A Novel Disturbance Observer for Multiagent Tracking Control with Matched and Unmatched Uncertainties

In this paper, multiagent tracking control problem of second-order multiagent systems with unknown leader acceleration, input saturation, and matched and unmatched disturbances is investigated. An auxiliary system is constructed to approximate system position states, and a novel sliding mode disturbance observer is designed to estimate matched and unmatched uncertainties. A sliding mode disturbance observer-based control protocol is proposed by constructing a novel sliding mode manifold based on the sliding mode disturbance observer outputs. In addition, the input saturation and the unknown leader acceleration become a part of lumped uncertainties by using mathematic transformation. The lumped uncertainties estimated by the sliding mode disturbance observer are compensated by the sliding mode disturbance observer-based control protocol. Stability of the secondorder multiagent systems is guaranteed via Lyapunov method. Finally, a simulation example is proposed to exhibit advantages and availability of the developed techniques.


Introduction
Recently, cooperative control in MASs has received significant attention due to its widespread application in engineering such as unmanned aerial vehicles [1], spacecraft formation [2], and autonomous underwater vehicles [3].Among these researches, multiagent tracking control, i.e., leaderfollowing control, is extensively scattered in multiagent control, such as adaptive multiagent control [4], backstepping multiagent control [5], intermittent multiagent control [6], and hybrid multiagent control [7].The multiagent tracking control between leader and followers means that, by partial information, the followers could track a leader reference trajectory and maintain state stability.Significant works of the multiagent tracking control, which deal with different practical conditions, have been conducted in the past decades [8][9][10][11].However, performance of the MASs is usually affected by several uncertainty factors, such as input saturation, unknown leader acceleration, and unknown disturbances of multiagent systems.Therefore, it is an interesting research for designing a control protocol to obtain good performance of MASs with matched and unmatched uncertainties.
A multiagent control problem for second-order MASs in presence of uncertain dynamics and unknown external timevarying disturbances has been investigated in [12].A robust adaptive neural network controller has been developed for a multiagent tracking control of higher-order nonlinear MASs in which each follower is modeled by a higher-order integrator incorporated with unknown nonlinear dynamics and unknown disturbances [13].Moreover, methods for a single agent system with matched uncertainties are also applicable to MASs.In [14,15], a way to deal with matched uncertainties of control systems is that the uncertainties are estimated by an effective observer and compensated by a controller.In [14], an adaptive fuzzy observer design approach has been proposed for control systems with matched uncertainties.An adaptive neural network-based observer has been given in [15] in which the proposed observer could estimate external disturbances online so as to decrease computation burden of control systems with matched uncertainties.Furthermore, results proposed in [16][17][18] were only focused on the socalled matched uncertainties, which means that the matched uncertainties can only be compensated from input channels.It is well known that unmatched uncertainties are frequently encountered in various engineering systems, such as permanent magnet motor systems and magnetic levitation suspension systems.Due to the fact that the mismatched uncertainties cannot be easily compensated by control inputs, many researchers have devoted their efforts to solve the problem.In [19], an observer-based sliding mode control method has been developed to counteract mismatched disturbances.Reference [20] addressed a finite-time output control problem for control systems with mismatching uncertainties.
On the other hand, input saturation, which may severely deteriorate performance of control systems and even lead to control systems instability, is a common feature for most practical control systems [21].Furthermore, as the most important nonsmooth nonlinearity, input saturation makes control systems design more complicated.For attitude control system, actuator saturation has been considered in [22].Discrete-time double-integrator consensus control for MASs with switching topologies and input saturation has also been studied in [23].In [24], hyperbolic tangent functions have been employed to prevent input saturation, whereas saturation functions have been used in [25].In [26], a spacecraft finite-time controller with input saturation constraint has been taken into account for the first time, and a global saturated finite-time control scheme has been proposed.
In this paper, a SMDOB based control protocol is proposed for multiagent tracking control of second-order MASs with matched and unmatched uncertainties which are composed of input saturation, unknown leader acceleration, and matched and mismatched disturbance.A novel SMDOB is designed to estimate the matched and unmatched uncertainties of the MASs.An auxiliary system is constructed to approximate the second-order MASs position states.A Lyapunov method is used to show stability of the secondorder MASs.An illustrative example is constructed to verify the proposed method developed in this paper.The main contributions of this paper are summarized as follows (i) A novel SMDOB and an observer-based sliding manifold are proposed to deal with matched and unmatched uncertainties which include input saturation, unknown leader acceleration, and matched and mismatched disturbance in the second-order MASs.
(ii) Compared with literature [2,27], assumptions of nonlinear functions of all followers in the secondorder MASs are developed.
(iii) An auxiliary system is introduced to approximate the second-order MASs position states.A SMDOB based control protocol is proposed based on auxiliary system states.
This paper is organized as follows: Section 2 details the problem formulation and graph theory.Two useful lemmas are also given in this section.The main result of our research is elaborately presented in Section 3. A numerical simulation is shown to verify the effectiveness of the proposed method in Section 4. Finally, the paper is concluded by Section 5.

