2018 Annual American Control Conference (ACC)

June 27–29, 2018. Wisconsin Center, Milwaukee, USA

Dynamics, Hover Configurations, and Rotor Failure Restabilization of a

Morphing Quadrotor
Trevor Avant, Unsik Lee, Brian Katona and Kristi Morgansen

Abstract— We present the design of a morphing quadrotor In this paper, we will consider a quadrotor which is
which is capable of arm rotation and extension. Using La- capable of arm rotation and extension. This design has been
grangian mechanics, we derive the multibody dynamics. We considered in [8, 9]. Due to the complexity of the system, the
also derive the set of equations which describe all possible hover
configurations of the vehicle. Using these equations, we develop full dynamics of the design were never derived, as [9] only
a technique to find a hover configuration in the event a rotor considered symmetric motions of the vehicle (e.g. opposite
has limited or zero thrust. We then develop a control technique arms move mirrored to each other). In this paper, we derive
to stabilize the vehicle at these configurations after a rotor has the full dynamics of the vehicle, which is only tractable
failed. Finally, we present a simulation which demonstrates the through the use of symbolic math software.
viability of recovering from complete and sudden rotor failure.
There are several enhanced capabilities of a quadrotor with
morphing arms. The capability we will explore in this paper
I. I NTRODUCTION is related to the fact that there are a variety of possible
shape configurations with which the vehicle can hover. As
Over the past several years, quadrotors have been the a consequence of this unique property, an appropriately
subject of a large amount of interest. As a result, there is designed vehicle can hover using only three rotors, which
currently a long list of potential quadrotor applications, for makes it possible to recover after rotor failure.
which many different control designs have been developed. The inability to restabilize after rotor failure has recently
As quadrotors only have the speeds of the four propellers as come into consideration. As a conventional quadrotor cannot
control channels, it is appealing to look beyond controller hover using only three rotors, several studies have considered
design to enhance the vehicle’s capabilities. It is natural the task of trying to regain some control of the vehicle [10–
to ask the question: how can a quadrotor’s capabilities be 12]. These control techniques are all similar in that they
enhanced by adding additional actuation? sacrifice control of the yaw direction in order to obtain a
There are many conceivable ways to add additional ac- desired position. The result is that these vehicles have a non-
tuation to the standard quadrotor. One of the simplest is to zero yaw rate which puts the vehicles into a spin. These
add additional rotors. This has led to the more general class studies have developed ways to control the position of the
of vehicles known as multirotors (e.g. hexarotors, octorotors, quadrotor which allow for landing or trajectory following.
etc.). Another idea is to tilt the rotors to change the direction However, these techniques result in an inability to control
of thrust. For example, designs have considered rotors which the yaw which renders the vehicle incapable of maneuvering
can rotate about their arm axis [1, 2] and rotors which can tilt the way in which it was designed, and may prevent certain
in a way in which both horizontal and vertical flight modes sensors from functioning properly or obtaining accurate state
can be achieved [3]. estimation. Several studies have also considered rotor failure
Another way to create additional actuation, which has in multirotors with greater than four rotors [13, 14]. Due to
seldom been explored, is to actively change the shape of their rotor redundancy, certain multirotor configurations can
the vehicle. Doing so will change the vehicle’s dynamics still hover despite failure of one or more rotors.
by creating internal inertial effects, as well as change the II. DYNAMICS
locations of thrust. This consideration leads to the broader
class of multibody dynamics, as the common rigid body A. System Overview
assumption of quadrotor dynamics is no longer valid. One We consider a quadrotor which has two morphing mech-
of the few designs which has explored multibody quadrotor anisms. The first mechanism is that each arm can change
dynamics is [4], which considered an actuated pendulum length. The second mechanism is that each arm is able
beneath the vehicle. Other designs have considered vehicles to rotate around the center of the vehicle, in the direction
with multibody dynamics [5–7], but the multibody aspect of perpendicular to the rotor axis. The vehicle consists of a main
these designs came from the addition of external loads or body in the center, four arm assemblies, and four rotors - one
manipulators. at the end of each arm. We model each arm as a mechanism
which consists of two pieces: a proximal arm closer to the
The authors are with the Department of Aeronautics and Astronautics, main body and a distal arm further from the main body. The
University of Washington, Seattle, WA 98195-2400 proximal arm is modelled as a hollow cylinder and the distal
This work was supported in part by an NDSEG Fellowship, in part by
ONR N000141712623, and in part by a grant from the Washington Joint arm is modelled as a solid cylinder which slides in and out
Center for Aerospace Technology Innovation. of the proximal arm (a prismatic joint). The proximal arm

