Aerospace Science and Technology 82–83 (2018) 105–118

Contents lists available at ScienceDirect

Aerospace Science and Technology

www.elsevier.com/locate/aescte

Cooperative aerial lift and manipulation (CALM)

Hossein Rastgoftar ∗ , Ella M. Atkins
Aerospace Engineering Department, University of Michigan, Ann Arbor, MI, USA

a r t i c l e i n f o a b s t r a c t

Article history: This paper proposes a novel paradigm for aerial payload transport and object manipulation by an
Received 30 April 2018 unmanned aerial vehicle (UAV) team. This new paradigm, called cooperative payload lift and manipulation
Received in revised form 4 September 2018 (CALM), applies the continuum deformation agent coordination approach to transport and manipulate
Accepted 4 September 2018
objects autonomously with collision avoidance guarantees. CALM treats UAVs as moving supports during
Available online 10 September 2018
transport and as stationary supports during object manipulation. Constraints are formulated to assure
Keywords: sufficient thrust forces are available to maintain stability and follow prescribed motion and force/torque
Scalability profiles. CALM uses tensegrity muscles to carry a suspended payload or a manipulation object rather
Customizability than cables. A tensegrity structure is lightweight and can carry both the tension and compression forces
Manipulation required during cooperative manipulation. During payload transport, UAVs are categorized as leaders and
Locomotion followers. Leaders define continuum deformation shape and motion profile while followers coordinate
Planning through local communication. Each UAV applies input–output (IO) feedback linearization control to track
Unmanned aerial vehicles
the trajectory defined by continuum deformation. For object manipulation, the paper proposes a new
hybrid force controller to stabilize quadcopters when smooth or sudden (impulsive) forces and moments
are exerted on the system.
© 2018 Elsevier Masson SAS. All rights reserved.

1. Introduction
Natural disasters such as earthquakes, wildfires, and hurricanes
have created devastating disaster zones where survival relies on
emergency responder support. Access roads in disaster zones may
be covered with debris or destroyed, making it difficult for sur-
vivors to access supplies such as clean water, food, and medicine
even after the supplies have been delivered to the greater re-
gion. Small UAVs (unmanned aerial vehicles) offer an alternative
to manned helicopters that can act as a force multiplier for supply
delivery in disaster zones. A team of small UAVs can carry an ap-
preciable package and can be flexibly deployed from local supply
stations to maneuverably transport payload to people in need.
This paper develops the underlying theory to enable cooperative
payload transport in support of disaster relief and other missions
requiring payload carriage through a constrained environment with
unimproved launch and delivery sites. Because a small UAV team
will need to potentially pickup and manipulate its payload at each
remote site, this paper also develops a theory for cooperative object
manipulation. Because cable and tether systems can only support
tension loads, this work presumes a lightweight tensegrity struc-
* Corresponding author.
E-mail addresses: hosseinr@umich.edu (H. Rastgoftar), ematkins@umich.edu
(E.M. Atkins).
ture replaces the cable connecting UAV and payload so that the
UAV team can provide the required tension and compression forces
to achieve the required payload (grasping) manipulation forces and
torques. The proposed multi-UAV system provides a comprehen-
sive capability we denote Cooperative Aerial Lift and Manipulation
(CALM).
Manipulation Robots (MRs) are now able to accomplish a va-
riety of tasks that include but are not limited to grasping, push-
ing, sliding, tipping, rolling, and throwing. Although MRs can have
multiple degrees of freedom, the number of robots collaboratively
directing a single manipulation task has been limited in previous
work. Furthermore, most MRs require grounding or connection to a
grapple fixture. Our goal is to assemble and coordinate many UAVs
as stable supports for aerial object manipulation in environments
difficult to access or stabilize with ground-based manipulation. In
this manner, objects can be grappled, released, and manipulated,
enabling the UAV team to clear debris, pick up and deploy supply
packages, or perform repairs or construction tasks given appropri-
ate manipulation tooltip designs.
Relevant background and contributions are presented below.
Mathematical preliminaries are followed by a problem statement
and specification for the CALM system. Transport and stationary
support MUS (multi UAV system) control strategies are presented
and analyzed, and simulation results are presented to illustrate and
evaluate the CALM specification.

https://doi.org/10.1016/j.ast.2018.09.005
1270-9638/© 2018 Elsevier Masson SAS. All rights reserved.
106 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118
1.1. Background
Cooperative control of multiple unmanned aerial vehicles (UAVs)
has been an active area of research for over two decades. For-
mation flight [1], air traffic control [2], transportation engineering
[3], aerobiological sampling of agricultural lands [4], cooperative
manipulation [5] and general team-based surveillance are some
applications of UAV cooperative control. A UAV team can improve
mission efficiency, reduce cost, and offer increased resilience to
failures including individual UAV loss. Group cooperation also im-
proves fault detection and ability for the team to recover from
anomaly conditions [6,7]. Coordination of a multi-agent system
(MAS) using Laplacian control has been considerably studied over
the last decade [7]. Laplacian control methods include consensus
algorithms [8–12] and partial differential equation (PDE) methods
[13–15]. A containment approach [16–20] to coordination has also
been proposed. Consensus is the most common approach used for
distributed control in multi-agent systems. Consensus has been
applied for distributed motion control [21–24], distributed sensing
[25,10], medical applications [26], and power distribution systems
[27,28].
In containment control multiple leaders guide the MAS toward
a desired target shape using consensus to update positions [29,
30,19,16]. Go-Stop hybrid control strategies in [20,30] enable the
leaders to guide followers to a desired target. Containment con-
trol under fixed and switching MAS communication topologies are
developed in [31,32]. Containment control of a MAS formation
modeled as double integrators was presented in [18]. Addition-
ally, containment control with complex communication weights
was investigated in [18]. MAS containment control under directed
communication topology was studied in [33,34], event-based con-
tainment control of multi-agent systems was defined in [35,31],
and finite-time containment control of a multi-agent system is
studied in [36,37].
UAVs have been applied in payload transport, grasping and ma-
nipulation tasks. These tasks require UAVs to manage total forces
and moments applied to external object(s) along with their col-
lective motions. Manipulation can be used to achieve perching or
deploy/pickup payloads, or the cooperative team can carry slung
loads [38,39]. Swing-free trajectory tracking is demonstrated by a
single quadcopter carrying a payload in [40]. Aerial manipulation
using a single quadcopter is studied in [41]. Ref. [42] shows how
multiple quadcopters can participate in a cooperative manipulation
task. Stabilization of a helicopter carrying a payload is studied in
[43,44]. In Ref. [45,46] stabilization of a single quadcopter carrying
a single payload is analyzed. The connecting cable between each
UAV and the slung payload is modeled by serially-connected bar
elements in [45]. Ref. [46] assumes that the cable is active only if
it is taut. Ref. [46] proposes a differentially flat hybrid controller to
stabilize a slung payload with a single quadcopter.
Modeling and control of multiple UAVs deployed for coopera-
tive manipulation are investigated in [47,41,5], while cooperative
grasping of a payload using multiple UAVs is studied in [48,47].
Adaptive control for a slung load transport mission using a single
rotorcraft is studied in [49–51]. Stability of a slung load carried by
multiple single-rotor helicopters is investigated in [52], while an
inverse kinematics formulation for aerial payload transport is pre-
sented in [53]. Cooperative manipulation and transportation with
three quadcopters are modeled and experimentally demonstrated
in [3]. Refs. [54,55] suggests a geometric control scheme to control
payload transport with multiple UAVs.
1.2. Contribution
The proposed Cooperative Aerial Lift and Manipulation (CALM)
system offers payload transport and object manipulation capabil-
ities. The multi-UAV (unmanned aerial vehicle) system (MUS) and
tensegrity [56] structure tethering system are the main CALM ele-
ments. The MUS contains a large number of small unmanned aerial
vehicles (UAVs) that provide moving supports for payload trans-
port. Each tensegrity arm is comprised of a finite set of tensegrity
muscles. A tensegrity muscle consists of a finite number of tenseg-
rity prism cells, where each prism cell is made of tethers and rigid
bars. A tensegrity arm is lightweight but sufficiently stiff to carry
tension and compression forces.
The first contribution of this paper is to characterize and en-
hance manipulation deformability and scalability. The paper ap-
plies the principles of continuum mechanics to coordinate CALM
UAVs in a three-dimensional motion space [57]. Because contin-
uum deformation is homeomorphic, no two separate particles of
a continuum (or deformable body) occupy the same position in
a continuum deformation, and inter-agent distances can signifi-
cantly change. Treating UAV coordination as continuum deforma-
tion, we formally verify safety by providing inter-agent collision
avoidance guarantees [57]. By ensuring inter-agent collision avoid-
ance through continuum deformation coordination, a large number
of UAVs can participate in a CALM mission. Increasing the number
of support UAVs offers the following advantages: (i) Manipulation
cost is manageable (low) while efficiency and robustness increase.
(ii) Fault-tolerance and reconfiguration capabilities in a cooper-
ative manipulation mission are improved. (iii) UAVs collectively
carry nontrivial payloads while MUS coordination maneuverability
is improved. (iv) Because inter-agent distances can be expanded
or contracted via continuum deformation, the UAV team can navi-
gate constrained environments including passing through a narrow
channel. This paper shows how a desired continuum deformation
can be acquired by support UAVs in real-time through local com-
munication with minimal computation overhead.
CALM supports customized operation modes of the MUS, offer-
ing Customizability as the second paper contribution. A MUS can
act either as moving supports in a payload transport mission or as
stationary supports in an object manipulation scenario. The pro-
posed CALM system can successfully accomplish two main tasks in
a cooperative fashion that include:
• Transportation tasks in which CALM carries a common teth-
ered payload from an initial location to a target destination. In
a transportation mission, support UAVs must be able to carry
rigid and deformable payloads and avoid collision with ob-
stacles. This work presumes a tensegrity structure connecting
each UAV to the common payload.
• Force control mode in which external forces exerted on UAV
supports can be either smooth or impulsive. The CALM force
control system must deal with discontinuous forces including
contact and impulse forces, as well as continuous forces. We
propose a hybrid control model to stabilize MUS when they
act in stationary support model for an object manipulation
task.
Decentralized payload transport is the third contribution of
the paper. Compared to related work [54], one CALM advantage
is that agent coordination is acquired through local communica-
tion in real-time with less computation cost. To the best of our
knowledge, available cooperative payload transport work relies on
centralized coordination approaches. Ref. [54] applies the Lagrange
method to calculate quadcopter desired thrust force in a cen-
tralized fashion. Given a desired trajectory for the slung payload,
Ref. [54] then applies the virtual structure method to coordinate
UAVs. Our work instead assumes that each UAV is equipped with a
force sensor to measure and indirectly regulate tensegrity muscle
force at any time t. Given force sensor readings, each UAV assigns
 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118 107