Problem Description and Preliminary
2.1.Algebraic Graph Theory.For a multiagent system with  connected agents, information can be transmitted between neighboring agents, so it is natural to describe the topology of the information flow by a weighted graph [28].Lemma 1 (see [28]).If G = {,,A} is an undirected connected graph, then the Laplacian matrix L is a symmetric matrix and its  real eigenvalues can be arranged in an ascending order where   = max 1⩽⩽ {  } and  2 is called the algebraic connectivity, which is used to analyze the rate of consensus convergence.What is more, the matrix L + B associated with  → G is also symmetric and positive definite.

Problem Description and Mathematical Preliminaries.
A second-order multiagent system (MAS) with an active leader and  followers is considered.A leader dynamic equation labeled by 0 is expressed as follows: where  0 , V 0 ∈ R  are leader position and velocity vectors, respectively;  0 ∈ R  is the leader's acceleration.An th follower's dynamic equation is described as where  = 1, . . ., ;   , V  ∈ R  denote the follower's position and velocity states, respectively;   (  , V  ) ∈ R  are known nonidentical functions; Δ  ∈ R  stand for matched uncertainties which come from the inherent system;   ∈ R  are the MAS input vectors.For simplicity, we will ignore declaration expressions (  , V  ) and  = 1, . . .,  in this paper.
To proceed with the design of disturbance observer, the following lemma is given.
Lemma 2 (see [29]).The following perturbed nonlinear differential equation is shown as follows: where () ∈ R  is a solution of ( 5), () is unknown bounded disturbance, and  0 is the upper value of the derivative of time-varying disturbance |()|.A solution () of ( 5) and its derivative ẋ () will converge to zeros in finite time   if parameters satisfy that  1 ≥ 1.5√ 0 ,  2 ≥ 1.1 0 .In addition, the convergent time   is decided by where (0) is the initial value of ().

Main Result
Our control objective is to design an appropriate control protocol for a second-order MAS with matched and unmatched uncertainties to make followers position and velocity states stability and tracking errors ‖  () −  0 ()‖, ‖V  () − V 0 ()‖ arbitrary small.