978-1-5386-5428-6/$31.00 ©2018 AACC 4855

Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on March 22,2025 at 10:01:31 UTC from IEEE Xplore. Restrictions apply.
 z� = z�� η = [ψ θ φ]T . Noting that the transposes of the matrices
in (1) represent clockwise rotations, the rotation matrix from
2 A to B is defined as RAB ∈ SO(3) and is given by
y� x ��
z�� RAB = RTx (φ)RTy (θ)RTz (ψ).
3
d�
γ� The rotation matrix from B to A is given by RBA = RTAB .
z� Each arm is capable of planar rotations, in a direction
x� �
rc counterclockwise around the z-axis of the B frame. We
1
define four arm-fixed reference frames as Ci for i = 1, 2, 3, 4.
α�
y� We define the angles of rotation of the Ci frames about the
4 B frame as αi for i = 1, 2, 3, 4. The rotation matrix which
x�
x� converts a vector in Ci to a vector in B is thus
RCi B = Rz (αi )
Fig. 1. Diagram of the vehicle. The vehicle is comprised of a main
body (blue), proximal arms (black), distal arms (gray), motors (red), and
propellers (purple). To avoid clutter, variables αi , di and γi are only shown
for i = 1, 2, 3, 4. Each arm is defined to lie on the pos-
for arm 1, but are defined similarly for all arms. More specifically, arm- itive x-axis of its associated Ci frame. We define α =
fixed axes xCi are measured at an angles αi counter-clockwise from the [α1 α2 α3 α4 ]T . As an example of a “default” arm con-
body-fixed axis xB , and propeller-fixed axes xDi are measured at angles
γi counter-clockwise from the corresponding xCi axis. Also, not all y-axes
figuration, consider a “×” configuration where all arms are
of the frames are not shown, but are defined using right-handed coordinate 90◦ apart, such that arm 1 lies in between the x and y
systems. axes of the B frame. This configuration is described by
α = [45◦ 135◦ 225◦ 315◦ ]T .
Finally, we consider the rotation of the propellers. Denote
rotates about the center of the vehicle as a hinge joint. We the propeller-fixed reference frame as Di for i = 1, 2, 3, 4.
consider both the proximal and distal arms to be of equal We define γi as the angle of Di with respect to the
length l. We denote the inner and outer radii of the proximal corresponding arm reference frame Ci , and define γ =
arm to be aah and aal , respectively. We denote the radius [γ1 γ2 γ3 γ4 ]T . Note that the propeller angles themselves are
of the distal arm as ab . We define the distance from center not very useful; we are more interested in their derivatives
of the main body to the ith rotor as di . We define the tip (i.e. the propeller angular velocities and accelerations). The
of the distal arm to be at length di as well, and define rotation matrix for each of the propeller frames is thus
d = [d1 d2 d3 d4 ]T . We denote the masses of the main
body, each proximal arm, each distal arm, and each rotor as RDi Ci = Rz (γi )
mc , ma , mb and mr , respectively. We define the total mass
for i = 1, 2, 3, 4. We designate rotors 1 and 3 to rotate
as mt = mc + 4ma + 4mb + 4mr .
clockwise (as viewed from the top) while rotors 2 and 4
B. Reference Frames rotate counter-clockwise. By definition of the Di frames,
propellers 1 and 3 will spin in the negative γ1 and γ3
We will use several reference frames, and start by defining
directions while 2 and 4 spin in positive γ2 and γ4 directions.
an inertial frame A. We next define a body frame B, which is
We define the angular rate of the B frame’s rotation about
fixed at the main body of the vehicle. Let rc = [X Y Z]T
the A frame, expressed in the coordinates of A, as ω A BA ∈
denote the vector from the origin of A to the origin of
R3 (subscript denotes “the angular velocity of the B frame
B. To construct the rotation matrix between A and B, we
relative to the A frame”, superscript denotes “expressed in
first define the matrices which represent counter-clockwise
the coordinates of A”). This vector can be found by solving
rotations of an angle ζ about the x, y and z-axes as
  the equation [15, Ch. 5]
1 0 0 d
RBA = ω A


Rx (ζ) = 0 cos ζ − sin ζ  BA × RBA (2)
dt
0 sin ζ cos ζ
  where
cos ζ 0 sin ζ  
Ry (ζ) =  0 1 0  (1) 0 −a3 a2
− sin ζ 0 cos ζ (a)× :=  a3 0 −a1 
  −a2 a1 0
cos ζ − sin ζ 0
Rz (ζ) =  sin ζ cos ζ 0 . for a = [a1 a2 a3 ]T so that (a)× b = a × b for all b ∈ R3 .
0 0 1 Expressed in B coordinates, this angular velocity is
We will use the 3-2-1 Euler angles, which by convention, ωB A
BA = (RAB )ω BA . (3)
describe a sequence of three clockwise rotations from the
inertial frame to the body frame: first yaw ψ, then pitch We can find the angular velocities ω CCii A and ω Di
Di A in a
θ, then roll φ [15]. Define the vector of Euler angles as similar manner, after noting that RCi A = RBA RCi B and