the control thrust force in real-time to achieve a desired trajectory This paper assumes the payload is a rigid body with an arbi-
assigned by local communication. trary orientation in a 3-D motion space. Therefore, payload con-
This paper is organized as follows. Preliminary graph theory no- figuration is specified at any time t only by knowing three unique
tions, a background on continuum deformation coordination, and points of the payload known as payload characteristic points. Pay-
position notations are presented in Section 2. The problem state- load characteristic points are denoted O 1 , O 2 , and O 3 . The ac-
ment in Section 3 is followed by the CALM system mathematical tual position of payload characteristic point O k is denoted r O k =
model in Section 4. CALM control systems including position con- x O k ê1 + y O k ê2 + z O k ê3 (k = 1, 2, 3). Payload characteristic points
trol and hybrid force control (HFC) are modeled in Section 5. In form a rigid triangle in the motion space.
Sections 6 and 7, we describe how quadcopters can act as moving
supports in a payload transport task, or as stationary supports in 2.3. Continuum deformation background
an object manipulation scenario. Stability of MUS-payload system
is discussed in Section 8. Simulation results presented in Section 9 2.3.1. Continuum deformation definition
are followed by concluding remarks in Section 10. By definition, a continuum is a domain consisting of an infi-
nite number of particles with infinitesimal size. [58]. This paper
2. Preliminaries considers a class of continuum deformation called homogeneous
transformation, given by
2.1. Graph theory notions
Yi , H T (t ) = Q (t )Ri + D(t ), (4)
Let collective motion in a 3D space be decomposed into a 2D
3×3
continuum deformation in the X –Y plane and a 1D continuum where Q ∈ R is the Jacobian matrix, ri , H T (t ) is the global
deformation along the Z axis. The graph G defines inter-agent desired position of quadcopter i initially positioned at Ri . Also,
communication in the X –Y plane. The graph G = G (V , E , W ) is a D = [ D 1 D 2 D 3 ] T is the rigid-body displacement vector. This pa-
weighted digraph defining inter-agent
 communication
 for a quad- per assumes the MQS deforms in the X –Y plane, so
copter team. Further, V = 1, 2, · · · , N is the node set, where ⎡ ⎤
i ∈ V represents a quadcopter’s index number. Three quadcopters,   Q 1,1 Q 1,2 0
Q CD 0
defining a 2D continuum deformation, are called leaders. The re- Q = = ⎣ Q 2,1 Q 2,2 0⎦ . (5)
maining quadcopters are called followers.
0 1
 Leader quadcopter index 0 0 1
numbers are defined by V L = 1, 2, 3 , and follower quadcopter
index numbers are defined by V F = {4, 5, · · · , N }. The edge set Elements of Q CD ( Q 1,1 , Q 1,2 , Q 2,1 , Q 2,2 ), D 1 and D 2 can be
E ⊂ V × V defines communication among quadcopters. The set defined based on the positions of the leaders [57]:
Ni = j  ( j , i ) ∈ EC D defines in-neighbor quadcopters of a quad- ⎡ ⎤ ⎡ ⎤⎡ ⎤
Q 1,1 (t ) X1 Y1 0 0 1 0 x1, H T (t )
copter i. Leader quadcopters are assumed to move independently.
Thus, Ni = ∅, if i ∈ V L . The set W : E → (0, 1) defines follower ⎢ Q 1,2 (t ) ⎥ ⎢ X 2 Y2 0 0 1 0 ⎥ ⎢ x2, H T (t ) ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
communication weights. ⎢ Q 2,1 (t ) ⎥ ⎢ X 3 Y3 0 0 1 0 ⎥ ⎢ x3, H T (t ) ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥. (6)
⎢ Q 2,2 (t ) ⎥ ⎢ 0 0 X1 Y1 0 1 ⎥ ⎢ y 1, H T (t ) ⎥
⎣ D (t ) ⎦ ⎣ 0 0 X2 Y2 0 1
⎦ ⎣ y 2, H T (t )
⎦
2.2. Position terminology 1
D 2 (t ) 0 0 X3 Y3 0 1 y 3, H T (t )
It is assumed that position is expressed with respect to a fixed
ground coordinate system. The unit bases of the ground coordinate Polar decomposition of the continuum deformation Jacobian matrix
system are mutually orthogonal and denoted ê1 , ê2 , and ê3 . The Using polar decomposition, Q CD can be expressed as
actual position of quadcopter i is denoted ri = xi ê1 + y i ê2 + zi ê3 .
Initial position of quadcopter i is denoted by Ri = X i ê1 + Y i ê2 + Q CD = RCD U CD , (7)
Z i ê3 . We define where U CD is a positive definite (and symmetric) matrix and RCD
is an orthogonal matrix, e.g RCDT
RCD = I 2 . Eigenvalues of matrix
Yi = xi ê1 + y i ê2 (1)
U CD are positive and real and denoted by λ1 and λ2 (0 < λ1 ≤ λ2 ).
as the actual output of quadcopter i. The global desired output
of quadcopter i is given by a continuum deformation in the X –Y Key feature of a homogeneous transformation: It is assumed that the
plane and denoted by three leaders are non-aligned at all times t. Therefore,
 
Yi , H T = xi , H T ê1 + y i , H T ê2 . (2) ∀t ≥ t 0 , Rank r2, H T − r1, H T r3, H T − r1, H T = 2. (8)
The local desired position of quadcopter i is given by The three leaders form a leading triangle in the X –Y plane. Under
 a homogeneous transformation, X and Y components of the global
Yi , H T i ∈ VL desired position of quadcopter i ∈ V can be expressed as in [57]:
Yd,i =  . (3)
j ∈Ni w i , j Y j i ∈ VF
   3  
xi , H T (t ) x j , H T (t )
Note that w i , j is the communication weight of follower i with t ≥ t0 , Yi , H T = = αi , j , (9)
agent j. w i , j is positive if j ∈ Ni ; otherwise, it is zero. In Sec- y i , H T (t ) y j , H T (t )
j =1
tion 2.3.2, communication weight characteristic equations are ob-
tained; communication weights are consistent with agents’ po- where αi,1 , αi,2 , and αi,3 are time-invariant parameters and αi,1 +
sitions at initial time t 0 . If followers apply the communication αi,2 + αi,3 = 1. αi,1 , αi,2 , are αi,3 are computed from quadcopter i
weights given in Section 2.3.2 a MQS continuum deformation will initial position and the three leaders as follows [57]:
be prescribed. ⎡ ⎤⎡ ⎤ ⎡ ⎤
X1 X2 X3 αi,1 Xi
⎣ Y1 Y2 Y 3 ⎦ ⎣ αi ,2 ⎦ = ⎣ Y i ⎦ . (10)
Remark. At initial time t 0 , actual, local and global desired outputs
are identical: [ X i Y i ] T = Yi (t 0 ) = Yi , H T (t 0 ) = Yd,i (t 0 ). 1 1 1 αi,3 1
108 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118

2.3.2. Continuum deformation acquisition

Continuum deformation is acquired by followers through lo-
cal communication. Inter-agent communication is defined by a
weighted graph with the following properties [57]:

1. Follower quadcopter i interacts with three  in-neighbor quad-

copters defined by the set Ni = i 1 , i 2 , i 3 .
2. In-neighbor quadcopters’ initial positions satisfy the following
rank condition:
Fig. 1. CALM operation modes.
 
Rank Ri 2 − Ri 1 R i 3 − R i 1 = 2. (11)

An example continuum deformation communication graph is

shown in Fig. 2.
3. Given initial positions of quadcopter i ∈ V F and i 1 , i 2 , i 3 , de-
noted by Ri , Ri 1 , Ri 2 , and Ri 3 , communication weights are
obtained from
⎡ ⎤⎡ ⎤ ⎡ ⎤
X i1 X i2 X i3 w i ,i 1 Xi
⎣ Y i1 Y i2 Y i 3 ⎦ ⎣ w i ,i 2 ⎦ = ⎣ Y i ⎦ . (12)
1 1 1 w i ,i 3 1

The last row of Eq. (12) ensures w i ,i 1 + w i ,i 2 + w i ,i 3 = 1. Note that

w i ,i 1 , w i ,i 2 , and w i ,i 3 are positive because quadcopter i ∈ V F is
inside the triangle defined by the in-neighbor agents.
We can set up a weight matrix W ∈ R( N −3)× N with the i j entry
defined as follows:
⎧
⎪
⎨ w i +3 , j i + 2 ∈ V F , j ∈ N i +3
W i j = −1 j=i+3 . (13)
⎪
⎩
0 else

Let W = [ B A ], where B ∈ R( N −3)×3 is the leader-follower com-

munication matrix and A ∈ R( N −3)×( N −3) is the follower-follower Fig. 2. Inter-agent communication graph used by followers to acquire the desired
communication matrix. The follower-follower communication ma- continuum deformation.
trix A has the following properties [57]:

1. All diagonal entries of the matrix A are −1.

2. Off-diagonal elements of A are non-negative. • Tensegrity arm: Each tensegrity arm consists of a finite num-
3. The matrix A is Hurwitz. ber of tensegrity muscles. A tensegrity muscle is made of
serially-connected prism tensegrity cells [56]. Schematic of a
With communication weights assigned by Eq. (12), quadcopter X prism tensegrity cell is shown in Fig. 3(c). Each prism cell is
and Y position components satisfy the following: made of bars and tethers. A tensegrity muscle can carry both
 T  T tension and compression forces.
q = X, Y , A q4 ··· qN + B q 1 q 2 q 3 = 0. • Gripper: A tensegrity arm is connected to a distinct gripper;
each is commanded by a spiral motor.
Therefore,
• Mechanical joints: To avoid topological obstructions, tenseg-
 T  T rity muscles are attached to a multicopter and a gripper or
q = X, Y , q4 ··· qN = W L q1 q2 q3 , fixture by a three-DOF revolute wrist.
where • Payload: The payload can be either a rigid or deformable body.
⎡ ⎤ It can be a live creature (e.g. such as a human or animal) or a
α4,1 α4,2 α4,3 rigid object that should be transported between two locations.
⎢ .. .. .. ⎥ ,
W L = − A −1 B = ⎣ . . . ⎦ (14) Note the payload object can be a single manipulator with mul-
tiple degrees of freedom to maximize manipulation capability.
αN ,1 αN ,2 αN ,3
αi,1 , αi,2 , and αi,3 are the α parameters satisfying Eq. (10). As shown in Fig. 1, a CALM mission can support two primary
tasks: (i) Transportation Task (TT) and (ii) Force Control (FC). In
3. Problem statement TT, the MUS carries a payload. Therefore, the UAVs act as moving
supports of the CALM system.
Consider a CALM system with the schematic shown in Fig. 3. For FC, the MUS provides stationary supports of the tensegrity
The CALM structure consists of five main elements: arms carrying a single/multiple degree of freedom manipulation
system. During FC, the forces and moments applied by the grippers
• MUS: It is assumed that a MUS consists of N multicopters to a fixture or payload can be either smooth or impulsive. Note
cooperating as moving or stationary supports for the CALM that gentle forces and moments exerted on the grippers are not
structure (See Fig. 3 (a,b)). necessarily continuous but they are also not harsh and impulsive.
 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118 109

⎡ ⎤
1 0 − sin θi
Rω,i = ⎣ 0 cos φi cos θi sin φi ⎦ . (17)
0 − sin φi cos θi cos φi

Note that φ̈i = u φ,i , θ̈i = u θ,i , and ψ̈i = u ψ,i in Eq. (16) are sub-
stituted by u φi , u θi , and u ψi assigned in Section 5. Also, ω,i =
ω1,i − ω2,i + ω3,i − ω4,i , where ω1,i through ω4,i are the angular
speeds of rotors 1, 2, 3, and 4 uniquely related to F T ,i , T φ,i , T θ,i ,
and T ψ,i (See Ref. [59]).

4.2. Tensegrity muscle

The paper assumes tensegrity muscle i ∈ V has negligible mass,

N
so muscle i is a two-force member with stiffness ki ( m ) and struc-
tural damping c i ( N.s m
). Tensegrity muscle i can carry both ten-
sion and compression forces. Each tensegrity muscle connects to
a unique attachment point on the payload. Let payload connecting
points be denoted O 1 , O 2 , and O 3 . Then, ζ : V → { O 1 , O 2 , O 3 } as-
signs a connecting point to each quadcopter i ∈ V . Let ζi = ζ (i )
(i ∈ V → { O 1 , O 2 , O 3 }). Then,
⎡ ⎤
ẋζi − ẋi
ki   ci
Fig. 3. (a) Schematic of a CALM system carrying a payload. (b) Simulation exam- F̄Muscle,i = l i − L f ,i ni + ⎣ ẏ ζi − ẏ i ⎦ , (18)
mi mi
ple of the CALM structure with three grippers O 1 , O 2 , and O 3 . (c) Schematic of a żζi − żi
tensegrity prism cell with bar elements and tethers. A tensegrity prism cell is made  2  2  2
of three bars and nine tensile elements.
where li = xi − xζi + y i − y ζi + zi − zζi , L f ,i is the free
[(xζi −xi ) ( y ζi − y i ) (zζi −zi )]T
4. CALM system mathematical model length of muscle i, and ni = li
. Throughout
the paper, we assume that F̄EXT,i = F̄Muscle,i is the external force
4.1. UAV model exerted on quadcopter i.

Assume a MAS consisting of N quadcopters moves as a group 4.3. Payload dynamics

in a 3-D motion space. The dynamics of quadcopter i ∈ V is given
by The MUS carries a 3-D payload with mass m p . The principal
ṙi = vi body axes of the payload are denoted by î p , ĵ p , and k̂ p . Define
⎡ ⎤ ⎡ ⎤ payload roll angle φ p , payload pitch angle θ p , and payload yaw
0 C φi S θi C ψ i + S φi S ψ i angle ψ p . These quantities are related to ê1 , ê2 , and ê3 by
v̇i = ⎣ 0 ⎦ + F̄ T ,i ⎣ C φi S θi S ψi − S φi C ψi ⎦ + F̄EXT,i , (15)  T  T
−g C φi C θi
îb ĵb k̂b = R p ê1 ê2 ê3 , (19)
 T  T
F̄¨ T ,i φ̈i θ̈i θ̈i = u T ,i u φ,i u θ,i u ψ,i where
⎡ ⎤
where φi , θi , ψi are quadcopter i roll, pitch, and yaw angles, ri = C θp C ψp C θp S ψp − S θp
xi ê1 + y i ê2 + zi ê3 , vi = v x,i ê1 + v y ,i ê2 + v z,i ê3 , g = 9.81 m/s2 is the Rp = ⎣ S φp S θp C ψp − C φp S ψp C φp C ψp + S φp S θp S ψp S φp C θp ⎦ .
F EXT,i
gravity, F̄ EXT,i = is the external force per unit mass, F̄ T ,i = C φp S θp S ψp + S φp S ψp C φp S θp S ψp − S φp S ψp C φp C θp
mi
F T ,i (20)
is the thrust force per unit mass, and mi is the quadcopter
P = ωx, p î p + ω y , p ĵ p +
mi The payload angular velocity is denoted by
mass. Furthermore, u T ,i , u φ,i , u θ,i , and u ψ,i are the control inputs.
Note that C (.) and S (.) are the abbreviations for cos(.) and sin(.),
ωz, p k̂ p , where ωx, p , ω y, p , and ωz, p are related to φ̇ p , θ̇ p , and ψ̇ p
by
respectively.
⎡ ⎤ ⎡ ⎤⎡ ⎤
The rotational dynamics of quadcopter i is given by [59] ωx, p 1 0 − sin θ p φ̇ p
⎡ ⎤ ⎡ ⎤ ⎣ ω y , p ⎦ = ⎣ 0 cos φ p cos θ p sin φ p ⎦ ⎣ θ̇ p ⎦. (21)
T φ,i I xx,i 0 0
ω z, p 0 − sin φ p cos θ p cos φ p ψ̇ p
⎣ T θ,i ⎦ = ⎣ 0 I y y ,i 0 ⎦
T ψ,i 0 0 I zz,i Payload mass moment of inertia values, expressed with respect
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ to the payload body frame, are denoted by I xx, p , I y y , p , and I zz, p .
 φ̈i φ̇i ωx,i 
The quadcopters are connected to the payload at three connecting
× Rω,i ⎣ θ̈i ⎦ + Ṙω,i ⎣ θ̇i ⎦ + I r ,i ⎣ ω y ,i ⎦ ω,i
points O 1 , O 2 , and O 3 . Define payload centroid position by r p =
ψ̈i ψ̇i 0
⎡   ⎤ x p ê1 + y p ê2 + z p ê3 . Then the payload motion dynamics is modeled
I zz,i − I y y ,i θ̇i C  φi + ψ̇i C θi Sφi −θ̇i S φi + ψ̇i C θi C φi by
−⎣ I xx,i − I zz,i  φ̇  i − ψ̇i S θi  − θ̇i S φi + ψ̇i C θi C φi
⎦,
I y y ,i − I xx,i φ̇i − ψ̇i S θi θ̇i C φi + ψ̇i C θi S φi  T  T 
N
m p ẍ p ÿ p z̈ p = mp 0 0 −g − FMuscle,i + FG ,
(16)
i =1
where (22a)
110 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118

⎡ ⎤ ⎡  ⎤
I xx, p ω̇x, p I y y , p − I zz, p  ω y , p ωz, p
⎣ I y y , p ω̇ y , p ⎦ = M p + MG + ⎣ I zz, p − I xx, p ωz, p ωx, p ⎦ . (22b)
 
I zz, p ω̇z, p I xx, p − I y y , p ωx, p ω y , p
N
Note that M p = i =1 mMuscle,i , where mMuscle,i is the moment of
the tensegrity muscle force −FMuscle,i about the center of grav-
ity. Furthermore, FG and MG are the external forces and moments
originating from wind-induced forces and/or manipulation forces
applied by grippers O 1 , O 2 , and O 3 . Fig. 4. Quadcopter i position controller block diagram.

4.4. Motion constraints The quadcopter i control system consists of three operation
modes: (i) a closed-loop position control (PC), (ii) a closed-loop
This paper assumes tensegrity muscles are sufficiently stiff that force control (FC), and (iii) an open-loop rigid body motion (RBM).
length of connecting tensegrity muscle i is almost constant at any Position control tracks a desired trajectory defined by a continuum
time t. Let L 0,i ∈ R+ be the length of muscle i. Then, deformation. In force control mode the MUS acts as stationary sup-
 2  2  2 ports for the tensegrity muscles connected to grippers O 1 , O 2 , and
i ∈ V , ∀t ≥ t 0 , Ci = xi − xζi + y i − y ζ i + zd , i − z ζ i ∼ = L 0, i . O 3 . In RBM mode, the thrust force exerted on quadcopter i re-
Given xi , y i , xζi , y ζi , and zζi , zd,i is obtained as follows: mains constant.

