Professional Documents
Culture Documents
Abstract: The problem of control law design for a small scale quadrotor helicopter enabling
both speed and pitch control is considered. The tradeoff between speed and pitch control is first
investigated with reference to a linear design problem. The overall control problem is then posed
on the pitch control and is formulated in terms of an outer (position) loop and an inner (attitude)
one. The design of the controller for position dynamics is based on the flatness property, while
attitude control is dealt with by means of an (almost) globally stabilising control law.
can be used to recover uniform handling qualities for trim φ̇ = p + sin(φ) tan(θ)q + cos(φ) tan(θ)r
conditions corresponding to different values of rotor rpm.
Finally, the flatness-based control architecture first pro- θ̇ = cos(φ)q − sin(φ)r (3)
posed in Formentin and Lovera (2011) is briefly reviewed sin(φ) cos(φ)
and adapted to the variable-pitch control problem. ψ̇ = q+ r
cos(θ) cos(θ)
where p, q and r are the attitude velocities in the B-frame,
2. QUADROTOR DESCRIPTION AND MODELING m is the vehicle mass, Ix , Iy and Iz are the body principal
moments of inertia, Jm is the motor inertia and ωR reads
The equations of motion for the quadrotor will be pre- ωR = −ω1 − ω3 + ω2 + ω4 . As is common practice in the
sented with reference to two coordinate frames: the Earth quadrotor literature, the input vector (U1 , U2 , U3 , U4 ) has
inertial reference frame (E-frame) and the Body-fixed been defined so that the state rates are linear in the control
reference frame (B-frame) - see Figure 1. The angular variables. More precisely, the Ui s are defined in terms of
the thrust and torque generated by each rotor (Li and TLi ,
i = 1, . . . , 4 respectively) as
4
X
U1 = Li ,
i=1
U2 = l(L4 − L2 ), (4)
U3 = l(L3 − L1 ),
U4 = −TL1 + TL2 − TL3 + TL4 ,
where l is the distance from the center of gravity of the
Fig. 1. Quadrotor reference frames. vehicle to the center of each of the rotors.
The flight mechanics model used in this study is based on
position (or attitude) of the quadrotor is defined by the the rigid body equations of motion (see (1)-(3)). The rotor
orientation of the B-frame with respect to the E-frame, forces and moments are computed using a combination of
whereas position is defined on the E-frame. Furthermore, blade element theory and momentum theory (see Prouty
we will rely on the following assumptions: (1990)). The lift curve slope for the rotor blades and the
(1) the origin of the B-frame is located at the center of rotor drag coefficients (2nd order polynomial formulation)
mass of the vehicle; are calculated from an experimental thrust vs power
(2) the body is rigid; curve obtained through fixed-point rotor thrust tests. A
(3) the axes of the B-frame coincide with the body linear inflow distribution along the blades is assumed,
principal axes of inertia. with a Prandtl tip-loss correction. The vehicle fuselage
is modeled as an aerodynamic surface, its influence being
Define the six degrees of freedom of the rigid body as described through suitable look-tables for the aerodynamic
q = [x, y, z, φ, θ, ψ]T , where the triple (x, y, z) represents coefficients. The latter are functions of the fuselage angle of
the position of the center of mass (in the E-frame) and attack with respect to the total flow, due to rotor induced
the ”roll-pitch-yaw” (φ, θ, ψ) set of Euler angles is the velocity and relative to air vehicle motion.
representation of the orientation of the quadrotor in the
same reference frame. The dynamical model can be derived The control variables are the collective pitch angles of the
by following, e.g., the Lagrangian approach and is given rotors in case of variable pitch (fixed RPM) architectures,
by or the rotor angular velocities in case of variable RPM
(fixed pitch) architectures. The considered quadrotor (see
(cos ψ cos φ sin θ + sin ψ sin φ) Table 1 for a summary of the main parameters) has
ẍ = U1 teetering rotor hubs allowing flapping motion; the rotor
m blades are assumed to be rigid and quasi-steady flapping
(sin ψ cos φ sin θ − sin φ cos ψ) is considered (computing only first harmonic flapping
ÿ = U1 (1)
m motion) in the model.