4856
Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on March 22,2025 at 10:01:31 UTC from IEEE Xplore. Restrictions apply.
RDi A = RBA RCi B RDi Ci (i.e. we use (2) and (3) with Ci E. System energies
or Di in the place of B). We note that we have previously derived vectors for the
Later on, we will use the Euler-Lagrange equation to positions of the centers of mass of each component in the
derive the dynamics. Their are 18 degrees of freedom in our inertial frame A. We recall that the kinetic energy of a rigid
system, and we define the generalized coordinates q ∈ R18 body can be expressed as the sum of the translational kinetic
as energy of the center of mass plus the rotational kinetic energy
about the center of mass. Since the main body is fixed in the
q = [rTc η T αT dT γ T ]T .
B frame, its angular velocity expressed in B coordinates is
C. Position vectors ωB BA . The kinetic energy of the main body is thus

Recall that the ith arm and rotor are defined to lie on the 1 1
Tc = mc ṙTc ṙc + (ω B )T Jc ω B
BA .
positive x-axis of the corresponding Ci frame. The distances 2 2 BA
from the center of the Ci frame to centers of mass of the ith The rotors and distal arms translate, but do not rotate in the
proximal arm, distal arm, and rotor are l/2, di − l/2, and di , Ci frame, therefore the ith arms and rotors all have the same
respectively. So, in Ci we have angular velocity. The kinetic energy of the arms and rotors
  is
l l
rC,ai = e1 , rC,bi = di − e1 , rC,ri = di e1 1 1
2 2 Tai = ma ṙTai ṙai + (ω CCii A )T Ja ω CCii A
2 2
for i = 1, 2, 3, 4 where we let e1 , e2 , e3 ∈ R3 denote the 1 1
Tbi = mb ṙTbi ṙbi + (ω CCii A )T Jb ω CCii A
standard basis vectors. In the inertial frame A, we can denote 2 2
the position of the ith proximal arm, distal arm, and rotor as 1 1
Tri = mr ṙri ṙri + (ω CCii A )T Jr ω CCii A .
T
2 2
rai = rc + RBA (RCi B rC,ai ) Since the propellers have relatively low mass, we can neglect
rbi = rc + RBA (RCi B rC,bi ) their translational kinetic energy and potential energy. For the
rri = rc + RBA (RCi B rC,ri ) . ith propeller, the total kinetic energy is thus
1
Note that the terms in parentheses are the vectors to the Tpi = (ω D i
) T Jp ω D
Di A .
i

centers of mass of the arms and rotors in the B frame. 2 Di A

The total kinetic energy of the system is the sum of
D. Moments of inertia the kinetic energies of the main body, arms, rotors, and
We will model most bodies as cylinders. The inertia matrix propellers:
of a cylinder centered at the origin, coaxial with the x-axis, 4
X
and of mass m, radius r, and length h is T = Tc + (Tai + Tbi + Tri + Tpi ) .
i=1
m
diag 6r2 , 3r2 + h2 , 3r2 + h2


Jcyl,x (m, r, h) = Defining the potential energy of the system as V , we have
12
4
where diag(·) forms a diagonal matrix from its inputs. X
V = mc grTc e3 + g ma rTai e3 + mb rTbi e3 + mr rTri e3 .


Similarly, if this cylinder is coaxial with the z-axis, its inertia
m i=1
matrix is Jcyl,z (m, r, h) = 12 diag(3r2 + h2 , 3r2 + h2 , 6r2 ).
The main body is modeled as a cylinder of uniform mass F. Forces and generalized forces
mc with radius ac and height hc . In the B frame, its inertia The vehicle will be subject to non-conservative forces and
tensor about its center of mass is Jc = Jcyl,z (mc , ac , hc ). torques generated by the propellers. We will use the simple
We denote the moment of inertia about the center of mass of assumption that both are proportional to the square of the
the proximal arm and distal arm as Ja and Jb , respectively. propeller rotation rate. We describe the thrust force generated
The proximal arm is modeled as a hollow cylinder, p and to by the ith rotor as
simplify notation, we define the constant aa := a2ah + a2al .  2
In the Ci frame, we have Ja = Jcyl,x (ma , aa , l) and fri = c1 eT3 ω D i
Di A (4)
Jb = Jcyl,x (mb , ab , l). Each rotor is modeled as a cylinder where c1 is a constant determined by the propeller. This is a
of uniform mass mr , radius ar , and height hr . In the Ci very simplified relationship which models the air around the
frame, its inertia matrix with respect to its center of mass is propeller as static, and is most accurate when the vehicle is
Jr = Jcyl,z (mr , ar , hr ). hovering. We define the vector of the rotor forces as
The propellers have a relatively small mass, but spin at
high velocities, which means we must consider their inertia. f r = [fr1 fr2 fr3 fr4 ]T .
Consequently, we can neglect all propeller inertia except Since the thrust forces are in the positive z-direction of
that about the propeller axis. We define the propeller inertia the Di frame, they can be represented in the inertial frame
matrix as Jp = diag(0, 0, Jp,zz ). Finally, note that all inertia as
matrices are constant since they are defined in their respective
frames. Fri = RDi A (fri e3 ) .