 2  2 5.1. Position control
zd , i = z ζ i + L 0,i − xζi − xi + y ζi − y i . (23)
(iv) Quadcopter i applies input–output (IO) feedback linearization
By taking time derivatives of Ci , żd,i · · · , zd,i are assigned. In this
control to track the desired trajectory rd,i . The controller block di-
paper, the zi component of quadcopter i is updated by the follow-
agram is shown in Fig. 4.
ing fourth-order and second-order dynamics:
The second time derivative of the Eq. (15) second row is:
Position control: ⎡ ⎤
   T u T ,i
 = M T φθ,i ⎣ u φ,i ⎦ + N T φθ,i + F̄¨ EXT,i ,
4
d4 zi d4 zd,i d4− j zd,i d4− j zi (iv)
xi yi
(iv) (iv)
zi (27)
= + γ j ,i − (24a) u ψ,i
dt 4 dt 4 dt 4− j dt 4− j
j =1
Force control: where
   
d2 zi d2 zd,i 
2
d2− j zd,i d2− j zi M T φθ,i (φi , θi , ψi ) = ∈ R3×3 , (28a)
= + γ j ,i − , (24b) 1, i 2, i 3, i
dt 2− j dt 2− j
N T φθ,i = ˙ 0,i + ˙ 1,i F̄˙ T ,i + ˙ 2,i φ̇i + ˙ 3,i θ̇ T ,i ∈ R3×1 ,
dt 2 dt 2
j =1 (28b)
⎡ ⎤
where γ1,i through γ4,i are selected such that dynamics (24) is − C φi S θi S ψ i + S φi C ψ i
stable. Position control and force control will be discussed in Sec- 0,i = ψ̇i
⎣ C φi S θi C ψ i + S φi S ψ i ⎦ , (28c)
tions 5.1 and 5.2, respectively. 0

5. CALM control system 1, i = 1,i (φi , θi , ψi ), 2, i = 2,i (φi , θi , ψi ), and 3, i = 3,i (φi ,
θi , ψi ) .
The dynamics of quadcopter i can be expressed in the following On the other hand,
affine form:  T
(iv) (iv) (iv)
⎧  xi yi zi = Ui . (29)
⎪
⎪ Vi
⎪
⎨Ẋi = Fi (Xi ) + G u + Fd,i 4 d4− j Yi μi d j Yd,i
μ
ψ,i
, (25) Defining Pi = j =1 γ j ,i and Pd,ii = j =0
γ4− j,i , we
⎪ dt 4− j dt j
⎪
⎪ r i = [ x i y i z i ] T
choose
⎩  μ
Yi = [ xi y i ] T
Pd,i − Pi
i

Ui = . (30)
where Xi = [xi y i zi v x,i v y ,i v z,i F̄ T ,i φi θi ψi F̄˙ T ,i φ̇i θ̇i ψi ] T ∈ zi
(iv)

R14×1 is the control state, and ri and Yi denote the actual position
and control output of quadcopter i, respectively. Moreover, Vi = Notice that μi = 3 if i ∈ V F ; otherwise, μi = 4 and γ0,i = 1. Also,
(iv)
[u T ,i u φ,i u θ,i ] T ,G = [04×10 I 4 ] T , Fd,i = [01×6 F̄EXT
T
, i 0 1×5 ]
T zi is updated by Eq. (24a). Furthermore, γ1,i , γ2,i , γ3,i , γ4,i > 0
are chosen such that the MUS collective dynamics is stable. Note
Fi = [ v x,i v y ,i v z,i f 4 f 5 f 6 F̄˙ T ,i φ̇i θ̇i ψ̇i 0 0 0 0] T that MUS collective dynamics stability is discussed in Section 6.
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ By equating the right-hand sides of Eqs. (27) and (29), the control
f 4, i 0 C φi S θi C ψ i + S φi S ψ i
⎣ f 5,i ⎦ = ⎣ 0 ⎦ + F̄ T ,i ⎣ C φi S θi S ψi − S φi C ψi ⎦ + F̄EXT,i . Vi = [u T ,i u φ,i u θ,i ] T is obtained as follows:
! "
f 6, i −g C φi C θi
Vi = M − 1
Ui − N T φθ,i − F̄¨ EXT,i . (31)
T φθ,i
Yaw
 control:
  we assume that u ψ,i = ψ̈d,i +
In this paper,
kψ̇i ψ̇d,i − ψ̈i + kψi ψ̇d,i − ψ̈i , where kψi and kψ̇i are positive 5.2. Force control
constants and ψd,i , ψ̇d,i , and ψ̈d,i are known. Therefore, ψi is up-
dated by the following stable second order dynamics: For force control mode, force F̄EXT,i = F̄Muscle,i and rd,i are the
      reference inputs. It is assumed that quadcopter i’s force sensor can
ψ̈i − ψ̈d,i + kψ̇i ψ̇i − ψ̇d,i + kψi ψi − ψd,i = 0. (26) accurately measure the force F̄EXT,i exerted by tensegrity muscle i.
 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118 111

6. MUS as a moving support

In the transportation task, each quadcopter acts as a moving

Fig. 5. Quadcopter i force controller block diagram.
support and applies position control from Section 5.1 to track de-
sired trajectory. When quadcopters act as moving supports, MUS
desired coordination is decomposed into two modes: (i) MUS con-
tinuum deformation and (ii) Constrained motion along the z axis.
Note that MUS coordination in the X –Y plane is free. The z com-
ponent of quadcopter i coordination is determined by the motion
constraints from Section 4.4.

6.1. Agent coordination

Fig. 6. Quadcopter i rigid body motion (RBM) block diagram. MUS continuum deformation in the X –Y plane can be defined
based on the trajectories chosen by three leaders in Section 6.2.
Furthermore, quadcopter i’s desired position, denoted by rd,i , is Followers acquire a desired continuum deformationthrough lo-
known and constant. Let cal communication. Let local desired position Yd,i = j ∈N i w i , j Y j
(i ∈ V F ) be substituted in Eqs. (29) and (30) and followers commu-
dYi nication weights satisfy Eq. (12).
L i = γ1 , i + γ2 , i Y i (32)
dt Given γ j ,i , ( j = 1, 2, 3, 4, i ∈ V ), we define the following
where γ1,i , γ2,i > 0. Yi asymptotically converges to zero, if Yi is positive-definite diagonal gain matrices:
updated by the following second order dynamics:  
G j , L = diag γ1, j , γ2, j , γ3, j ∈ R3×3 , (39a)
2
d Yi  
= Yd,i − Li . (33) G j , F = diag γ4, j , · · · , γN , j ∈ R(N −3)×( N −3) . (39b)
dt 2
Before obtaining MUS coordination dynamics, we define the fol-
For force control,
lowing vectors: Z TL,q = [q1 q2 q3 ], Z TF ,q = [q4 · · · q N ], Z TL,q, H T =
⎡ ⎤ ⎡ ⎤
u i ,1   0 [q1, H T q2, H T q3, H T ], and Z TF ,q, H T = [q4, H T · · · q N , H T ] (q = x, y).
⎣ ⎦ Yd,i − Li
Ui = u i ,2 = + ⎣ 0 ⎦ − F̄EXT,i , (34) Note that Z F ,q, H T = W L Z L ,q, H T assigns component q of the fol-
z̈i
u i ,3 g lower quadcopter global desired positions given by continuum de-
formation (Eqs. (10) and (14)).
where z̈i is updated by Eq. (24b). Given Ui , desired thrust per mass
F̄ T ,d,i and Euler angles φd,i and θd,i are obtained as: 
4
d4− j Z L ,q, H T
UMUS,q = G j,L (40)
F̄ T ,d,i = Ui , (35a) dt 4− j
j =0
  
φd,i = − sin−1 u 1,i S ψi − u 2,i C ψi / Ui , (35b) as the reference input of the coordination dynamics. If the position
   of every quadcopter is updated according to the dynamics in Eqs.
θd,i = tan−1 u 1,i C ψi + u 2,i S ψi /u 3,i . (35c)
(29) and (30), then MUS collective dynamics becomes
Quadcopter i input Vi = [u T ,i u φ,i u θ,i ] is chosen as:
T
 
d4 Z L ,q d4 Z L ,q, H T 
4
d4− j Z L ,q, H T − Z L ,q
⎡ ⎤ ⎡  ⎤ = + G j,L (41a)
u T ,i −k Ṫ i F̄˙ T ,i + k T i F̄ T ,d,i − F̄T ,i dt 4 dt 4 dt 4− j
⎣ u θ,i ⎦ = ⎢ ⎥ j =1
⎣ −kφ̇i φ̇i + kφi  φd,i − φi ⎦, (36)
u ψ,i −k θ̇i + kθ θd,i − θi d4 Z F ,q 
4
d4− j Z F ,q 
4
d4− j Z L ,q
θ̇i i = G j, F A + G j, F B . (42)
dt 4 dt 4− j dt 4− j
where k T i , k Ṫ i , kφi , kφ̇i , kθi , kθ̇i > 0. The quadcopter i force controller j =1 j =1
is shown in Fig. 5.
Theorem 1. Define E L ,q = Z L ,q − Z L ,q, H T and E F ,q = Z F ,q − Z F ,q, H T as
5.3. Rigid-body motion (RBM) the error signals specifying deviation of leader and follower quadcopters
from global desired positions given by the continuum deformation. The
In rigid body motion mode the thrust force per mass exerted error signals E L ,q and E F ,q are updated by the following fourth order
on quadcopter i, denoted by F̄ TH,i , is constant. Let F̄EXT ,i = F̄Muscle,i
H H dynamics:
is the force per mass exerted by quadcopter i at the static equi-  
librium (hovering) condition. At the hovering condition v̇i vanishes d4 E L ,q 
4
d4− j E L ,q
= G j,L (43a)
(see the second row of Eq. (15)), and the hovering thrust force per dt 4 dt 4− j
j =1
mass F̄ TH,i is obtained as follows:
 
d4 E F ,q 
4
d4− j E F ,q d4− j E L ,q d4 Z L ,q, H T
F̄ TH,i = H
g ê3 + F̄EXT . (37) = G j, F A +B − WL .
,i
dt 4 dt 4− j dt 4− j dt 4
j =1
In the RBM mode, the acceleration of quadcopter i is obtained us-
ing Newton’s second law:
(43b)

r̈i = F̄ TH,i = F̄EXT,i + F̄ TH,i kb,i − g ê3 . (38) d4 Z L ,q, H T

Proof. By subtracting from both sides of Eq. (41a), Eq.
dt 4
The RBM dynamics mode is illustrated in Fig. 6. (41a) can be rewritten as Eq. (43a) and (42) becomes
112 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118

 
d4 Z F ,q 
4
d4− j Z F ,q d4− j Z L ,q, H T d4− j E L ,q
= G j, F A +B +B .
dt 4 dt 4− j dt 4− j dt 4− j
j =1

(44)

Substituting B = − AW L , Eq. (44) can be rewritten as follows:

Fig. 7. HFC hybrid dynamic model.
d4 Z F ,q 
4
d4− j E F ,q 
4
d4− j E L ,q
= G j, F A + G j, F B . (45)
dt 4 dt 4− j dt 4− j 7. An MUS as stationary supports
j =1 j =1

d4 Z d4 Z The MUS is commanded to remain stationary to simulate

By subtracting F ,q, H T
dt 4
= W L dtL,q4, H T from both sides of Eq. (45), a fixed-base manipulation platform when grippers carry multi-
E F ,q is updated by Eq. (43b). 2
degree-of-freedom manipulation arms. To stabilize the MUS, we
 propose a hybrid force control (HFC) model with the following
Remark. In collective dynamics characteristic equation,  s4 I N −3 − three operation modes:
3 

j =0 s G 4− j , F A = 0, roots are all placed in the left-hand side of
j

the s-plane. 1. HM: quadcopter i “Hover Mode” condition is active.

