Assumption 1. The Lipschitz constants are $l_{x} = 1.5$ and $l_{v} = 0.26$ .

In the simulations, the continuously time-varying topology of the system is shown in Figure 3. The left-hand side figures represent the decomposed two constant topologies and the right-hand side figure is the time-varying topology in polytopic model. The corresponding Laplacian matrix is

\begin{array}{l} L (t) + B (t) = & α_{1} (t) (\begin{matrix} 0.8 & - 0.8 & 0 \\ - 0.8 & 2.3 & - 1.2 \\ 0 & - 1.2 & 1.2 \end{matrix}) \\ + α_{2} (t) (\begin{matrix} 1.5 & - 0.3 & - 0.7 \\ - 0.3 & 0.3 & 0 \\ - 0.7 & 0 & 1.7 \end{matrix}) \end{array}

where the corresponding scheduling functions are $α_{1} (t) = 0.3 {sin}^{2} (t), α_{2} (t) = 1 - α_{1} (t)$ .

Figure 3.

The time-varying topology in polytopic model.

The derivative of $α_{1} (t)$ is ${\dot{α}}_{1} (t) = 0.6 sin (t) cos (t)$ . Then, the derivative of $α_{2} (t)$ is ${\dot{α}}_{2} (t) = - {\dot{α}}_{1} (t)$ . Thereafter, the assumption ${\dot{α}}_{k} (t) \leq α_{d k}$ is satisfied with $α_{d 1} = α_{d 2} = 0.6$ . In the simulations, $c_{1} = 1.5, c_{2} = 2.8, c_{3} = 3.5$ are the selected control parameters and $η = 1.5, β = 0.16$ are the designed reaching parameters. Based on the conditions in Theorem 1 with an identity matrix Z, we can obtain the below feasible solutions for matrices P ₁ and P ₂

P_{1} = P_{2} = (\begin{matrix} 1.3680 & 0 & 0 \\ 0 & 5.3466 & 0 \\ 0 & 0 & 7.2659 \end{matrix})

The disturbances of the robots are

\begin{array}{l} d_{1} = & [1.5 sin (2 t); 2 cos (0.8 t); 3 cos (1.5 t)] \\ d_{2} = & 1.5 \times [1.5 sin (2 t); 2 cos (0.8 t); 3 cos (1.5 t)] \\ d_{3} = & 0.75 \times [1.5 sin (2 t); 2 cos (0.8 t); 3 cos (1.5 t)] \end{array}

Now, we conduct the simulation for tracking system under the effect of disturbances (25) and the sliding mode control law (9) without the adaptive part $- \frac{| S |}{S} \hat{δ}$ . Figure 4 shows the trajectories of the target and tracking robots, while the position and velocity tracking errors are depicted in Figures 5 and 6, respectively. It can be seen that, for target tracking system under continuously time-varying topology, the tracking robots tend to converge to the target under the effect of coordination control without the designed adaptive part. However, it is obvious that the tracking errors fluctuate in a relatively large range.

Figure 4.

Tracking trajectories for systems under disturbances without adaptive control law.

Figure 5.

Position error trajectories for systems under disturbances without adaptive control law: (a) robot 1, (b) robot 2, and (c) robot 3.

Figure 6.

Velocity error trajectories for systems under disturbances without adaptive control law: (a) robot 1, (b) robot 2, and (c) robot 3.

Comparison simulation of tracking system with designed adaptive sliding mode control

In the following, we show the advantage of the proposed adaptive sliding mode control law (9). All the initial configurations of the robots and the target are the same as those in the above simulation. Figure 7 shows the trajectories of the target and three tracking robots with the proposed adaptive control under the same effect of disturbances, as used in the above simulation. Figures 8 and 9 show the robots’ position and velocity tracking errors suffering disturbances and with the adaptive control law, respectively. Compared to the results in Figures 5 and 6, the tracking errors are significantly reduced. Under the continuously time-varying topology (24), it is shown that the tracking robots can track the moving target successfully.

Figure 7.