Control Protocol Design Based on a Novel SMDOB.
Here, a SMDOB is designed to estimate the effect of MAS uncertainties.A SMDOB based control protocol is proposed for a MAS to achieve tracking objective.Meanwhile, assume that all the MAS states can be measured.An assumption, which always holds in practical applications, is given as follows.
In general, the leader position keeps changing throughout the entire motion process and its behaviour is independent of its followers.What is more, for th follower of a MAS, matched uncertainty Δ  satisfies that ‖Δ  ‖ ⩽ Δ  , where Δ  is a positive scalar.
Remark 4.Many practical systems can be described as the second-order MAS (4), such as spacecraft attitude dynamics (with some mathematical transformation), satellite orbital control system, and robotic dynamics.Assumption 3 is a common assumption used in most literature works, which means finite available energy of  0 and bounded change rate of uncertainties Δ  .
The th follower velocity takes the following ideal form [30]: where   is the element of Laplacian matrix L;  > 0 is the tracking control gain, which is used to adjust the convergence speed.From [30], it is obtained that V  is considered as a reference signal for each follower.Moreover, if the conditions V  → V  are satisfied, then the control objectives   →  0 and V  → V 0 are also obtained.State tracking error variables are introduced.
By invoking the tracking state errors  V into the secondorder MAS (4), a tracking error dynamic equation is obtained as follows: where ) are known function vectors and Δ  = Δ  −  0 are total uncertainties.
In order to estimate the uncertainties Δ  , an auxiliary dynamic equation is introduced as where  V , Δ F ∈ R  are the estimation of the tracking error  V (8) and the uncertainties Δ  , respectively.

Mathematical Problems in Engineering
Define the estimation errors  V =  V −  V .Considering ( 9) and ( 10), one has the following.
Inspired by [31] Furthermore, note that  1 ,  1 should be large enough to make the estimation errors   = ṡ V − ŝ as small as possible.Thus, via tuning these parameters properly, estimation errors   are limited to be small enough, which means that ŝ converge into a neighborhood of actual states ṡ V .
From Lemma 2, a novel SMESO is constructed as follows: where   =  V + ŝ .By (11), the following is shown.
From Lemma 2, if the conditions that a positive constant and observer parameters  2 ,  3 such that  2 ⩾ 1.5√ 1 and  3 ⩾ 1.1 1 are satisfied, then it is obtained that   and ṡ  will converge to zeros in finite time.From (11), we can also conclude that the SMDOB output Δ F will approach its real value Δ  .Meanwhile, it also means that ‖ V ‖ and ‖ ṡ V ‖ would converge to zero in finite time according to (15).Remark 6.From [29], we can know that the estimation error  1 can converge to a bounded region, and its derivative  1 could also converge to zeros.
With Δ F , which is estimated by the SMESO (13), a SMESO based control protocol ( 16) is proposed as follows: where   > 0 is a designed control gain.
The following theorem summarizes the feasibility of the SMESO control protocol (16).
Theorem 7. Based on the SMESO (13) and the proposed control protocol (16), positions of the second-order MAS (4) with matched uncertainties will track the desired trajectory  0 .Proof.To facilitate the stability analysis, the following Lyapunov function is constructed.
With the auxiliary dynamic system (10), the SMESO (13), and the proposed control protocol ( 16), the time derivative of  is shown as follows.
According to (18), it is shown that all system variables are stable, i.e.,  V → 0 as  → ∞.Since  V and ṡ V are convergent in finite time, it is obtained that  V → 0, V  → V  by considering  V =  V −  V ,  V = V  − V  .Moreover, from (8), the distributed tracking objectives   →  0 , V  → V 0 have also been achieved.This concludes the proof.Remark 8.For traditional extended disturbance observers, ṡ V involve the term of observer output Δ F which includes tedious analytic derivatives computation and increases system computation burden.To avoid this, a novel SMDOB is employed to estimate ṡ V .However, it should be noted that the estimation errors are unavoidable with the application of SMDOB.On the other hand, it should be also emphasized that the undesired effect of the unknown leader acceleration  0 is combined with uncertainties Δ  as compounded uncertainties.
Remark 9.In the SMDOB, derivatives of disturbances are required to be bounded.While taking Assumption 3 and the estimation of the ESO into consideration, it is obtained that the bounded assumption of the time derivative of total lumped disturbance is reasonable.
Remark 10.By the definition of graph Laplacian matrix L, it follows that   ̸ = 0 if and only if there is information exchange between th follower and th follower.For the diagonal matrix B,   ̸ = 0 if and only if there is information exchange between the leader and the th follower.Therefore, the proposed control protocol can only use the information of its neighbors.Hence, the proposed protocol belongs to the decentralized design fashion with directed communications.