4857
Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on March 22,2025 at 10:01:31 UTC from IEEE Xplore. Restrictions apply.
We note that the bodies which these forces are acting on are Summing all of the generalized forces, the total general-
the propellers, and rri are vectors in the inertial frame to the ized force vector is
lines of action of these forces. The generalized forces can be 4
X

determined by the relationship between a force and its point Q= Qf ri + Qτ ri + Qτ α + Qf d + Qτ γ .
of application (or any point on its line of action) [16], i=1
 T We define the inputs associated with the rotors, arm angle
∂ ṙri
Qf ri = Fri torques and arm extension forces as
∂ q̇
uα = [τα1 τα2 τα3 τα4 ]T
for i = 1, 2, 3, 4.
ud = [fd1 fd2 fd3 fd4 ]T
Next, we consider the aerodynamic torque on the pro-
pellers. Recalling the direction of propeller rotation, the uγ = [τγ1 τγ2 τγ3 τγ4 ]T
torque on the rotors will be in the positive γi direction for respectively. We define our full system input vector u ∈ R12
rotors 1 and 3, and in the negative γi direction for rotors as
2 and 4. We model this torque as being proportional to the
square of the speed of the propeller. For i = 1, 2, 3, 4 we u = [uTα uTd uTγ ]T .
have
G. Application of the Euler-Lagrange Equation
 2
τri = (−1)i−1 c2 eT3 ω Di
Di A (5) The Lagrangian of the system is the kinetic minus potential
energies,
where c2 is a constant determined by the propeller. The
L = T − V,
torque vector in the inertial frame is
and the Euler-Lagrange equation is
τri = RDi A (τri e3 ) .
d ∂L ∂L
− = Q.
We note that the bodies which these forces are acting on are dt ∂ q̇ ∂q
the rotors, which have an angular velocity of ω A Di A in the This differentiation above will result in an equation of the
inertial frame A. The generalized force can be determined by form Mq̈ + h = Q where the positive definite mass matrix
the following relationship between the torques in the inertial M is a function of q, and h is a function of both q and q̇.
frame and the angular velocity of the body on which they Solving for q̈ we have q̈ = M−1 (Q − h). Defining
are applied, T
x = qT q̇T

!T
∂ω ADi A
Qτ ri = τri where x ∈ R36 , the dynamics are given by
∂ q̇  
q̇
ẋ = G(x, u) := . (6)
for i = 1, 2, 3, 4. M−1 (Q − h)
We define the torques actuating the arm joints as ταi . Since The derivation of these dynamics is far too complicated
these torques solely actuate the α generalized coordinates, and tedious to perform by hand. Instead, we have used
the generalized coordinate associated with ταi is solely αi . symbolic math software. Also, the mass matrix M can
Noting that α are the 7th , 8th , 9th and 10th generalized be inverted symbolically, but yields a very complicated
coordinates, the generalized forces associated with the arm expression which is intractable to work with. Instead, we
joint torques are perform all integration and differentiation numerically, which
means we only have to invert numerical instances of M.
Qτ α = τα1 e7 + τα2 e8 + τα3 e9 + τα4 e10
H. Final Notes on the Dynamics
where e8 , e9 , e10 , e11 ∈ R18 are the standard unit vectors.
Besides those on the propellers, we did not consider any
We define the forces which extend the distal arm from the
aerodynamic drag forces on the vehicle. We have also used a
proximal arm as fdi , and the torque which actuates the ith
simple propeller model which neglects aerodynamic effects
propeller as τγi . Note that we have defined di as the extent
including forces perpendicular to the propeller plane (“H-
of the ith rotor from the body, and γi as the ith propeller
forces”). Since this paper mainly considers operations around
angle. Using reasoning similar as in the previous paragraph,
hover, we deem these assumptions to be appropriate.
by noting the locations of d and γ in q, the generalized
In summary, the morphing quadrotor is modeled as an 18
forces for arm extension and propeller torque are
degree of freedom system with 12 actuators. More or less, we
Qf d = fd1 e11 + fd2 e12 + fd3 e13 + fd4 e14 have taken a conventional quadrotor and added 8 additional
Qτ γ = τγ1 e15 + τγ2 e16 + τγ3 e17 + τγ4 e18 degrees of freedom (i.e. the arm rotations and extensions).
Since each of these degrees of freedom has its own actuator,
respectively. The “e” vectors in these equations are the 18- our vehicle has the same relative degree of underactuation
dimensional standard unit vectors. as a conventional quadrotor.