Tracking trajectories for systems under disturbances with adaptive control law.

Figure 8.

Position error trajectories for systems under disturbances with adaptive control law: (a) robot 1, (b) robot 2, and (c) robot 3.

Figure 9.

Velocity error trajectories for systems under disturbances with adaptive control law: (a) robot 1, (b) robot 2, and (c) robot 3.

To clearly show the effectiveness of proposed adaptive sliding mode control for multirobot systems, tracking errors are compared in a quantitative way. The maximum and minimum values of position and velocity errors over time interval $[4 s,10 s]$ are tabulated in Tables 1 and 2, respectively. Specifically, for robots 1 and 3, the position tracking errors with adaptive control law are only almost 1/50 of that without adaptive control law. For robot 2, the position errors are reduced to nearly 1/30 of the value without adaptive control. In terms of velocity errors, under the effect of adaptive control, they are shrunk to about 1/30, 1/10, and 1/70 of the values without adaptive control for robots 1, 2, and 3, respectively.

Table 1.

Quantitative comparison of position errors with and without adaptive control law.

e_p	$r 1_{adap}$	$r 2_{adap}$	$r 3_{adap}$	$r 1$	$r 2$	$r 3$
Maximum	0.0033	0.0064	0.0001	0.1486	0.1696	0.0761
Minimum	−0.0050	−0.0002	−0.0002	−0.2111	−0.1521	−0.1340

Table 2.

Quantitative comparison of velocity errors with and without adaptive control law.

e_v	$r 1_{adap}$	$r 2_{adap}$	$r 3_{adap}$	$r 1$	$r 2$	$r 3$
Maximum	0.0085	0.0194	0.0020	0.2231	0.2103	0.1404
Minimum	−0.0059	−0.0164	−0.0019	−0.2711	−0.2145	−0.1648

Conclusions and discussion

In this work, an adaptive timing consensus tracking method for second-order multirobot systems with disturbances is proposed. The time-varying communication topology between the tracking robot and the target is modeled by constant Laplacian matrices and corresponding time-varying scheduling functions, which is called polytopic model. Through constructing the polytopic model, the tracking strategy is less conservative. Numerical examples show that the proposed adaptive sliding mode control tracking strategy is superior to previous related work. With designed adaptive sliding mode control, the disturbed robots can effectively track the target with nonlinear dynamic characteristics under continuously time-varying topology.

This article mainly deals with disturbed target tracking systems continuously time-varying topology and some significant results have been presented. However, there are still some issues to be solved in future research. For example, the time delays are existing in most systems since the robots are exchanging information using wireless communication techniques. How the time delays will affect the control performance of designed adaptive sliding mode control method especially with polytopic structure is worth further investigating.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Natural Science Foundation of China under Grant Nos. 61903022 and 61803006, funded by Beijing Advanced Innovation Center for Intelligent Robots and Systems under Grant No. 2019IRS11.

ORCID iD

Lijing Dong

References

Kim

Kashino

Pineros

, et al. A high-performance millirobot for swarm-behaviour studies: swarm-topology estimation. Int J Adv Robot Syst 2019; 16(6): 1–18.

Sabattini

Secchi

Fantuzzi

. Coordinated dynamic behaviors for multirobot systems with collision avoidance. IEEE Trans Cybern 2017; 47(12): 4062–4073.

Zhao

Liu

. Self-adaptive collective motion of swarm robots. IEEE Trans Autom Sci Eng 2018; 15(4): 1533–1545.

Tang

Ding

, et al. Swarm robots search for multiple targets based on an improved grouping strategy. IEEE ACM Trans Comput Bio Bioinf 2018; 15(6): 1943–1950.

Farmani

Sun

Pack

. A scalable multi-target tracking system for cooperative unmanned aerial vehicles. IEEE Trans Aero Elec Syst 2017; 53(4): 1947–1961.

Jin

, et al. Dynamic task allocation in multi-robot coordination for moving target tracking: a distributed approach. Automatica 2019; 100: 75–81.