2. RBM: quadcopter i is at the “Rigid Body Motion” mode.
3. FC: quadcopter i is at the “Force Control” Mode.
Theorem 2. Let d s be the distance between the closest quadcopters at
initial time t 0 and db be the closest distance of a quadcopter from the
Let quadcopter i’s desired X Y position, denoted by Yd,i =
sides of the leading triangle forms by three leaders. Suppose λCD,min > 0
[xd, yd,i ] T , be constant and let the threshold distance dThresh , time
assigns a lower-limit for eigenvalues of the pure deformation matrix U CD ,
threshold T Thresh , threshold force magnitudes f Thresh , threshold
e.g λCD,min ≤ λ1 (U CD ) ≤ λ1 (U CD ). Assume each quadcopter is enclosed
force rate ḟ Thresh all be known. We define the following atomic
by a ball with radius . Define
propositions to construct transitions between the HFC discrete
1  nodes:
δmax = (ds − 2 ) , db − . (46)
2 i = 1, · · · , N , d,i := Yi − Yd,i − dThresh ≤ 0,
Inter-agent collision in a MUS continuum deformation is avoided, if the i = 1, · · · , N ,  f ,i := F̄EXT,i − f Thresh ≤ 0,
following inequalities are both satisfied:
i = 1, · · · , N , T ,i := t − t RBM − T Thresh ≤ 0.
∀i ∈ V , Yi − Yi , H T ≤ δ (47a) Note that t RBM and t denote the time system entered the RBM
δ+ and the current time. The HFC hybrid dynamic model is shown in
≤ λCD,min . (47b) Fig. 7. Note the ¬, ∧, and ∨ are logical operators implying “nega-
δmax +
tion”, “conjunction”, and “disjunction”, respectively.
HFC interpretation: Functionality of the HFC is described by
Proof. See the proof in [57]. 2
the following statements:

6.2. Leaders’ evolution during TT • Quadcopter i is at the hovering condition until d,i = Yi −
Yd,i − dThresh ≤ 0 is violated.
The leader quadcopters’ coordinated global desired trajectory is • If Yi − Yd,i − dThresh ≤ 0 is violated but  f ,i = F̄EXT,i −
assigned by solving a constrained optimal control problem. Let x f Thresh ≤ 0 is satisfied, “FC” mode is activated.
and y components of the leaders’ desired positions be updated by • Quadcopter i activates the “RBM”, if both d,i = Yi − Yd,i −
the following second order dynamics: dThresh ≤ 0 and  f ,i = F̄EXT,i − f Thresh ≤ 0 are violated.
• Quadcopter i leaves the “RBM” and activates the “FC” mode,
i ∈ V L , q = x, y , q̈i , H T = u q,i , H T (48) if  f ,i = F̄EXT,i − f Thresh ≤ 0 is violated or  T ,i := t − t 0 −
T Thresh ≤ 0 is not satisfied.
where u x,i , H T and u y ,i , H T are inputs. The area of the leading trian- • Quadcopter i leaves the “FC” modes returns to mode “HM”, if
gle must remain constant at any time t; therefore, desired leaders’ d,i = Yi − Yd,i − dThresh ≤ 0 is satisfied.
trajectories must satisfy the following constraint:
  8. MUS-payload system stability
 x1, H T x2, H T x3, H T 

C L =  y 1, H T y 2, H T y 3, H T  − 2a0 = 0, (49) Tensegrity muscles can carry both tension and compression
 1 1 1  forces, and tensegrity arms are connected to the payload at three
non-aligned points O 1 , O 2 , and O 3 . Therefore, the payload is al-
where a0 is the area of the leading triangle at initial time t 0 . With-
most rigid with respect to the MUS and rotation angles φ p and θ p
are small (φ p ∼
= θp ∼
out loss of generality, we consider
= 0). Let
#
T mission
3 !
 " i = 1, · · · , N , rζi = r p + dζi , (51a)
J= u 2x,i , H T + u 2y ,i , H T dt (50)
i = 1, · · · , N , dζi = dx,ζi ê1 + d y ,ζi ê2 + d z,ζi ê3
i =1
0  T (51b)
= dx,ζi d y ,ζi d z,ζi ,
as the optimal control cost, where T mission is fixed. The solution of
the above optimal control problem is presented in Appendix A and where ζi was previously defined in Section 4.2. Eq. (51) implies
[60]. that
 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118 113

1  
i = 1, · · · , N , xζi = x p + dx,ζi (52a) C p ,xyz = c1 c2 ··· c N ∈ R1× N , (55b)
mp
i = 1, · · · , N , y ζi = y p + d y ,ζi . (52b)
01× N ∈ R1× N , 01×2N ∈ R1×2N , 02× N ∈ R2× N , and 02×2N ∈ R2×2N
Assuming ψ̇ p is small, then, ḋζi ∼ = 0, ḋx,ζi ∼
= 0, ḋ y,ζi = 0, ḋz,ζi = 0, are the zero-entry matrices, and ki and c i denote the stiffness and
d̈ζi ∼
= 0, d̈x,ζi ∼
= 0, d̈ y,ζi = 0, and d̈z,ζi = 0 (i = 1, · · · , N). Therefore, damping of tensegrity muscle i, respectively. The MUS-payload dy-
q = x, y , z, q̇ p =q̇ O 1 = q̇ O 2 = q̇ O 3 namics is given by
⎡ ⎤
q̈ p =q̈ O 1 = q̈ O 2 = q̈ O 3 BPC
SYS,xy 04N ×3 04N × N
⎢ 04N ×3 BPC 04N × N ⎥
ẊPC PC PC ⎢ ⎥ PC PC
SYS = ASYS XSYS + ⎣ ⎦ USYS + fSYS , (56)
SYS,xy
8.1. Position control mode1 04N ×3 04N ×3 BPC
SYS, z
06×3 06×3 06× N
For the position control mode, quadcopter i ∈ V is modeled by
a feedback linearizable dynamics and position of quadcopter i is where
updated by the fourth-order dynamics given in (29) and (30). De- ! "T ! "T ! "T T
 T
fine XPC
SYS = XPC
SYS,x XPC
SYS, y XPC
SYS, z XSYS, p
q = x, y , z,
... ... ∈ R(12N +6)×1 , (57a)
XPC
SYS,q = [q1 · · · q N q̇1 · · · q̇ N q̈1 · · · q̈ N q 1 · · · q N ]T ∈ R4N ×1 , ⎡ PC ⎤
xy 04N 04N 04N ×6
(53a) ⎢ PC
04N ×6 ⎥
  APC ⎢ 04N xy 04N ⎥ ∈ R(12N +6)×(12N +6) ,
SYS = ⎣
XSYS, p = x p yp zp ẋ p ẏ p ż p ∈ R6×1 , (53b) 04N 04N PC
z 04N ×6 ⎦
  PC PC PC PC
j = 1, · · · , 4, G j = diag γ1, j , γ2, j , · · · , γN , j ∈ RN ×N , (53c) p ,x p, y p ,z p, p
  (57b)
−I3 03×( N −3)
= ∈ RN ×N , (53d)  T
B A BPC
SYS,xy = 03×3N I3 03×( N −3) ∈ R4N ×3 , (57c)
⎡ ⎤  T
0N IN 0N 0N 4N × N
BPC = 0 N ×3N IN ∈R , (57d)
⎢ 0N 0N IN 0N ⎥ SYS, z
xy = ⎢
PC
⎣ 0N
⎥ ∈ R4N ×4N , (53e)  T  N T
0N 0N IN ⎦
fPC
SYS = 01×(12N +5) −g + 01×(12N +3) − m1p T
i = 1 k ζ i dζ .
G 4 G 3 G 2 G 1 i
⎡ ⎤ (57e)
0N IN 0N 0N
⎢ 0N ⎥
PC ⎢ 0N 0N IN ⎥ ∈ R4N ×4N , Defining
z =⎣ 0 0N 0N IN ⎦
(53f)
N  T
−G 4 −G 3 −G 2 −G 1 d4 zd,1 d4 zd, N
 UPC
MUS, z = ···
N dt 4 dt 4
− m1p i =1 k i I3 03
 p, p = N ∈R 6×6
, (53g) 
4  T
03 − m1p i =1 c i 03 d4− j zd,1 d4− j zd, N
+ Gj ··· ,
j =1
dt 4− j dt 4− j
where A ∈ R( N −3)×( N −3) and B ∈ R(N −3)×3 were previously de-
fined in (13), I ∈ R N × N and I ∈ R3×3 are the identity matrices ( N +6)×1 is given by
SYS ∈ R
UPC
N 3
and 03×( N −3) ∈ R3×( N −3) and 0 N are zero-entry matrices. Further-
! "T ! "T ! "T T
more, we define
⎡ ⎤ UPC
SYS = UPC
MUS,x UPC
MUS, y UPC
MUS, z
, (58)
K p ,xyz 01× N 01×2N
⎢ 02× N 02× N 01×2N ⎥
 p ,x = ⎢ ⎥ ∈ R6×4N , MUS,x = UMUS,x and UMUS, y = UMUS, y were previously de-
where UPC
PC PC
⎣ 01× N (54a)
C p ,xyz 01×2N ⎦ fined in Eq. (40).
02× N 02× N 01×2N
⎡ ⎤
01× N 01× N 01×2N Remark. Matrix  p , p and z are Hurwitz. A is also Hurwitz, so
⎢ K p ,xyz 01× N 01×2N ⎥  and xy are Hurwitz as well. Matrix ASYS is Hurwitz because
⎢ ⎥
PC
p , y = ⎢ 02× N 02× N 02×2N ⎥ ∈ R6×4N , (54b) all block-diagonal matrices are Hurwitz. Consequently, the MUS-
⎣0 C p ,xyz 01×2N
⎦ payload dynamics, given by Eq. (56), is stable.
1× N
01× N 01× N 01×2N
⎡ ⎤ 8.2. Force control mode2
02× N 02× N 01×2N
⎢ K 01× N 01×2N ⎥
PC ⎢ p ,xyz ⎥ ∈ R6×4N ,
p ,z = ⎣ 0 02× N 01×2N ⎦
(54c) To analyze MUS-Payload stability in a force control task, UAVs
2× N
are modeled as double integrators. Define
01× N C p ,xyz 01×2N
2N ×1
where q = x, y , z, XFC T
SYS,q = [q 1 · · · q N q̇ 1 · · · q̇ N ] ∈ R , (59a)
 