4858
Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on March 22,2025 at 10:01:31 UTC from IEEE Xplore. Restrictions apply.
 Also, in contrast to most models of conventional quadro- However, noting that at zero pitch and roll all gravitational
tors, we have included the propeller dynamics. This was and thrust forces point in the z-direction, the net force in
accomplished by adding an additional 4 degrees of free- the x and y directions is always zero. Thus, there are four
dom to our model, which is easily incorporated into the equations which describe hover equilibrium. These equations
Lagrangian framework. Although doing so makes our model are given in (7) where Fz denotes the net force in the z
more realistic, it also leads to poorer scaling since the direction of the B frame and Mx , My , Mz denote the net
propellers have much less inertia and much larger velocities moment about the x, y, z axes of the B frame.
than the other bodies.
In order to determine the required hover forces and
III. H OVER C ONFIGURATIONS torques, we note that both the motor torque and the arm
torque must counteract the aerodynamic torque on the pro-
A. Equilibrium Equations
peller. Also, there will be no external force on the arm joint.
A quadrotor with parallel thrust vectors can achieve hover So the forces and torques at hover are
at any location and at any yaw angle, but roll and pitch must c2 c2
be zero, otherwise the gravitational forces cannot balance τγi = (−1)i fri , ταi = (−1)i fri , fdi = 0 (8)
c1 c1
the thrust forces. Translational and rotational velocities must
also be zero, and thus we can denote the equilibrium point where the first two equations can be determined by using (4)
of our state x at hover as and (5).
T B. A 3-Parameter Technique to Find a Hover Configuration
x := [qT q̇ ]T
T Given That a Rotor is Failed or Degraded
q := [X Y Z ψ 0 0 αT d γ T ]T
T
The system (7) represents all possible static hover con-
q̇ := [0T14 γ̇ ] figurations of the vehicle. We now consider solving these
equations in the event a rotor has limited or zero thrust. Since
where 0j denotes the length-j column vector of zeros and α,
(7) consists of four nonlinear equations in twelve degrees
d, and γ̇ are defined as parameters which make it possible to
of freedom, analysis is tricky. In order to simplify, we will
hover. Note that there may be configurations of the vehicle
design a technique which captures the main mechanisms
which involve zero translation and rotation of the center of
which can be used to compensate for the unbalanced forces
the vehicle (i.e. rc and η do not change in time), but involve
associated with deficient thrust from a rotor. The result will
motion of the arms. However, we will define “hover” as zero
be that (7) reduces to one equation in three parameters.
dynamics of all bodies except the propellers.
These parameters will represent the two main methods to
The only way in which a conventional fixed quadrotor
compensate for a degraded rotor: moving adjacent rotors
can achieve hover is by equally distributing thrust among all
towards the failed rotor, and contracting the failed rotor arm
rotors. The geometric morphing capabilities of a morphing
and its opposite arm. Both of these shape changes serve the
quadrotor, however, allow hover to be achieved in many dif-
same function of compensating for the non-zero net moment
ferent thrust/geometry profiles. Since a hover configuration
on the vehicle due to loss of thrust from the degraded rotor.
is one in which the dynamics equal zero, we can find these
Describing hover conditions by three parameters will both
profiles by considering the statics problem of zeroing the net
simplify the analysis and allow us to guarantee when the
force and moment on the vehicle.
hover technique will work.
Proposition 1. Consider the morphing quadrotor described We consider the case in which a rotor is degraded in some
in Section II. A vehicle with a fixed configuration will be at way which limits its thrust. Without loss of generality we
hover equilibrium if and only if the following equations are designate rotor 4 as the degraded rotor. We assume that we
satisfied: lose control of rotor 4, and that it will follow a given profile
Fz = 0 = fr1 + fr2 + fr3 + fr4 − mt g over time. Accordingly, we lose τγ4 as a control input, and
we denote our reduced control vector as
Mz = 0 = τr1 + τr2 + τr3 + τr4
Mx = 0 = v := [uTα uTd τγ1 τγ2 τγ3 ]T .
4
Now u is a function of the reduced input v, and only has
   
X l l
sin αi −ma g − mb g di − − mr gdi + fri di 11 parameters which can be controlled.
i=1
2 2
In order to reduce (7) down to one equation, we will
My = 0 = restrict the vehicle geometry in the following way, which
4
is shown in Fig. 2. To start, we consider any default con-
   
X l l
cos αi ma g + mb g di − + mr gdi − fri di . figuration such that all arms are 90◦ apart. We define the
i=1
2 2
vector of arm angles α for such a configuration as α90 .
(7)
Without loss of generality, we define this configuration as
Proof. Static equilibrium in three dimensions can be defined α90 = [0 π/2 π 3π/2]T . We set arms 1 and 3 to be of
by six equations which correspond to zero force in x, y and length d1,3 and also restrict them to rotate mirrored to each
z directions and zero moment about the x, y and z axes. other, each at the variable angle α1,3 measured from the xB