Park

Hutchinson

. Robust rendezvous for multi-robot system with random node failures: an optimization approach. Auton Robots 2018; 42: 1807–1818.

Zheng

Wang

. Consensus of hybrid multi-agent systems. IEEE Trans Neur Net Learn Syst 2018; 29(4): 1359–1365.

Wang

Cheng

, et al. Second-order consensus in multiagent systems via distributed sliding mode control. IEEE Trans Cybern 2017; 47(8): 1872–1881.

10.

Yoo

. Distributed consensus tracking of a class of asynchronously switched nonlinear multi-agent systems. Automatica 2018; 87: 421–427.

11.

Gao

. Distributed adaptive tracking control for multi-agent systems with uncertain dynamics. Nonlinear Dyn 2017; 90(4): 2729–2744.

12.

Sakthivel

Kaviarasan

Ahn

, et al. Observer and stochastic faulty actuator-based reliable consensus protocol for multi-agent system. IEEE Trans Syst Man Cybern: Syst 2018; 48(12): 2383–2393.

13.

Zuo

. Nonsingular fixed-time consensus tracking for second-order multi-agent networks. Automatica 2015; 54(C): 305–309.

14.

Mishra

Sinha

. Event-triggered sliding mode based consensus tracking in second order heterogeneous nonlinear multi-agent systems. Eur J Control 2019; 45: 30–44.

15.

Mondal

Ghommam

Saad

. Homogeneous finite-time consensus control for higher-order multi-agent systems by full order sliding mode. J Syst Sci Complex 2018; 31(5): 1186–1205.

16.

Haghighi

Mobayen

. Design of an adaptive super-twisting decoupled terminal sliding mode control scheme for a class of fourth-order systems. ISA Trans 2018; 75: 216–225.

17.

Qin

Zhang

Zheng

, et al. Adaptive sliding mode consensus tracking for second-order nonlinear multiagent systems with actuator faults. IEEE Trans Cybern 2019; 49(5): 1605–1615.

18.

Huang

Jia

. Fixed-time consensus tracking control for second-order multi-agent systems with bounded input uncertainties via NFFTSM. IET Control Theory Appl 2017; 11(16): 2900–2909.

19.

Khanzadeh

Pourgholi

. Fixed-time leader-follower consensus tracking of second-order multi-agent systems with bounded input uncertainties using non-singular terminal sliding mode technique. IET Control Theory Appl 2018; 12(5): 679–686.

20.

Nair

Behera

Kumar

. Event-triggered finite-time integral sliding mode controller for consensus-based formation of multirobot systems with disturbances. IEEE Trans Contr Syst Technol 2019; 27(1): 39–47.

21.

Zhong

Liu

, et al. Output feedback stabilizer design of Boolean networks based on network structure. Front Inform Technol Electr Eng 2020; 21: 247–259.

22.

Liu

Lou

, et al. Set stabilization of probabilistic Boolean control networks: a sampled-data control approach. IEEE Trans Cybern. Epub ahead of print 25 September 2019. DOI: 10.1109/TCYB.2019.2940654.

23.

Wen

Zheng

. On constructing multiple Lyapunov functions for tracking control of multiple agents with switching topologies. IEEE Trans Automat Contr 2019; 64(9): 3796–3803.

24.

Wen

, et al. Coordination and control of complex network systems with switching topologies: a survey. IEEE Trans Syst Man Cybernetics: Syst 2020; 1–16. DOI: 10.1109/TSMC.2019.2961753.

25.

Rezaei

Kabiri

Menhaj

. Adaptive consensus for high-order unknown nonlinear multi-agent systems with unknown control directions and switching topologies. Inform Sci 2018; 459: 224–237.

26.

Zhang

Duan

. Distributed fixed-time consensus tracking for high-order uncertain non-linear multi-agent systems with switching topologies. IET Control Theory Appl 2019; 13(11): 1761–1772.

27.

Zhang

Wei

. Delay-dependent exponential stability of cellular neural networks with time-varying delays. Chaos Soliton Fract 2005; 23(4): 1363–1369.