(cos θ cos φ)
z̈ = U1 −g
m name symbol value m.u.
mass m 5.1 kg
inertia on x Ix 0.1743 kg m2
(Iy − Iz ) 1 Jm inertia on y Iy 0.1774 kg m2
ṗ = qr + U2 − q ωR
Ix Ix Ix inertia on z Iz 0.3206 kg m2
(Iz − Ix ) 1 Jm rotor-motor polar inertia Jm 0.000658 kg m2
q̇ = pr + U3 + p ωR (2) body center-propeller distance l 0.415 m
Iy Iy Iy Table 1. Parameters of the considered (Aer-
(Ix − Iy ) 1 matica ANTEOS) quadrotor.
ṙ = pq + U4
Iz Iz
207
2013 IFAC ACA
September 2-6, 2013. Würzburg, Germany
3. OPTIMAL RPM SCHEDULING • the single rotor total power curve both in case of
nominal rotor RPM and using the optimal RPM (see
Figure 3);
• the percentage reduction of single rotor total power
using the optimal RPM respect to the nominal one
This Section investigates the possibility of varying the (see Figure 4);
rotor RPM on a variable pitch quadrotor as a function of • the percentage of rotor collective pitch as a function of
the forward velocity, so as to minimize power consumption forward velocity both for nominal and optimal RPM
(and therefore maximize rotor efficiency) and increase en- (see Figure 5);
durance (a critical aspect for electrically powered UAVs).
A similar strategy has been studied for manned conven- It appears from the Figures that the nominal rotor RPM
tional helicopters in, e.g., Gandhi et al. (2008). is optimized for hover at the maximum TOW. Therefore,
for decreasing TOW and increasing forward velocity larger
Performing model linearization through an iterative pro- power savings become possible. More precisely, it is pos-
cess, rotor controls (RPM or collective pitch) and vehicle sible to save up to 1% of required power for each rotor at
attitude angles were found to satisfy the vehicle forces maximum TOW, up to 1.7% at medium TOW and up to
and moments equilibrium equations for a defined steady- 4% at minimum TOW, with, respectively, a reduction of
state flight condition (trim solution). Analyzing the trim rotor RPM up to 9% for maximum TOW, 11% for medium
solution for increasing forward flight velocity (starting TOW and 17% for minimum TOW. The variation in
from hover), a well known behavior can be observed: the rotor RPM needed to reach the optimal condition is small
rotor total power (sum of induced, drag and H-force power enough to be handled without problems by the electrical
contributions) decreases with respect to the hover value motor and the related electronic speed controller, so that
until a minimum is reached at a certain velocity, and this strategy can be implemented in practice.
then increases again. This is due to the (well known) fact
that for increasing velocity the induced power decreases Remark 1. Rotor RPM reduction for increasing velocity
continuously but drag and H-force power increase. The has two potential drawbacks. First of all, fixing the per-
trim collective pitch (fixed RPM quadrotor architecture) centage of collective pitch to 70% limits the margin for
shows the same behavior as total power for increasing control and maneuver that would otherwise be available
forward velocity. in forward flight. Second, changing the RPM setting af-
fects the spectrum of vehicle vibration and therefore has
The above phenomena can be exploited when considering an impact on the design of the isolation of the Inertial
the preliminary design problem for a variable pitch quadro- Measurement Unit.
tor. Indeed, the maximum Take-Off Weight (TOW) is
0
obviously limited by the maximum available rotor thrust; min TOW
med TOW
furthermore, the pitch necessary for hover at maximum −2 max TOW
trim is less than the 70% used in hover and it becomes −12
possible to decrease the rotor RPM to reestablish the 70%
threshold for each value of forward velocity, obtaining a −14
208
2013 IFAC ACA
September 2-6, 2013. Würzburg, Germany
1.5
X Position [m]
1
100
0.5
−0.5
90 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0 1 2 3 4 5 6 7 8 9 10 Time [s]
Forward speed [m/s]
15
−1.5
Fig. 6. Linear control: responses to a step in desired longi-
−2 tudinal position at different flight conditions ranging
from 0 to 10 m/s (top: position; middle: quadrotor
−2.5 pitch attitude; bottom: commanded blade pitch).