1   0N IN
K p ,xyz = k1 k2 ··· k N ∈ R1× N , (55a) FC
xy = ∈ R2N ×2N , (59b)
mp G 2 G 1

1 2
The superscript PC refers to “Position Control”. The superscript FC is referred to “Force Control”.
114 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118

 
0N IN
FC
z = ∈ R2N ×2N , (59c) 9. Simulation results
−G 2 −G 1
⎡ ⎤ Consider a CALM consisting of 18 quadcopters. Each quadcopter
K p ,xyz 01× N connects to one of the grippers by a tensegrity muscle. We assume
⎢ 02× N 02× N ⎥ N
 p ,x = ⎢
FC ⎥ ∈ R6×2N , (59d) that tensegrity muscles all have the same stiffness ki = 50 m and
⎣ 01× N C p ,xyz ⎦ c i = 5 Nm·s . The CALM system consists of three arms. Arm 1 con-
02× N 02× N nects quadcopters 1, 4, 7, 8, 13, 14 to gripper O 1 ; Arm 2 connects
⎡ ⎤ quadcopters 2, 5, 9, 10, 15, 16 to gripper O 2 ; Arm 3 connects
01× N 01× N
quadcopters 3, 6, 11, 12, 17, 18 to gripper O 3 .
⎢ K p ,xyz 01× N ⎥
⎢ ⎥ We simulate dynamics of CALM in the “TT”, where CALM acts as
FC
p , y = ⎢ 02× N 02× N ⎥ ∈ R6×2N , (59e) moving supports. We also simulate CALM dynamic behavior, where
⎣0 C p ,xyz
⎦
1× N it acts as stationary supports and controls the required reaction
01× N 01× N force.
⎡ ⎤
02× N 02× N 9.1. MUS as moving support
⎢ K p ,xyz 01× N ⎥
FC ⎢
 p ,z = ⎣ ⎥ ∈ R6×2N . (59f)
02× N 02× N ⎦ When the MUS cooperatively carries a payload, agent coordi-
01× N C p ,xyz nation is defined by a continuum deformation. Continuum defor-
mation coordination is defined by the trajectories of the leaders
Note that  ∈ R N × N , K p ,xyz ∈ R1× N , C p ,xyz ∈ R1× N , G 1 ∈ R N × N ,
1, 2, and 3 and acquired by the follower quadcopters through
and G 2 ∈ R N × N were previously defined in Section 8.1. The MUS-
local communication. The graph shown in Fig. 2 defines inter-
payload dynamics is given by
agent communication among the quadcopters. Given initial posi-
⎡ ⎤ tions shown in Fig. 2, followers’ communication weights, calculated
BPC
SYS,xy 02N ×3 02N × N
using Eq. (12), are as follows:
⎢ 02N ×3 BPC 02N × N ⎥
ẊFC FC FC ⎢ ⎥ FC FC
SYS = ASYS XSYS + ⎣ ⎦ USYS + fSYS
SYS,xy (60)
02N ×3 02N ×3 BPC w 4,1 = w 5,2 = w 6,3 = w 7,4 = w 8,4 = w 9,5 = w 10,5 = w 11,6
SYS, z
06×3 06×3 06× N = w 13,7 = w 14,8 = w 15,9 = w 16,10 = w 17,11
where = w 18,12 = 0.55.
! "T ! "T ! "T T The remaining communication weights are all 0.225. x and y com-
 T ponents of the leaders’ desired trajectories are assigned by solving
XFC
SYS = XFC
SYS,x XFC
SYS, y XFC
SYS, z XSYS, p
the optimal control problem defined in Section 6.2 and formu-
∈ R(6N +6)×1 , (61a) lated in [60]. Leaders initiate their motion from rest at ( X 1 , Y 1 ) =
(5.5, 0), ( X 2 , Y 2 ) = (−2.5, 4), and ( X 3 , Y 3 ) = (−2.5, −4). It is
⎡ ⎤
⎡ ⎤ 06×1 aimed that leaders reach their target destinations at
2 d2− j Z L ,x, H T ⎢ d2 zd,1 ⎥  
⎢ j =0 G j , L dt 2− j ⎥ ⎢ ⎥ x1, H T ( T mission ) , y 1, H T ( T mission ) = (28, 30) ,
⎢ ⎥ ⎢ dt 2 ⎥
UFC = ⎢ 2 d2− j Z L , y , H T ⎥+⎢
⎢ . ⎥
⎥
 
SYS
⎣ j =0 G j , L ⎦ ⎢ .. ⎥ x2, H T ( T mission ) , t 2, H T ( T mission ) = (20, 38) ,
2− j  
dt ⎣ 2 ⎦
0 N ×1 d zd , N x3, H T ( T mission ) , y 3, H T ( T mission ) = (20, 30)
dt 2 in T mission = 150 s. Note that initial area of the leading triangle is
⎡ ⎤T
06×1 A 0 = 32 m2 . The leaders’ desired trajectories must satisfy equality
⎢ d2− j zd,1 ⎥ constraint (49). x and y components of the leaders’ optimal tra-
2  ⎢⎢ ⎥
 ⎥ jectories are plotted versus time in Fig. 8 (a). Followers apply the
06 06× N ⎢ dt 2− j ⎥
+ ⎢ .. ⎥ , (61b) communication graph shown in Fig. 2 to acquire the desired coor-
0 N ×6 Gj ⎢ . ⎥
j =1 ⎢ ⎥ dination defined by leaders. x, y, and z components of follower 14
⎣ d2− j zd, N ⎦ position are shown in Fig. 8 (b). Tensegrity muscle force per mass
dt 2− j F̄EXT,14 = F̄Muscle,14 and the magnitude of the thrust force of
⎡ ⎤ quadcopter 14 are plotted versus time in Fig. 8 (c).
FC
xy 02N 02N 02N ×6 Collision avoidance verification: We observe that quadcopters
⎢ 02N FC 02N 02N ×6 ⎥ 12 and 18 has minimum separation d s = 0.7104 m at t = 0.
AFC ⎢ ⎥ ∈ R(6N +6)×(6N +6) ,
SYS = ⎣
xy
(61c)
02N 02N FC
z 02N ×6 ⎦ Quadcopter 4 is closest to the boundary (3 − 1) of leading trian-
gle (db = 0.5792 m) at the initial time. We assume = 15 cm
FC FC
p ,x  p , y FC
p ,z FC
p, p
is the radius of a ball enclosing each quadcopter. Given d s and
 T db , δmax is computed from Eq. (46): δmax = min 0.5 (d s − 2 ) ,
BFC
SYS,xy = 03× N I3 03×( N −3) ∈ R2N ×3 , (61d) 
db − = 0.2052. Given leaders trajectories’ shown in Fig. 8 (a)
 T
BFC
SYS, z = 0 N IN ∈ R2N × N , (61e) and (b), eigenvalues of matrix U D are plotted versus time in
Fig. 8 (d). We choose λCD,min = 0.78 as the lower limit for the
 T  N T matrix U D eigenvalues, e.g. 0.78 < λ1 (t ) ≤ λ2 (2), ∀t ≥ 0. There-
fFC
SYS = 01×(6N +5) −g + 01×(6N +3) − m1p i =1 kζi dζTi . fore, δ = λCD,min (δmax + ) − = 0.1271 m. Inter-agent collision is
avoided if Yi − Yi , H T ≤ 0.1661 m. In Fig. 8 (e), Yi − Yi , H T (∀i ∈ V )
(61f)
is plotted versus time. It is seen that followers’ continuum defor-
mation satisfy inequality (47a); therefore, inter-agent collision is
Remark. Matrix AFC
SYS is Hurwitz. Therefore, MUS-payload dynamics avoided. Note that the quadcopters are initially in a static equilib-
(60) is stable. rium condition so that Yi − Yi , H T is small (∀i ∈ V ).
 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118 115

Fig. 9. (a) Tensegrity muscle force magnitude F EXT,14 vs. time. (b) Position com-
ponents of quadcopter 14 vs. time.

Fig. 10. (a) Tensegrity muscle force magnitude F EXT,14 vs. time. (b) Position com-
ponents of quadcopter 14 vs. time.

Quadcopter initial positions are shown in Fig. 2. We consider

dThresh = 0.01 m, f Thresh = 25 N, T Thresh = 3 s as the threshold dis-
tance, threshold force, and threshold time, respectively. We sim-
ulate CALM dynamic behavior for gentle and impulsive external
force FG and moment MG .

9.2.1. Gentle (smooth) external excitation

Fig. 8. (a) Leaders’ desired trajectories. (b) Follower 14 position components.
(c) Quadcopter 14 thrust per mass F̄ T ,14 and tensegrity force per mass FEXT,14
vs. time. (d) Eigenvalues of matrix U CD vs. time. (e) Yi − Yi , H T vs. time (∀i ∈ V ).
(For interpretation of the colors in the figure(s), the reader is referred to the web
version of this article.)
9.2. The MUS as a stationary support system
In this section, we assume that grippers O 1 , O 2 , and O 3 grasp
three connection points of the payload. The payload or payload
connection fixture is a triangle with mass m p = 100 kg and geom-
etry shown in Fig. 3 (b). The payload can be considered as the cap
of the manipulation arms grasped at points O 1 , O 2 , O 3 . We as-
sume that force FG and moment MG are exerted on the centroid
of the triangular cap.
Assume that constant force FG = [20 20 40] T N and moment
MG = 0.1FG N.m are exerted at the centroid of the payload. In
Fig. 9 (a) tensegrity muscle force FEXT,14 is plotted versus time.
Because FEXT,14 < f Thresh = 25 N, the force control mode is ac-
tivated. Quadcopter force controllers stabilize quadcopters at the
initial destinations shown in Fig. 2. Position components of quad-
copter 14 are plotted in Fig. 9 (b).
9.2.2. Impulsive external excitation
Let