4859
Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on March 22,2025 at 10:01:31 UTC from IEEE Xplore. Restrictions apply.
 y� We first note some properties of this equation. Our domain
2
of interest is fr4 ∈ [0, m4t g ] and α1,3 ∈ [0, π/2). On this
domain, the RHS of (12) has a non-negative numerator and
d2,4 positive denominator since the term −2lma is always made
x� positive by the 4d1,3 ma term, since d1,3 ≥ l. Therefore,
the RHS is a continuous, non-negative function which is
α1,3 d1,3 d1,3 α1,3
decreasing in both fr4 and d1,3 . Additionally, both d2,4
d2,4 and sin α1,3 are positive and continuous on the domain of
3 4 1 interest, so the LHS is continuous, increasing in α1,3 and
decreasing in d2,4 .
We need to find conditions for which the LHS can achieve
Fig. 2. Top view of vehicle geometry parameterized by d1,3 , d2,4 and α1,3 ,
corresponding to (9). Propellers are drawn differently to indicate velocity: equality for all fr4 ∈ [0, m4t g ]. The RHS reaches its
dark disks indicate high velocity, light disks indicate medium velocity, and minimum of 0 when fr4 = mt g/4 and its maximum at fr4 =
blades indicate near zero velocity. This diagram assumes the default 90◦ 0. The LHS can achieve equality at fr4 = mt g/4 when
configuration is α90 = [0 π/2 π 3π/2]T .
α1,3 = 0. So, by noting the continuity and monotonicity of
the LHS and RHS, a necessary and sufficient condition for
axis. We set arms 2 and 4 to be of the same length d1,2 . We showing (12) can be satisfied for all fr4 is showing that
will also define f r such that rotor 2 compensates for rotor the maximum of the RHS can be achieved by the LHS.
4’s reduced thrust (see Theorem 1). These restrictions are Thus, we want to maximize the LHS and minimize the RHS
summarized mathematically in the following equations: at fr4 = 0. The LHS is maximized at α1,3 = αmax and
 mt g  
−α1,3
 
d1,3
 d2,4 = dmin , and the RHS is minimized when d1,3 = dmax ,
4
 mt g − fr4   0  d2,4  which implies that (12) is a necessary and sufficient condition
fr =  2
 mt g  , α = α90 +  α1,3  , d = d1,3  . for (10).
    
4
fr4 0 d2,4 Theorem 2 reduces the task of determining the feasibility
(9) of the hover parameterization (9)-(10) to satisfying the in-
Theorem 1. Consider the morphing quadrotor described in equality (11). However, this inequality does not tell us much
Section II. Using (9) to define f r , α and d, hover can be at first glance. To gain some insight, consider the case in
achieved if and only if the following equation is satisfied: which dmin = l and dmax = 2l (i.e. the distal arm can slide
  all the way in and out of the proximal arm), and αmax = π/4
fr4
d2,4 mt − 4 = (i.e. arms can rotate 45◦ in each direction). In this case (11)
g
reduces to
[d1,3 (mc + 4ma ) + 2l(mb − ma )] sin α1,3 . (10)
mc + .6ma & 6.2mb + 9.7mr .
Proof. The full equations for hover equilibrium are ex-
pressed in (7), so we only need to show (7) becomes (10) Building a vehicle with main body, arm and rotor masses
under (9). Note that (4) and (5) imply τri = (−1)i−1 cc21 fri . which satisfy this inequality is plausible considering that
So, all equations in (7) evaluate to zero except Mx = 0, arms and rotors are usually much less massive than the
which becomes (10). main body where heavier components such as batteries and
In summary, Theorem 1 states that (9)-(10) guarantee payload are usually located.
hover. Next, we present a theorem which determines if the
IV. ROTOR FAILURE R ESTABILIZATION
physical design of a given morphing quadrotor can utilize
the hover parameterization in (9)-(10). A. Restabilization Technique
Theorem 2. Consider the morphing quadrotor described in We will now present a technique to restabilize the vehicle
Section II. Let the extension of all arms be constrained in the case a rotor’s thrust goes to zero instantly. This
by dmin ≤ di ≤ dmax and the rotation of the arms be assumption is meant to serve as a worst-case scenario which
constrained by |α1,3 | ≤ αmax . Designate the degraded rotor could, for example, represent failure due to a propeller break-
as rotor 4. If this rotor has any degraded thrust value, i.e. ing. Throughout the rest of the paper, we assume that we
fr4 ∈ [0, m4t g ], the 3-parameter hover parameterization (9)- can detect rotor failure instantaneously. The restabilization
(10) can achieve hover if and only if dmin , dmax and αmax technique is as follows.
satisfy • Stage I: As soon as a fault is recognized in a rotor or a
sin αmax mt rotor fails, brake all rotors and bring the arms into the
≥ . (11) hover configuration described by α and d in (9) which
dmin dmax (mc + 4ma ) + 2l(mb − ma )
Proof. We start by rewriting (10) as satisfy (10).
• Stage II: Once the arms are in the correct places,
sin α1,3 mt − 4 fgr4 reinitiate the three working rotors and stabilize the
= . (12)
d2,4 d1,3 (mc + 4ma ) + 2l(mb − ma ) vehicle using an equilibrium-stabilizing controller (14).