−3
be seen from the Figure, even though the nominal rotor
RPM is decreased for increasing velocity, the scheduling
−3.5
mechanism can compensate for changes in the open-loop
−4
flight mechanics of the quadrotor and leads to almost
0 1 2 3 4 5 6
Forward speed [m/s]
7 8 9 10
identical responses over the considered range of velocities.
Similarly, the control effort is hardly affected by the
Fig. 4. Single rotor total power reduction (%) using opti- changes in the operating condition of the quadrotor.
mal RPM respect to nominal - varying TOW.
5. NONLINEAR FLATNESS-BASED CONTROL
70
opti RPM max TOW A nonlinear system ẋ(t) = f (x(t), u(t)), with time t ∈ R,
nominal RPM max TOW
opti RPM med TOW state x(t) ∈ Rn and input u(t) ∈ Rm , is said to be
nominal RPM med TOW
opti RPM min TOW
(differentially) flat (see, e.g., Fliess et al. (1995); Levine
65
nominal RPM min TOW (2009) for overviews of the theory of flat systems) if
there exists a set of m differentially independent variables
w = [w1 , . . . , wm ]T , called flat outputs, such that:
Collective pitch %
60
209
2013 IFAC ACA
September 2-6, 2013. Würzburg, Germany
Flatness-based quadrotor control It is apparent from (2) following: the new control vector is no longer given by
that (as already noticed in, e.g., Koo and Sastry (1999)) v = [v1 v2 ]T , but rather by v∗ = [v1∗ v2∗ ]T , where v∗ is
the control task can be decoupled in two subproblems defined as
concerning attitude control, handled via U2 , U3 and U4 , v1∗ = v1 cos(ψ) − v2 sin(ψ)
and position control, that can be handled with U1 and v2∗ = v2 cos(ψ) + v1 sin(ψ) (7)
the reference values φ◦ , θ◦ and ψ ◦ . Therefore the architec- v3∗ = v3 .
ture of the quadrotor control system is composed by the It follows that the real control variables can still be
following blocks. recovered from v, but this one now has to be computed as
• A trajectory generation block, that provides the v1 = w1 cos(ψ) + w2 sin(ψ)
quadrotor desired path in terms of the quadruple v2 = −w1 sin(ψ) + w2 cos(ψ) (8)
(x, y, z, ψ). v3 = w3 .
• A position control block, based on flatness of the po- The v∗ vector can be used to place the eigenvalues of
sition dynamics and exact feedforward linearization the linearized system as before via extended PID. Notice
theory (see Hagenmeyer and Delaleau (2003)). that in both the cases ψ = 0 and ψ 6= 0, this approach
• Three passivity-based attitude controllers, see For- allows complete freedom in the specification of the desired
mentin and Lovera (2011) for further details. closed loop performance as the nominal eigenvalues are
• A control allocation block, which determines the de- user-defined.
sired pitch angles on the basis of the computed Ui s. Attitude control is based on the so-called Modified Ro-
The position controller can be designed by considering the drigues Parameters (MRPs) σ. This set of coordinates is
reduced-order model given by (1), where the state-vector such that σ = eE tan (αE /4), where eE and αE are the
is x = [x, ẋ, y, ẏ, z, ż], the input vector is u = [U1 , φ, θ] Euler axis and angle that describe the three consecutive
and m, g and ψ are system parameters. The following rotations roll, pitch and yaw (see again Shuster (1993)).