[10 50 200]T ( N ) t ≤ 0.1s
FG =
0 otherwise
and MG = 0.1FG (N.m) be exerted at the centroid of the trian-
gular payload grasped by grippers O 1 , O 2 , and O 3 . As shown
116 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118

in Fig. 10 (a) tensegrity muscle force magnitude F EXT,14 sud- where σ = [σ1 · · · σ12 ] ∈ R12×1 is the costate vector and  is the
denly increases, causing Quadcopter 14 to activate the RBM over Lagrange multiplier. Using calculus of variation, X and Y compo-
t ∈ [0, T Thresh ]. At t = T Thresh = 3s, the quadcopter leaves the RBM nents of the leaders’ optimal trajectories are assigned by solving
because  T ,i := t − t RBM − T Thresh ≤ 0 is violated, so it activates FC the following dynamics system:
mode until safely returning to hover. Quadcopter 14 position time
history is shown in Fig. 10 (b). Ṡ = A s S , (65)

10. Conclusion where

 
P
S= , (66a)
The paper has applied continuum deformation to an application σ
in which multiple quadcopters carry a single parcel cooperatively. ⎡ ⎤
0 I6 0 0
The cooperative payload transport mission can be accomplished in
⎢ − K 1
− I6 ⎥
1
a decentralized fashion, where continuum deformation is defined ⎢ 1 0 0 ⎥
by leaders and acquired by follower UAVs through local communi- As = ⎢ 2
⎢1 2 1
2 ⎥,
⎥ (66b)
cation with less computation cost. Because continuum deformation ⎣  K2 0 0 K1 ⎦
2 2
of an MUS is scalable, a large MUS team can cooperatively carry a 0 −2 K 1 − I 6 0
heavy payload without the need for a heavy-lift quadcopter design. ⎡ ⎤
Continuum deformation coordination also provides the ability for
0 0 0 0 1 −1
⎢ 0 0 0 −1 0 1 ⎥
the team to pass through a narrow channel without collision as ⎢ ⎥
demonstrated in a eighteen-quadcopter case study. The proposed ⎢ 0 0 0 1 −1 0 ⎥
K1 = ⎢ ⎥, (66c)
CALM approach can be also used in a cooperative object manipu- ⎢ 0 −1 1 0 0 0 ⎥
⎣ 1 0 −1 0 0 0
⎦
lation task, where quadcopters act as stationary supports. For this
purpose, we designed a hybrid force controller to stabilize the UAV −1 1 0 0 0 0
team when external forces exerted on each quadcopter are either ⎡ ⎤
2 −1 −1 0 0 0
impulsive or smooth.
⎢ −1 2 −1 0 0 0 ⎥
⎢ ⎥
⎢ −1 −1 2 0 0 0 ⎥
Conflict of interest statement K2 = ⎢ ⎥. (66d)
⎢ 0 0 0 2 −1 −1 ⎥
⎣ 0 0 0 −1 2 −1 ⎦
None declared.
0 0 0 −1 −1 2
Appendix A Note that  in Eq. (65) can be expressed as
2τ − ξ
In this Appendix, we find the leaders’ optimal trajectories min- = , (67)
imizing cost function (50) and satisfying equality constraints (49). ρ
Dynamics (48) can be given in the following state-space form [57]: where

Ṗ = A p P + B p U p (62) ρ = ( p 5 − p 6 )2 + ( p 6 − p 4 )2 + ( p 4 − p 5 )2 + ( p 3 − p 2 )2
, (68a)
+ ( p 1 − p 3 )2 + ( p 2 − p 1 )2
where P = [ p i ] ∈ R12×1 , p 1 = x1, H T , p 2 = x2, H T , p 3 = x3, H T , p 4 =
y 1, H T , p 5 = y 2, H T , p 6 = y 3, H T , p 7 = ẋ1, H T , p 8 = ẋ2, H T , p 9 = ẋ3, H T , τ = p 7 ( p 11 − p 12 ) + p 8 ( p 12 − p 10 ) + p 9 ( p 10 − p 11 )
, (68b)
p 10 = ẏ 1, H T , p 11 = ẏ 2, H T , p 12 = ẏ 3, H T , + p 10 ( p 9 − p 8 ) + p 11 ( p 7 − p 9 ) + p 12 ( p 8 − p 7 )
 T
U p = u 1, H T u 2, H T u 3, H T v 1, H T v 2, H T v 3, H T ξ = σ7 ( p 5 − p 6 ) + σ8 ( p 6 − p 4 ) + σ9 ( p 4 − p 5 ) + σ10 ( p 3 − p 2 )
 T  T .
0 I6 0 + σ11 ( p 1 − p 3 ) + σ12 ( p 2 − p 1 )
Ap = , Bp = .
0 0 I6 (68c)

Note that 0, I6 ∈ R6×6 are zero-entry and identity matrices, re- Moreover, optimal control inputs are obtained as follows:
spectively. Constraint (49) can be rewritten as [60] 1
u 1, H T = − (σ7 + ( p 5 − p 6 )), (69a)
 2
C L :=u 1, H T ( p 5 − p 6 ) + u 2, H T ( p 6 − p 4 ) + u 3, H T ( p 4 − p 5 )
1
+ v 1, H T ( p 3 − p 2 ) + v 2, H T ( p 1 − p 3 ) + v 3, H T ( p 2 − p 1 ) u 2, H T = − (σ8 + ( p 6 − p 4 ), (69b)
(63) 2
+ p 7 ( p 11 − p 12 ) + p 8 ( p 12 − p 10 ) + p 9 ( p 10 − p 11 ) 1
u 3, H T = − (σ9 + ( p 4 − p 5 )), (69c)
+ p 10 ( p 9 − p 8 ) + p 11 ( p 7 − p 9 ) + p 12 ( p 8 − p 7 ) = 0. 2
1
Notice that constraint (63) is obtained by taking the second v 1, H T = − (σ10 + ( p 3 − p 2 )), (69d)
2
derivative of Eq. (49) with respect to time t. Leaders are initially 1
at rest, and they stop within a finite horizon time T mission . We as- v 2, H T = − (σ11 + ( p 1 − p 3 )), (69e)
2
sume that leaders’ initial and final positions are given.
1
Let the cost functional (50) be rewritten in the following aug- v 3, H T = − (σ12 + ( p 5 − p 6 )). (69f)
2
mented form [60]:
Numerical solution: Leaders’ optimal trajectories are determined
#
T mission
by using a distributed gradient algorithm. Let k (t , t 0 ), P k , and σk
 
J = U p T U p + σ T ( A p P + B p U p − Ṗ ) + C L dt (64) denote state transition matrix, control state, and costate at attempt
(k = 1, 2, 3, . . . ). Then [60],
0
 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118
∀t ∈ [0, T mission ],
   