4860
Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on March 22,2025 at 10:01:31 UTC from IEEE Xplore. Restrictions apply.
 Stage I should happen as quickly as possible since the TABLE I
vehicle will be in free fall during this stage. Although this C ONSTANTS USED IN SIMULATION .
is undesirable, if shape reconfiguration can be accomplished Constant Value Units
quickly, a significant loss in height will not occur. The advan-
αmax 52 degrees
tages of our technique are that it is simple, computationally
c1 1.2 × 10−6 N s2 /rad2
undemanding, and would be easily implementable in real- c2 1.2 × 10−8 N m s2 /rad2
time. It may be possible to achieved better performance with mc , ma , mb , mr 400, 60, 30, 30 g
other control techniques, which is a possible direction of l, aal , aah , ab 13, 1.5, 2.2, 1.4 cm
future work. ac , hc , ar , hr 5, 7, 1.5, 5 cm
Jp,zz 4 × 10−6 kg m2
B. Controller Design
Stage I in the technique above requires determination of
a hover configuration, but does not require sophisticated parameterization (9)-(10). Before applying the Stage II con-
control techniques. In simulation we used proportional- troller, we let the Stage I controller operate over 1/4 second.
derivative controllers to brake the propellers and reconfigure We consider the state at the time of rotor failure as
the arms. Control for Stage II is slightly more complicated T
x0 = [0 0 .3 0 .3 .2 αT90 (1.5l)1T4 0T4 0T14 γ̇ eq ]T
as it requires stabilization of the full system around an
equilibrium point. Accordingly, we propose the following where 0j and 1j are length-j column vectors of 0’s and 1’s,
control for Stage II. and γ̇ eq is defined to be propeller velocities of a symmetric
Denote v as the control corresponding to hover equi- hover configuration where thrust is equally distributed be-
librium, determined by (8) and (9). We let α and d be tween all four rotors. Our desired final hover state is x with
determined by (9), where α1,3 , d1,3 and d2,4 are determined rc = η = 0.
by finding a solution to (10). We can find γ̇ by noting that After some tuning, the Q and R matrices in the LQR
the term in parentheses in (4) equals γ̇i at hover. We define problem were chosen to be
the error dynamics as
Q = blockdiag(103 I6 , 104 I4 , 107 I3 , 10−12 I3 , I14 , 10−2 I4 )
e := x − x
x R = blockdiag(I4 , I4 , 102 I3 )
e := v − v
v
where Ij denotes the j-dimensional identity matrix and
and we will linearize about the dynamics (6) as blockdiag(·) forms a block diagonal matrix of its inputs.
Snapshots in time are shown in Fig. 3, and the states and
∂G(x, u) ∂G(x, u) controls are plotted in Fig. 4. All states are stabilized despite
A= , B= . (13)
∂x ∂v the fact that the vehicle initially had non-zero pitch and roll.
x=x x=x
v=v v=v Although the controller is based on a linearization about
We will use the infinite-horizon continuous- the equilibrium point x, the simulation results suggest that
time linear-quadratic regulator (LQR) to generate the region of attraction extends a reasonable amount beyond
controls x. To further investigate, we tested several initial conditions
R ∞ for the system.
The LQR cost function is
J = 0 xT Qx + vT Rv dt. Selection of Q and R will offset from x with non-zero pitch and roll. The results, shown
be addressed in the next section. Solving the LQR problem in Fig. 5, show that the region of attraction includes most
e = −Ke
results in the control v x for the error dynamics, and attitudes in which the vehicle is not upside down.
thus for the system the control is
V. C ONCLUSIONS AND F UTURE W ORK
v = v − K(x − x). (14) We have presented the dynamics of a morphing quadrotor
and shown in simulation that it can regain hover after rotor
Note that we have implicitly assumed that the linearized failure. We now note some practical considerations regarding
system (13) is stabilizable. In simulation we have always the design of this vehicle.
found the linearized system to be stabilizable or marginally In our design, the rotors can’t slide more than halfway
stabilizable. Explicit calculation of stabilizability and con- in. If they could slide in closer, it would be easier to
trollability would be beneficial direction of future work. accomplish the 3-rotor hover. Also, the control inputs in Fig.
4 are achievable given appropriately chosen actuators (e.g.
C. Simulation
standard hobby servo and DC motors for a ∼1 kg vehicle).
The rotor failure recovery scenario in the previous section One concern about the design of this quadrotor is that
was simulated. We assume full arm extension, i.e. dmin = l if adjacent rotors are positioned too close to one another,
and dmax = 2l. The physical constants used in the simulation the propellers may collide. In this case, smaller diameter
are given in Table I. propellers may be preferable. A more effective idea would
With d2,4 = 1.05l and d1,3 = 1.95l, and assuming be to mount opposite pairs of propellers above and below the
αmax > 55◦ , Theorem 2 implies we can apply the hover arms, so that they do no collide. However, the aerodynamic