result holds. As was shown in Formentin and Lovera Although MRPs still include the occurrence of discontin-
(2011), the system (1) is differentially flat. Indeed, consider uous jumps in the parameter space when incrementing
w = [w1 , w2 , w3 ]T = [x, y, z]T as the flat output vector and the rotation, only 2π-rotations are forbidden. Moreover,
note that w is a function of the state, specifically the linear it holds (see Tsiotras (2002)) that the linear Proportional
selection Derivative (PD) control law u = −Kp σ − Kd ω almost
"
10 0 0 00
# globally asymptotically stabilizes the attitude dynamics
w = G(x) = Gx , G = 0 0 1 0 0 0 . at the origin, for any positive value of Kp and Kd 1 . The
00 0 0 10 linear controllers can be tuned based on the linearized
model for the attitude.
Obviously, first derivatives of x are linearly connected to
the first derivative of w and x = fx (w, ẇ), with fx = I6×6 . Finally, control allocation, i.e., mapping the Ui s into the
Moreover, concerning the input vector, it can be shown rotor pitch angles αi s, amounts to the solution of equations
that (4) with
p p
U1 = m ẍ2 + ÿ 2 + (z̈ + g)2 = m w¨1 2 + w¨2 2 + (w¨3 + g)2 = mẅ, ∂L ∂L
Li = Ωi + αi , i = 1, . . . , 4 (9)
and that φ and θ are expressed as in (5). Therefore, ∂Ω ∂α
w = G(x) = Gx is such that x and u can be written ∂T ∂T
as functions of w and its derivatives, hence w is a flat Ti = Ωi + αi , i = 1, . . . , 4 (10)
∂Ω ∂α
output with ρ = 1. where the numerical values of the control derivatives have
Assuming now that the yaw angle ψ is zero, it can be been derived from the experimental characterisation of the
shown that the original control variables U1 , φ◦ and θ◦ considered rotors. As the above equations are linear with
are obtained as functions of new inputs vi , i = 1, . . . , 3, as respect to the αi s, the control allocation problem can be
q solved just like in the case of variable RPM quadrotors.
U1 = m v12 + v22 + (v3 + g)2 The proposed control strategy has been tested using,
!
◦ v2 again, the model for the Aermatica ANTEOS quadrotor.
φ = arcsin − p 2 (6) The response of the closed-loop system to a step excitation
v1 + v22 + (v3 + g)2
of the reference for longitudinal motion is illustrated
v1
θ◦ = arctan . in Figure 7. As can be seen from the Figure, a very
v3 + g satisfactory response can be achieved.
The new inputs are then chosen as the combination of
a feedforward part taking into account the desired rates 6. CONCLUSION
of the states, returned by the trajectory generator, and a
feedback part Λ taking the tracking error into account: In this paper the control of variable pitch quadrotors
vi = ẋoi,2 + Λi (e), i = 1, 2, 3. For the correction term has been addressed. First, the problem of maximising the
Λ, in this work extended PID control is considered, see efficiency of the quadrotor by scheduling the rotor RPM as
Hagenmeyer and Delaleau (2003); Formentin and Lovera a function of the flight condition has been considered and it
(2011). Consider now the more general case in which ψ 6= 1 Note that almost global asymptotic stability can only be guaran-
0. Note that the effect of yaw on position dynamics can teed, as obviously singular points are not considered. This is a general
be interpreted geometrically as a rotation of the control point, since the topological structure of the attitude motion is not
action on the plane (x, y), so that the modification to a contractible space and it does not allow for globally continuously
the control law to account for nonzero yaw is simply the stabilizing control laws.
210
2013 IFAC ACA
September 2-6, 2013. Würzburg, Germany
w¨1 sin ψ − w¨2 cos ψ w¨1 cos ψ + w¨2 sin ψ
φ = arcsin , θ = arcsin (5)
ẅ w¨1 sin ψ − w¨2 cos ψ
ẅ cos arcsin
ẅ
1.2
tion. In 50th IEEE Conference on Decision and Control
1 and European Control Conference, Orlando, USA.