P k (t ) P (t )
= k (t , t 0 ) k 0
σk (t ) σk (t 0 )
   (70)
11k (t , t 0 ) 12k (t , t 0 ) P (t 0 )
= .
21k (t , t 0 ) 22k (t , t 0 ) σ (t 0 )
Thus,
P ( T mission ) = P k ( T mission ) = 11k ( T mission , t 0 ) P (t 0 )+ 12k σk (t 0 ).
(71)
Notice that P (t 0 ) and P ( T mission ) are both given at initial time
t 0 and final time T mission . Hence, σk (t 0 ) can be expressed as fol-
lows [60]:
σk (t 0 ) = 12k ( T mission , t 0 )−1 ( P ( T mission )
− 11k ( T mission , t 0 ) P (t 0 )). (72)
Therefore, P k (t ) and σk (t ) can be updated using Eq. (70) until
∀t ∈ [t 0 , T mission ], |k (t ) − k−1 (t )| → 0. (73)
References
[1] J.A. Fax, R.M. Murray, Information flow and cooperative control of vehicle for-
mations, IEEE Trans. Autom. Control 49 (9) (2004) 1465–1476.
[2] F. Paterno, C. Santoro, S. Tahmassebi, Formal models for cooperative tasks: con-
cepts and an application for en-route air traffic control, in: Design, Specification
and Verification of Interactive Systems’ 98, Springer, 1998, pp. 71–86.
[3] N. Michael, J. Fink, V. Kumar, Cooperative manipulation and transportation with
aerial robots, Auton. Robots 30 (1) (2011) 73–86.
[4] D.G. Schmale Iii, B.R. Dingus, C. Reinholtz, Development and application of an
autonomous unmanned aerial vehicle for precise aerobiological sampling above
agricultural fields, J. Field Robot. 25 (3) (2008) 133–147.
[5] N. Michael, S. Kim, J. Fink, V. Kumar, Kinematics and statics of coopera-
tive multi-robot aerial manipulation with cables, in: ASME 2009 International
Design Engineering Technical Conferences and Computers and Information
in Engineering Conference, American Society of Mechanical Engineers, 2009,
pp. 83–91.
[6] A. Bazoula, M. Djouadi, H. Maaref, Formation control of multi-robots via fuzzy
logic technique, Int. J. Comp. Commun. Control 3 (3) (2008) 179–184.
[7] R.M. Murray, Recent research in cooperative control of multivehicle systems, J.
Dyn. Syst. Meas. Control 129 (5) (2007) 571–583.
[8] C.-L. Liu, F. Liu, Asynchronously compensated consensus algorithm for discrete-
time second-order multi-agent systems under communication delay, IET Con-
trol Theory Appl. 8 (17) (2014) 2004–2012.
[9] J. Mei, W. Ren, J. Chen, Distributed consensus of second-order multi-agent sys-
tems with heterogeneous unknown inertias and control gains under a directed
graph, IEEE Trans. Autom. Control 61 (8) (2016) 2019–2034.
[10] L. Zhang, C. Hua, X. Guan, Distributed output feedback consensus tracking pre-
scribed performance control for a class of non-linear multi-agent systems with
unknown disturbances, IET Control Theory Appl. 10 (8) (2016) 877–883.
[11] A. Bidram, A. Davoudi, F.L. Lewis, Z. Qu, Secondary control of microgrids based
on distributed cooperative control of multi-agent systems, IET Gener. Transm.
Distrib. 7 (8) (2013) 822–831.
[12] W. Ren, R.W. Beard, E.M. Atkins, Information consensus in multivehicle cooper-
ative control, IEEE Control Syst. 27 (2) (2007) 71–82.
[13] J. Kim, K.-D. Kim, V. Natarajan, S.D. Kelly, J. Bentsman, Pde-based model ref-
erence adaptive control of uncertain heterogeneous multiagent networks, Non-
linear Anal. Hybrid Syst. 2 (4) (2008) 1152–1167.
[14] P. Frihauf, M. Krstic, Leader-enabled deployment onto planar curves: a pde-
based approach, IEEE Trans. Autom. Control 56 (8) (2011) 1791–1806.
[15] K.T. Magar, M.J. Balas, D.F. Gayme, Adaptive control of inter-area oscillations in
wind-integrated power systems using distributed parameter control methods,
in: American Control Conference, ACC, 2014, IEEE, 2014, pp. 903–907.
[16] Y. Cao, D. Stuart, W. Ren, Z. Meng, Distributed containment control for multiple
autonomous vehicles with double-integrator dynamics: algorithms and experi-
ments, IEEE Trans. Control Syst. Technol. 19 (4) (2011) 929–938.
[17] S.J. Yoo, Distributed containment control with predefined performance of high-
order multi-agent systems with unknown heterogeneous non-linearities, IET
Control Theory Appl. 9 (10) (2015) 1571–1578.
[18] Z. Lin, W. Ding, G. Yan, C. Yu, A. Giua, Leader–follower formation via complex
laplacian, Automatica 49 (6) (2013) 1900–1906.
117
[19] Y. Cao, W. Ren, Containment control with multiple stationary or dynamic lead-
ers under a directed interaction graph, in: Proceedings of the 48th IEEE Con-
ference on Decision and Control, 2009 Held Jointly with the 2009 28th Chinese
Control Conference, CDC/CCC 2009, IEEE, 2009, pp. 3014–3019.
[20] G. Ferrari-Trecate, M. Egerstedt, A. Buffa, M. Ji, Laplacian sheep: a hybrid, stop-
go policy for leader-based containment control, in: Hybrid Systems: Computa-
tion and Control, Springer, 2006, pp. 212–226.
[21] Q. Lu, Q.-L. Han, X. Xie, S. Liu, A finite-time motion control strategy for odor
source localization, IEEE Trans. Ind. Electron. 61 (10) (2014) 5419–5430.
[22] H. Li, G. Chen, Z. Dong, D. Xia, Consensus analysis of multiagent systems with
second-order nonlinear dynamics and general directed topology: an event-
triggered scheme, Inf. Sci. 370 (2016) 598–622.
[23] S. Martinez, J. Cortes, F. Bullo, Motion coordination with distributed informa-
tion, IEEE Control Syst. 27 (4) (2007) 75–88.
[24] W. Yu, P. Wang, Distributed node-to-node consensus of linear multi-agent sys-
tems with directed switching topologies, in: 2016 14th International Confer-
ence on Control, Automation, Robotics and Vision, ICARCV, IEEE, 2016, pp. 1–6.
[25] C. Li, X. Yu, W. Yu, T. Huang, Z.-w. Liu, Distributed event-triggered scheme
for economic dispatch in smart grids, IEEE Trans. Ind. Inform. 12 (5) (2016)
1775–1785.
[26] L.B. Seeff, J.H. Hoofnagle, National institutes of health consensus development
conference: management of hepatitis c, Hepatology 36 (5B) (2002).
[27] W. Zhao, M. Liu, J. Zhu, L. Li, Fully decentralised multi-area dynamic economic
dispatch for large-scale power systems via cutting plane consensus, IET Gener.
Transm. Distrib. 10 (10) (2016) 2486–2495.
[28] H. Xing, Y. Mou, M. Fu, Z. Lin, Distributed bisection method for economic power
dispatch in smart grid, IEEE Trans. Power Syst. 30 (6) (2015) 3024–3035.
[29] L. Cheng, Y. Wang, W. Ren, Z.-G. Hou, M. Tan, Containment control of multia-
gent systems with dynamic leaders based on a pî {n}-type approach, IEEE Trans.
Cybern. 46 (12) (2016) 3004–3017.
[30] M. Ji, G. Ferrari-Trecate, M. Egerstedt, A. Buffa, Containment control in mobile
networks, IEEE Trans. Autom. Control 53 (8) (2008) 1972–1975.
[31] W. Zhang, Y. Tang, Y. Liu, J. Kurths, Event-triggering containment control for a
class of multi-agent networks with fixed and switching topologies, IEEE Trans.
Circuits Syst. I, Regul. Pap. 64 (3) (2017) 619–629.
[32] W. Li, L. Xie, J.-F. Zhang, Containment control of leader-following multi-
agent systems with markovian switching network topologies and measurement
noises, Automatica 51 (2015) 263–267.
[33] M. Oussalah, D. Professor Ali Hessami, B. Qi, X. Lou, B. Cui, Containment control
of second-order multi-agent systems with directed topology and time-delays,
Kybernetes 43 (8) (2014) 1248–1261.
[34] C. Xu, Y. Zheng, H. Su, H.O. Wang, Containment control for coupled harmonic
oscillators with multiple leaders under directed topology, Int. J. Control 88 (2)
(2015) 248–255.
[35] K. Liu, Z. Ji, G. Xie, R. Xu, Event-based broadcasting containment control for
multi-agent systems under directed topology, Int. J. Control 89 (11) (2016)
2360–2370.
[36] Y. Zhao, Z. Duan, Finite-time containment control without velocity and accel-
eration measurements, Nonlinear Dyn. 82 (1–2) (2015) 259–268.
[37] H. Liu, L. Cheng, M. Tan, Z. Hou, Y. Wang, Distributed exponential finite-time
coordination of multi-agent systems: containment control and consensus, Int.
J. Control 88 (2) (2015) 237–247.
[38] M. Bernard, K. Kondak, I. Maza, A. Ollero, Autonomous transportation and de-
ployment with aerial robots for search and rescue missions, J. Field Robot.
28 (6) (2011) 914–931.
[39] M. Bernard, K. Kondak, G. Hommel, A slung load transportation system based
on small size helicopters, in: Autonomous Systems–Self-Organization, Manage-
ment, and Control, Springer, 2008, pp. 49–61.
[40] I. Palunko, R. Fierro, P. Cruz, Trajectory generation for swing-free maneuvers of
a quadrotor with suspended payload: a dynamic programming approach, in:
2012 IEEE International Conference on Robotics and Automation, ICRA, IEEE,
2012, pp. 2691–2697.
[41] S. Kim, S. Choi, H.J. Kim, Aerial manipulation using a quadrotor with a two dof
robotic arm, in: 2013 IEEE/RSJ International Conference on Intelligent Robots
and Systems, IEEE, 2013, pp. 4990–4995.
[42] H.-N. Nguyen, S. Park, J. Park, D. Lee, A novel robotic platform for aerial manip-
ulation using quadrotors as rotating thrust generators, IEEE Trans. Robot. 34 (2)
(2018) 353–369.
[43] P.E. Pounds, D.R. Bersak, A.M. Dollar, Stability of small-scale uav helicopters and
quadrotors with added payload mass under pid control, Auton. Robots 33 (1–2)
(2012) 129–142.
[44] M. Bernard, K. Kondak, Generic slung load transportation system using small
size helicopters, in: IEEE Intl. Conf. on Robotics and Automation, 2009, ICRA’09,
IEEE, 2009, pp. 3258–3264.
[45] F.A. Goodarzi, D. Lee, T. Lee, Geometric adaptive tracking control of a quadro-
tor unmanned aerial vehicle on se (3) for agile maneuvers, J. Dyn. Syst. Meas.
Control 137 (9) (2015) 091007.
[46] K. Sreenath, T. Lee, V. Kumar, Geometric control and differential flatness of a
quadrotor uav with a cable-suspended load, in: Decision and Control Confer-
ence, IEEE, 2013, pp. 2269–2274.
118 H. Rastgoftar, E.M. Atkins / Aerospace Science and Technology 82–83 (2018) 105–118
[47] D. Mellinger, Q. Lindsey, M. Shomin, V. Kumar, Design, modeling, estima-
tion and control for aerial grasping and manipulation, in: 2011 IEEE/RSJ
International Conference on Intelligent Robots and Systems, IEEE, 2011,
pp. 2668–2673.
[48] D. Mellinger, M. Shomin, N. Michael, V. Kumar, Cooperative grasping and trans-
port using multiple quadrotors, in: Distributed Autonomous Robotic Systems,
Springer, 2013, pp. 545–558.
[49] M. Bisgaard, A. la Cour-Harbo, J.D. Bendtsen, Adaptive control system for au-
tonomous helicopter slung load operations, Control Eng. Pract. 18 (7) (2010)
800–811.
[50] J. Potter, W. Singhose, M. Costelloy, Reducing swing of model helicopter sling
load using input shaping, in: International Conference on Control and Automa-
tion, ICCA, IEEE, 2011, pp. 348–353.
[51] S. Dai, T. Lee, D.S. Bernstein, Adaptive control of a quadrotor uav transport-
ing a cable-suspended load with unknown mass, in: 53rd IEEE Conference on
Decision and Control, IEEE, 2014, pp. 6149–6154.
[52] P.E. Pounds, D.R. Bersak, A.M. Dollar, Grasping from the air: hovering capture
and load stability, in: 2011 IEEE Intl. Conf. on Robotics and Automation, ICRA,
IEEE, 2011, pp. 2491–2498.
[53] Q. Jiang, V. Kumar, The inverse kinematics of cooperative transport with mul-
tiple aerial robots, IEEE Trans. Robot. 29 (1) (2013) 136–145.
[54] T. Lee, K. Sreenath, V. Kumar, Geometric control of cooperating multiple
quadrotor uavs with a suspended payload, in: Decision and Control Conference,
IEEE, 2013, pp. 5510–5515.
[55] G. Wu, K. Sreenath, Geometric control of multiple quadrotors transport-
ing a rigid-body load, in: Decision and Control Conference, IEEE, 2014,
pp. 6141–6148.
[56] R.E. Skelton, M.C. de Oliveira, Tensegrity Systems, vol. 1, Springer, 2009.
[57] H. Rastgoftar, Continuum Deformation of Multi-Agent Systems, Springer, 2016.
[58] W.M. Lai, D.H. Rubin, D. Rubin, E. Krempl, Introduction to Continuum Mechan-
ics, Butterworth-Heinemann, 2009.
[59] T. Luukkonen, Modelling and control of quadcopter, Independent research
project in applied mathematics, Espoo 22 (2011).
[60] H. Rastgoftar, S. Jayasuriya, Evolution of multi agent systems under a new
communication topology, in: 2014 Dynamic Systems and Control Conference,
American Society of Mechanical Engineers, 2014.