Protocol Design for Unmatched Uncertainties.
In practical engineering applications, MASs may suffer from unmatched uncertainties which always affect states directly rather than through the input channels.Compared with traditional matched disturbance observer-based control approaches, one of the intuitive difficulties induced by unmatched uncertainties is that these cannot be compensated in input channels directly.On the other hand, as one of the most important nonsmooth nonlinearity properties, input saturation should be explicitly considered in practical engineering.Here, we consider the multiagent tracking control problem under matched and unmatched uncertainties which include input saturation.
Followers' dynamic equations with input saturation and matched and unmatched uncertainties are considered as where Δℎ  , Δ  ∈ R  are the unmatched and matched uncertainties, respectively;   ∈ R  denotes unknown nonsymmetric saturation input.
The unknown nonsymmetric saturation input   can be expressed as follows: where  = 1, . . .,  and V  = [V 1 , . . ., V  ]  ∈ R  is the designed control input command.Note that  min ≥ 0 and  max ≥ 0, which are parameters of the unknown nonsymmetric saturation input, denote the upper and lower bounds of the nonsymmetric input saturation.
For further stability analysis, the followers dynamic equations (19) can be transformed as where   =   − V  + Δ  are lumped uncertainties.Due to the fact that  max and  min are unknown, the compounded uncertainties   are also unknown.The auxiliary dynamic equation is introduced as where Δ ĥ are the estimation of the unmatched uncertainties Δℎ  .Considering dynamic equations ( 21) and dynamic equation (22), an estimation error dynamic is considered as follows.
where   = ŝ −   is an ESO estimation error.The value of ESO parameters  1 ,  2 ,  1 , and  2 are the same as (12).According to (24), it is obtained that where  V ∈ R  is the estimation error.Due to the property of the DOB,  V could converge to an arbitrary small range by selecting appropriate parameters.
Defining   =   + ŝV , a novel SMESO is proposed as follows.
Similar to (7), the followers tracking signal, with input saturation and matched and unmatched uncertainties estimation Δ ĥ , yields the following.
Inspired by [19], a sliding mode manifold equation based on a disturbance observer is defined as follows: where  V = V  − V  are reference signal tracking errors.
Similar to (22), in order to estimate the unknown lumped uncertainties Δ  , an auxiliary system and a SMESO are proposed as where With the SMDOB providing the required estimation in (32), the control protocol is proposed as where   > 0 is the designed control gain.
The following theorem summarizes the proposed control protocol for the MAS (19) under matched and unmatched uncertainties which include input saturation and unknown leader acceleration.Theorem 11.Consider the multiagent tracking control of second-order MAS (19) in presence of matched and unmatched uncertainties which include input saturation and unknown leader acceleration.A finite-time converging performance is obtained according to a SMDOB (32) and a control protocol (33).Then, the MAS ( 19) is stable and the tracking error will converge to a bounded set as time going.Proof.To analyze stability of the MAS (19), we consider the following Lyapunov function candidate.
By the SMDOB (32) and the proposed control protocol (33), the time derivative of  is as follows.
Based on above analysis, the following is shown.
Hence, we can obtain that all followers can track the leader trajectory agreement with bounded error region.This concludes the proof.
Remark 12.In order to handle unmatched uncertainties, a novel sliding mode manifold is defined based on the SMESO (26).Compared with traditional sliding mode control, integral sliding mode control, and disturbance observer-based sliding mode control in [19], convergent accuracy of the SMESO ( 26) is dependent on the estimation error ‖Δℎ  −Δ ĥ ‖ rather than ‖Δℎ  ‖.Since the unmatched uncertainties have been precisely estimated by the SMESO (26), the magnitude of the estimation error ‖Δℎ  − Δ ĥ ‖, which is expected to converge to the neighbor region of zeros, can be kept much smaller than the magnitude of the uncertainties ‖Δℎ  ‖.It means the tracking trajectory could have the property of chattering reduction as well as excellent dynamic and static performance.The readers can also refer to [32][33][34] for the same argument.
Remark 13.For (21), the effects of unknown nonsymmetric input saturation are treated as a part of the lumped uncertainties and approximated by using the SMDOB (26).However, the system state has the feasibility of unknown nonsymmetric input saturation, and the bounded state tracking errors are still guaranteed by Lyapunov theory.Remark 14.The introduced auxiliary system, working together with the SMDOB (26) and the ESO (24), not only improves the control performances, but also reduces real-time computing burden of the MAS.With selecting appropriate parameters, the ultimate convergent sets of state tracking errors   −  0 can be tuned.Careful analysis indicates that increasing control gains and disturbances observer parameters could contribute to faster converging speed.Moreover, larger controller gains and observer parameters also lead to larger control power.Therefore, a compromise between control objective and converging speed should be made in practical problem.

