Furthermore, a dynamic leader is introduced to guide a
group of agents and treated as an ordinary agent by each agent
called a follower. Each follower can sense the information of the
dynamic leader if only if the distance between them is less than
the communication radius. The motion of the dynamic leader
is described by
xv
ll
=
, where x
l
∈ R
2
and v
l
∈ R
2
denote the
position and velocity vectors of the dynamic leader. Here,
without loss of generality, it is assumed that
vf
l
1
<
, where f is
a positive constant.
The control objective here is to derive a set of bounded
distributed controllers using only local information to steer the
followers to achieve velocity consensus and collision avoidance
with the dynamic leader with time-varying velocity, while guar-
anteeing the connectivity of the underlying communication
graph as the system evolves, provided the given graph is initially
connected.
III. MAIN RESULTS
3.1 Leader-follower flocking algorithm with
connectivity preservation
It is worth noting that the algorithms in [8,25,26,30] are
feasible only when each follower can access the accurate accel-
eration information of the dynamic leader, i.e.,
v
l
. However,
since not all the agents (robots, air vehicles, manipulators, etc.)
in practice are equipped with acceleration sensors, acceleration
measurements are more difficult to obtain than position and
velocity measurements. Moreover, an algorithm in the absence
of acceleration measurements has the advantage of decreasing
equipment cost and network traffic. Therefore, we are motivated
to design distributed connectivity-preserving leader-follower
flocking algorithms without using acceleration measurements.
To achieve the desired flocking motion, the explicit
bounded flocking control protocol for each follower i is devised
as follows:
uVxhVx
aavv
ixij
j
jl
ij i x
i
il il
j
ij ik i
i
i
i
=− ∇ − ∇
−−
∈
≠
∈
∑
∑
N
N
() ()
sgn (
α
kk
k
kl
ii l
ij
j
jk j k
k
i
i
hv v
aavv
)( )
sgn ( )
∈
≠
∈∈
∑
∑
+−⎧
⎨
⎩
⎫
⎬
⎭
⎧
⎨
⎩
⎫
⎬
⎭
+−
N
N
α
NN
j
kl
jj l
hv v
≠
∑
+−⎧
⎨
⎩
⎫
⎬
⎭
⎧
⎨
⎩
⎫
⎬
⎭
()
(3)
where
ht
it
i
l
()
,()
,
=
∈
⎧
⎨
⎩
0
1
N
otherwise
sgn(·) is the signum function,
α > 0 is the control gain, V
ij
,
∀∈j
i
N
is the bounded interactive
APF between agents i and j which is to be designed.
Remark 1. Note that, inspired by [27,35], the sliding mode
control approach is adopted here to serve as a distributed esti-
mator for all the followers without acceleration measurements to
the dynamic leader. The asymptotic stable flocking behavior of
the entire system can be guaranteed by properly adjusting the
control gain α, the determination of which will be detailed later.
The distinguished features of the proposed control protocol (3)
lie in the removal of the mild connectivity requirement in [27,35]
and the bounded control force.
Then, define the positive semi-definite function energy
function as
ψ
(, , , ) ( )( )
(, )
xvxv vv vv
Uxx
ll i l
T
il
i
N
il
i
N
=−−
+
=
=
∑
∑
1
2
1
2
1
1
(4)
where
Uxx V x hV x
i l ij ij
j
jl
iil il
i
(, ) ( ) ( )=+
∈
≠
∑
N
2
Note that U contains all the existing follower-follower potentials
and the leader-follower tracking potentials, which have the
physical meaning of characterizing all the interactive potential
energy of the entire system.
Further, define Ψ
max
which satisfies
ψ
max
max
(() ())(() ())
()
=− −
+
+
=
∑
1
2
00 00
1
2
1
vv vv
NN
V
il
T
il
i
N
(5)
Remark 2. Note that the first term and the second term in (5)
indicate the initial kinetic energy and the possible maximum poten-
tial energy of the system, respectively. The combination of both
giv es the total maximum mechanical energy and the overall leader-
follo wer multi-agent system in the context of the complete undi-
rected graph, which is vital for designing the bounded APF V
ij
.
In order to enable the overall system to achieve desired
stable flocking motion using only bounded control inputs, V
ij
(||x
ij
||)
should be well designed to be a bounded and nonnegative poten-
tial of the distance of ||x
ij
|| = ||x
i
− x
j
|| while integrating require-
ments of connectivity maintenance and collision avoidance, such
that:
1. V
ij
(||x
ij
||) is continuously differentiable for ||x
ij
|| ∈ (0, R);
2. V
ij
(||x
ij
||) is monotonically decreasing for ||x
ij
|| ∈ (0, d)
and monotonically increasing for ||x
ij
|| ∈ (d, R), where
ε
1
< d < R − ε
2
and
3. V
ij
(0) = Ψ
max
and V
ij
(R) = Ψ
max
.
ψ
max
max
(() ())(() ())
()
1
2
00 00
1
2
1
vv vv
NN V
il
T
il
i
N
−−
+
+
=
∑
(6)
where V
max
= max{V(ε
1
), V(R − ε
2
)} and
ε
1
0
2
0=
∈
min { ( ) }
,()ij
ij
x
E
.
3
Y. Mao et al.: Bounded Connectivity-Preserving Leader-Follower Flocking Algorithms Without Acceleration Measurements
© 2014 Chinese Automatic Control Society and Wiley Publishing Asia Pty Ltd