Hierarchical Decomposition Based Distributed Adaptive Control for
Output Consensus Tracking of Uncertain Nonlinear Systems*
Wei Wang
1
, Changyun Wen
2
, Zhengguo Li
3
, Jiangshuai Huang
2
Abstract— In this paper, we aim to design distributed adap-
tive controllers for output consensus tracking of multiple
nonlinear subsystems with intrinsic mismatched unknown pa-
rameters. The graph representing the communication status
among subsystems is assumed to have directed and fixed
topology. Only a small percentage of the subsystems can
obtain the desired trajectory information, which is regarded
as a virtual leader node added to the original communication
graph. We first split the communication graph into a hier-
archical structure according to the shortest possible path of
each subsystem originated from the virtual leader. Then local
adaptive controllers for subsystems in different layers can be
designed in a sequential order. By introducing the estimates of
the uncertainties of its neighbors located in the upper layer
into the local controller of a subsystem, the transmission of
parameter estimates among connected subsystems is avoided.
It is proved that output consensus tracking of the overall system
can be achieved asymptotically and all closed-loop signals are
ensured bounded. Simulation results show the effectiveness of
our scheme.
I. INTRODUCTION
In the past decades, distributed cooperative control of
a collection of multiple dynamic subsystems (also called
multi-agent systems) has attracted significant attention due
to a wide range of potential applications in mobile robot
networks, surveillance and reconnaissance systems, intel-
ligent transportation management systems, etc. To reach
consensus for some variables of the group members is a
typical control objective and several effective algorithms
have been proposed in this area. These approaches can
be roughly classified as constant consensus solutions ([1]–
[3]) and non-constant consensus solutions of tracking time-
varying trajectories ([4]–[6]). Many of early works were
devoted to investigating consensus algorithms for the sys-
tems with single-order dynamics, whereas more results by
considering systems with second or higher-order dynamics
have been reported only in recent years such as [7]–[10].
Apart from these, some other interesting topics have also
appeared, including finite-time consensus, consensus with
limited communication situations, about which a compre-
hensive overview can be found in [11].
* This work was supported by the National Natural Science Foundation
of China under Grant No. 61203068, 60974059.
1
W. Wang is with the Department of Automation, Tsinghua University,
Beijing, China 100084 wwang28@tsinghua.edu.cn
2
Changyun Wen and Jiangshuai Huang are with the School of Electrical
and Electronic Engineering, Nanyang Technological University, Singapore,
639798 ecywen@ntu.edu.sg,jhuang2@e.ntu.edu.sg
3
Zhengguo Li is with Signal Processing Department, Institute for Info-
comm Research, Singapore 138632 ezgli@i2r.a-star.edu.sg
All the results mentioned above were developed by as-
suming that precise intrinsic dynamics are available. How-
ever, such assumptions can hardly be satisfied especially
when complicated controlled plants are considered. Actually,
consensus control of multi-agent systems in the presence
of intrinsic model uncertainties has become a new issue
of interest in academic communities. In [12]–[14], intrin-
sic uncertainties including unknown parameters, unmodeled
dynamics and exogenous disturbances were handled with
by incorporating robust control techniques. Adaptive control
has also been proved as an promising tool in dealing with
this issue. In [15], a distributed consensus control strategy
based on model reference adaptive control (MRAC) tech-
nique was proposed for a team of linear subsystems with
unknown system parameters under an undirected commu-
nication mechanism. In [16], adaptive consensus tracking
controllers were designed for Euler-Lagrange swarm systems
with non-identical dynamics, unknown parameters and com-
munication delays. It was assumed that the desired trajectory
information was accessible to all subsystems. In [17], a
distributed neural adaptive control protocol was proposed for
a group of first-order nonlinear subsystems with unmodeled
dynamics, where the reference state is only available to
a subset of subsystems. By assuming the boundedness of
the basis neural network (NN) activation functions, global
uniform ultimate boundedness (GUUB) of all the signals in
the network under directed communication mechanism and
bounded consensus were achieved if sufficiently large local
control gains were chosen. The results were then extended to
more general class of systems with second and higher-order
dynamics in [18] and [19]. In [20] and [21], alternative dis-
tributed adaptive control design approaches were presented
for similar uncertain first-order systems based on information
exchange of local consensus errors and weighted one-to-one
consensus errors, respectively. With these methods, the inher-
ent nonlinear functions were not assumed bounded, however
undirected communication conditions were required. More
results on distributed adaptive control of multi-agent systems
could be found in [22]–[24].
In contrast to these results, output consensus tracking
problem for high-order parametric strict feedback systems
with mismatched parametric uncertainties was investigated
[25]. By including consensus reference estimates in the
subsystems which are un-noticed of the desired trajectory,
bounded output consensus tracking of the whole system
was ensured. However, it is not easy to check whether
the sufficient condition in the form of LMI is satisfied.
Moreover, the online parameter estimates of the neighbors