Simulation
In this section, a numerical simulation is presented to show the effectiveness of the proposed theorem.The simulation scenario is constructed as follows.
Considering a formation with four followers and a visual leader, the communication graph is shown in Figure 1.
Parameters of the SMDOB and ESO are listed in Table 1.
Figure 2 is the estimation error of the system uncertainties by using the designed observer.Based on the estimation of system uncertainties, the desired tracking performance and tracking errors for MASs are shown in Figures 3 and  4. Note that Figure 5 is the sliding surface.The evolution  of the saturation control inputs is illustrated in Figure 6.
Although, there exist the nonsymmetric input saturation and the time-varying system uncertainties, the system tracking performance is still satisfactory and the tracking errors converge to bounded regions.From the simulation results, we can obtain that the developed SMDOB control protocol is valid.The proposed protocol can force all agent states following the given desired trajectory; even only a subset of group members has access to the desired signal.

Conclusion
In this paper, a multiagent tracking control problem was discussed to make the networked agents achieve tracking objective under matched and unmatched uncertainties, which include input saturation and unknown leader acceleration.A novel SMDOB was proposed to estimate system uncertainties.An auxiliary system was constructed to approximate the follower dynamic.With the aid of designed observer-based sliding manifold, a feedback-type protocol was proposed and the stability conditions were also derived.In contrast to existing results on this aspect [19], the converging accuracies depended on the estimation errors ‖Δℎ  − Δ ĥ ‖ rather than ‖Δℎ  ‖.Moreover, the simulations illustrated good performance of the proposed SMDOB based protocol under complicated conditions.
The developed theoretical results can provide new insight into the studies of distributed tracking control for MASs in presence of complicated constraints.This protocol can be applied in some practical systems, such as robotic dynamic, satellite orbital control system.
In the future, there are still some interesting problems on this topic for further study, such as directed communication topology, time delay, and packet dropout.
is a derivative of  V ;  1 ,  1 ,  2 ,  1 ,  2 are observer parameters.If initial conditions  V (0)−ŝ V (0) and ṡ V (0)−ŝ  (0) are bounded, then ŝ can approximate derivative term ṡ V to any arbitrary accuracy.Stability of the ESO (12) has been obtained by selecting appropriate parameters  1 and  1 .Fundamental selections of parameters can be chosen as 1 =   + ζV are introduced auxiliary variables and Δ Ĥ is the estimation of the unknown lumped uncertainties Δ  .Considering Assumption 3 and estimation property of SMDOB, it is shown that ‖ ε  − ḋ  − ε  ‖ ⩽  2 with   = ζ  − ζ being the SMDOB approximation error.According to Lemma 2, one has that   and its derivative ζ  converge to zeros in finite time if SMDOB parameters satisfy  5 ⩾ 1.5√ 2 and  6 ⩾ 1.1 2 .Thus, the SMDOB estimation Δ Ĥ will converge to Δ  in finite time.
; note that ζV and ζ are estimations of ζ  and ζ  .