X Position [m]
10
Koo, T.J. and Sastry, S. (1999). Differential flatness
0
−10
based full authority helicopter control design. In IEEE
−20
Conference on Decision and Control, Phoenix, USA.
−30
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
La Civita, M. (2003). Integrated Modeling and Robust
Time [s]
Control for Full-Envelope Flight of Robotic Helicopters.
Ph.D. thesis, Carnegie Mellon University.
Fig. 7. Response of longitudinal axis: quadrotor position Lee, T., Leok, M., and McClamroch, H. (2010). Geometric
(top) and pitch attitude (bottom). tracking control of a quadrotor UAV on SE(3). In IEEE
has been shown that power reductions can be obtained by Conference on Decision and Control, Atlanta USA.
this approach. Subsequently, linear, gain-scheduled pitch- Levine, J. (2009). Analysis and Control of Nonlinear
based controllers for the quadrotor have been designed Systems: A Flatness-based Approach. Springer.
to recover uniform handling qualities for trim conditions Mellinger, D. and Kumar, V. (2011). Minimum snap
corresponding to different values of rotor RPM. Finally, a trajectory generation and control for quadrotor heli-
flatness-based nonlinear control architecture first proposed copters. In IEEE International Conference on Robotics
for variable RPM quadrotors has been adapted to the and Automation, Shanghai, China.
variable-pitch control problem. Nonami, K., Kendoul, F., Suzuki, S., Wang, W., and
Nakazawa, D. (2010). Autonomous Flying Robots:
REFERENCES Unmanned Aerial Vehicles and Micro Aerial Vehicles.
Springer.
Aermatica (2008). The ANTEOS remotely piloted air- Prouty, R.W. (1990). Helicopter Performance, Stability,
craft. http://www.aermatica.com/PRODUCTS.html. and Control. Krieger Publishing Company.
Bottasso, C., Leonello, D., and Savini, B. (2008). Path Shuster, M. (1993). A survey of attitude representations.
planning for autonomous vehicles by trajectory smooth- Journal of Austronautical Sciences, 41(4), 439–517.
ing using motion primitives. IEEE Transactions on Tsiotras, P. (2002). Further passivity results for the atti-
Control Systems Technology, 16, 1152–1168. tude control problem. IEEE Transactions on Automatic
Castillo Garcia, P., Lozano, R., and Dzul, A. (2010). Mod- Control, 43(11), 1597–1600.
elling and Control of Mini-Flying Machines. Springer. Van Nieuwstadt, M. and Murray, R. (1998). Real-time
Cowling, I., Yakimenko, O., Whidborne, J., and Cooke, trajectory generation for differentially flat systems. In-
A. (2010). Direct method based control system for ternational Journal of Robust and Nonlinear Control,
an autonomous quadrotor. Journal of Intelligent and 8(11), 995–1020.
Robotic Systems, 60, 285-316.
Cutler, M. and How, J. (2012). Actuator constrained tra-
jectory generation and control for variable-pitch quadro-
tors. In AIAA Guidance, Navigation and Control Con-
ference, Minneapolis, USA.
Cutler, M., Ure, N.K., Michini, B., and How, J. (2011).
Comparison of fixed and variable pitch actuators for
agile quadrotors. In AIAA Guidance, Navigation and
Control Conference, Portland, USA.
Das, A., Subbarao, K., and Lewis, F. (2009). Dynamic
inversion with zero-dynamics stabilisation for quadrotor
control. IET Control Theory & Applications, 3(3), 303–
314.
Fliess, M., Lévine, J., Martin, P., and Rouchon, P. (1995).
Flatness and defect of non-linear systems: introductory
theory and examples. International Journal of Control,
61(6), 1327–1361.
Formentin, S. and Lovera, M. (2011). Flatness-based con-
trol of a quadrotor helicopter via feedforward lineariza-
211