4861
Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on March 22,2025 at 10:01:31 UTC from IEEE Xplore. Restrictions apply.
Fig. 3. First .6 seconds of recovery maneuver as snapshots, stacked side-by-side from left to right. Propellers are drawn differently to indicate velocity:
dark disks indicate high velocity, light disks indicate medium velocity, and blades indicate near zero velocity.

Fig. 5. Region of attraction of x under (14) using simulation parameters

in Table I. Initial states are offset from x with non-zero pitch and roll.

[4] J. D. A. Becker, “Modeling and control of a quadrotor with dynamic

inertia,” Master’s thesis, University of Washington, 2013.
[5] F. A. Goodarzi, D. Lee, and T. Lee, “Geometric stabilization of
quadrotor UAV with a payload connected by flexible cable,” ArXiv
e-prints, September 2013.
[6] V. Lippiello and F. Ruggiero, “Cartesian impedance control of a UAV
with a robotic arm,” vol. 45, pp. 704 – 709, 2012. 10th IFAC
Symposium on Robot Control.
[7] S. Kim, S. Choi, and 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, pp. 4990–4995, Nov
2013.
[8] S. Datta, “Multicopters with variable flight characteristics,” July 2015.
US patent #: 15/112386.
[9] D. Wallace, “Dynamics and control of a quadrotor with active geo-
metric morphing,” Master’s thesis, University of Washington, 2016.
[10] A. Freddi, A. Lanzon, and S. Longhi, “A feedback linearization
Fig. 4. Evolution of states and controls in simulation. approach to fault tolerance in quadrotor vehicles,” IFAC Proceedings
Volumes, vol. 44, no. 1, pp. 5413 – 5418, 2011. 18th IFAC World
Congress.
[11] V. Lippiello, F. Ruggiero, and D. Serra, “Emergency landing for a
effects of this configuration would have to be carefully quadrotor in case of a propeller failure: A backstepping approach,”
considered. in 2014 IEEE/RSJ International Conference on Intelligent Robots and
Regarding future work, we plan to implement these results Systems, pp. 4782–4788, Sept 2014.
[12] M. W. Mueller and R. D’Andrea, “Relaxed hover solutions for mul-
in hardware on a prototype vehicle. We also plan to explore ticopters: Application to algorithmic redundancy and novel vehicles,”
other capabilities of morphing quadrotors. The International Journal of Robotics Research, vol. 35, no. 8,
pp. 873–889, 2016.
R EFERENCES [13] H. Alwi and C. Edwards, “Fault tolerant control of an octorotor using
LPV based sliding mode control allocation,” in 2013 American Control
[1] R. Falconi and C. Melchiorri, “Dynamic model and control of an over- Conference, pp. 6505–6510, June 2013.
actuated quadrotor UAV,” IFAC Proceedings Volumes, vol. 45, no. 22, [14] T. Schneider, G. Ducard, K. Rudin, and P. Strupler, “Fault-tolerant
pp. 192 – 197, 2012. control allocation for multirotor helicopters using parametric program-
[2] M. Ryll, H. H. Blthoff, and P. R. Giordano, “A novel overactuated ming,” in International Micro Air Vehicle Conference, 2012.
quadrotor unmanned aerial vehicle: Modeling, control, and experi- [15] R. M. Murray, S. S. Sastry, and Z. Li, A Mathematical Introduction
mental validation,” IEEE Transactions on Control Systems Technology, to Robotic Manipulation. Boca Raton, FL, USA: CRC Press, Inc.,
vol. 23, pp. 540–556, March 2015. 1st ed., 1994.
[3] K. T. Öner, E. Çetinsoy, E. Sirimoğlu, C. Hançer, M. Ünel, M. F. Akşit, [16] C. K. Liu and S. Jain, “A short tutorial on multibody dynamics,” Tech.
K. Gülez, and İ. Kandemir, “Mathematical modeling and vertical flight Rep. GIT-GVU-15-01-1, Georgia Institute of Technology, School of
control of a tilt-wing UAV,” Turkish Journal of Electrical Engineering Interactive Computing, 08 2012.
& Computer Sciences, vol. 20, no. 1, pp. 149–157, 2012.

4862
Authorized licensed use limited to: Polytechnic University of Bucharest. Downloaded on March 22,2025 at 10:01:31 UTC from IEEE Xplore. Restrictions apply.