Control Engineering Practice: P. Pounds, R. Mahony, P. Corke
Control Engineering Practice: P. Pounds, R. Mahony, P. Corke
Control Engineering Practice: P. Pounds, R. Mahony, P. Corke
a r t i c l e in fo abstract
Article history: Typical quadrotor aerial robots used in research weigh o 3 kg and carry payloads measured in hundreds
Received 27 April 2009 of grams. Several obstacles in design and control must be overcome to cater for expected industry
Accepted 9 February 2010 demands that push the boundaries of existing quadrotor performance. The X-4 Flyer, a 4 kg quadrotor
Available online 21 February 2010
with a 1 kg payload, is intended to be prototypical of useful commercial quadrotors. The custom-built
Keywords: craft uses tuned plant dynamics with an onboard embedded attitude controller to stabilise flight.
Robotics Independent linear SISO controllers were designed to regulate flyer attitude. The performance of the
Control system is demonstrated in indoor and outdoor flight.
Unmanned vehicles & 2010 Elsevier Ltd. All rights reserved.
Aerospace
Dynamics
Corresponding author.
1
E-mail addresses: [email protected] (P. Pounds), [email protected] Although similarly named, the ANU X-4 Flyer and CEA X4-flyer are quite
(R. Mahony), [email protected] (P. Corke). different craft.
0967-0661/$ - see front matter & 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.conengprac.2010.02.008
ARTICLE IN PRESS
692 P. Pounds et al. / Control Engineering Practice 18 (2010) 691–699
2. Drive system
Generally, hobby ESC microcontroller code and internals are is limited by slow rotor dynamics, this may be a crucial design
inaccessible; no direct ESC rotor speed measurement is available point.
externally, which may require additional sensors be added.
It was found that high-gain, closed-loop speed control around 3.1. Rigid body dynamics
the 50 Hz update rate of hobby RC equipment was not feasible for
the X-4 flyer. Programmable hobby ESCs now available can be
The inertial reference frame is denoted by I ¼ fEx ; Ey ; Ez g,
hooked up to PCs for fine-tuning—these may be adaptable for
where Ez is in the direction of gravity, and x ¼ ðx; y; zÞ is the origin
large quadrotor speed control. However, a commercial high-
of the body fixed frame A ¼ fEa1 ; Ea2 ; Ea3 g, where x is aligned with the
performance quadrotor will almost certainly use custom drive
front of the craft (see Fig. 2). The frame A is related to I by the
electronics, as is the case with the Ascending Technologies
rotation matrix R : A-I . Vectors v and x are the linear and
Hummingbird (Ascending Technologies GmbH, 2009).
angular velocities of the frame in A.
The equations are
2.3. Dynamic compensation
n_ ¼ Rv ð1Þ
Quadrotors must have fast thrust dynamics—the motors must X
be able to accelerate the rotors quickly to allow authorative mv_ ¼ mx v þ mgRT e3 þ ti ð2Þ
N;S;E;W
attitude stabilisation. Most current quadrotors have light rotors
that allow for fast speed changes without additional control. Large
quadrotors have heavier, high inertia rotors and thus need local R_ ¼ R skðxÞ ð3Þ
control to artificially improve the motor bandwidth. Reflected X
rotor inertia through any gearing should also be matched to the Ix
_ ¼ x Ix þ ½qi þmi ð4Þ
N;S;E;W
inertia of the motor to allow for maximum acceleration, although
this must be balanced against the added mass, complexity and 0 1
sina1s i
friction of a drive train. In practice, the closed-loop performance is B C
most heavily constrained by limits on the available instantaneous ti ¼ CT rAr2 o2i @ cosa1s i sinb1s i A ð5Þ
current draw on the batteries and this dominates the control cosb1s i cosa1s i
design.
Brushless motor speed dynamics are a single-pole dynamic qi ¼ CQ rAr 3 oi joi je3 ð6Þ
system, and proportional feedback control is suitable. The control
gain that can be realised by the torque-limited plant is bound by mi ¼ ti di ð7Þ
the maximum slew-rate that disturbance noise and sinusoidal
where m and I are the mass and rotational inertia of the flyer, g is
references may demand without inducing failure in the controller.
acceleration due to gravity, r is the density of air, r is the rotor
A method for calculating an optimised control design for a slew-
radius, and A is the rotor disc area. In Eq. (6), o is multiplied by its
saturated drive has been previously described (Pounds et al.,
magnitude to preserve the sign of rotation for counter-rotating
2009).
rotors.
Given sufficient bandwidth, the motor controller need not
Here sk(x) is the skew-symmetric matrix such that skðaÞb ¼ a
maintain precise rotor speed—the attitude control system for a
b for vectors in R3 .
full UAV will contain integral terms that will compensate motor
Rotors are indexed by their corresponding compass directions:
set-points to ensure flight stability of the vehicle.
North, South, East and West (NSEW), where N indicates the front
rotor. Correspondingly, di is the rotor displacement from the flyer
3. Quadrotor dynamics centre of mass:
dN ¼ ð0 d hÞ ð8Þ
Mathematical dynamic models of flight behaviour are essential
for good control design and analysis. A common model used to
represent quadrotor behaviour is that of Hamel, Mahony, Lozano,
& Ostrowski (2002). The most basic quadrotor model used
consists only of rigid body dynamics with abstract force and
torque actuators and no aerodynamics. The quadrotor is com-
monly represented as a rigid body mass with inertia and
autogyroscopics, acted upon by gravity and control torques.
Simple quadrotor dynamic models do not represent the
complex helicopter behaviour exhibited by real quadrotors. In
particular, they omit the blade flapping effect, which is critical to
understanding oscillatory helicopter modes, rotor flapping due to
yaw and variable rotor inflow velocities as a result of craft pitch
and roll.
Flapping dynamics are beginning to be recognised as impor-
tant aspects of quadrotor dynamics; even very small quadrotors
exhibit flapping (Huang, Hoffman, Waslander, & Tomlin, 2009).
The nature of the instability of quadrotor dynamics, oscillatory or
pure divergence, was shown to be dependent upon the height of
the rotor above the centre of mass; setting the rotors to be on, or
just above, the plane of the centre of gravity minimises the
sensitivity function of the system (Pounds, Mahony, & Corke,
2006). In the case of large quadrotors, where actuator bandwidth Fig. 2. Flapping quadrotor free-body diagram.
ARTICLE IN PRESS
694 P. Pounds et al. / Control Engineering Practice 18 (2010) 691–699
where DCT is the change induced by the changing inflow and g is the Lock Number (Leishman, 2006):
conditions. From Eq. (12), this is written as ra0 cr4
g¼ ð21Þ
a s Ib
DCTi ¼ 0 ðv þ x di Þe3 ð14Þ
4 oi r where Ib is the rotational inertia of the blade about the flapping
where a0 is the lift slope at the set point. hinge. Eq. (18) substitutes blade tip angle for collective pitch
The inflow velocity of the X-4’s rotors is very high with respect
to pitch, roll and translation velocities. Consequently, the vehicle
does not readily induce vortex ring states, even during aggressive
manoeuvres.
Aerodynamic parameters
Rotor, blade and aerodynamic parameters are obtained all masses, but includes all major masses—screws and
through measurement, computation, simulation or from fasteners are omitted (see Fig. 4).
references. These are listed in Table 1. Rotational inertia
Masses and displacements Computed from the previous values by treating the parts as
Component masses and distances measured with respect to point masses, the diagonal entries of the inertial matrix are
the rotor plane, (masses 70:005 kg, distances 7 0:005 m) are given in Table 3. The CoG is 0:0071 7 0:005 m above the rotor
given in Table 2. Note that this table is not a complete listing of plane.
ARTICLE IN PRESS
696 P. Pounds et al. / Control Engineering Practice 18 (2010) 691–699
@CT a 1 16
¼ s ð32Þ c3 ¼ ð45Þ
@y_ 8 oR go0
The characteristic equation of the system matrix determinant, The flapping angle is approximated as a linear function of x_
in canonical form of As3 + Bs2 + Cs + D, becomes and y_ :
1 @X 1 @M 2 g @M a1s ¼ c1 x_ þ c3 y_
s3 þ s þ ¼0 ð33Þ ð46Þ
m @x_ IYY @y_ IYY @x_
Using the system parameters and errors, the poles and zeros of
Solving for the roots of this polynomial gives the exponential the system are given in Table 4. The rotor height above the CoG is
components of the dynamic behaviour of the system. the predominant contributor to error, thus accurate knowledge of
Application of Routh’s Discriminant, as outlined in Prouty the rotor height is important to determining the dynamic model.
(2002, p. 602), uses the coefficients of the characteristic The unforced stability analysis demonstrated that h is also
polynomial, A, B, C and D, to determine the nature of the important in determining the behaviour of the dynamic system.
instability. From Eq. (33) The root locus for h shows that the structure of the open-loop
A¼1 ð34Þ
Table 4
Poles and zeros of the open loop pitch dynamics.
1 @X 1 @M
B¼ þ ð35Þ
m @x_ IYY @y_
Value Error
6. Attitude control
regulation performance of most controllers is within 7 21 of level discretised at ts = 0.02 s for the control design. The IMU returns
tracking, and the best in the range of 7 0:5213 . It is the authors’ both angle and rate information, which allows for a PID controller
assertion that the limiting factor in quadrotor dynamic control is in the improper form C = k(1 + i/s + d s) to be realised, where C is
the performance of the actuators. It has been suggested that less the controller transfer function, k is the proportional system gain,
complicated designs such as PID may, in fact, offer an advantage i and d are the integral and differential scalings and e is the system
due to their simplicity and potential robustness to parameter error. The complete discretised model, Gc ¼ y=do, is
variation (Bouabdallah, Noth et al., 2004). These qualities are
1:4343 105 ðz0:9916Þðz þ 1Þðz0:9997Þ
desirable for our full flapping model which is especially sensitive Gc ¼ ð49Þ
ðz0:2082Þðz0:9914Þðz1:038Þðz2 1:943z þ 0:9448Þ
to changes in h.
In addition to the attitude dynamics, the X-4 Flyer also has where do is the differential variation in rotor speed about the
important motor dynamics. The motor dynamics act in series with operating condition, 850 rad s1 . The additional zero at z = 1
the rigid body dynamics—fast motor response is important for comes from the matched pole-zero discretisation method.
authoritive attitude control of quadrotors. To this end, rotor speed
controllers have been developed to improve the natural perfor- 6.2. Controller design
mance of the rotor–motor system (Pounds et al., 2009). The
linearised closed-loop motor system transfer function, HM-CL, is
The proposed controller consists of a discrete PID controller.
68:85ðsþ 0:42Þ The transfer function of the controller, C, is
HM-CL ¼ ð48Þ
ðs þ 78:46Þðs þ 0:44Þ
0:02 ðz1Þ
C ¼ 400 1 þ 0:2 þ 0:3 ð50Þ
ðz1Þ 0:02
6.1. Discretised model As the motor dynamics are so fast, the dominant pole has little
interaction with the attitude mechanics. If it were slower, the
The controller runs at 50 Hz, the maximum frequency at which excess poles would diverge closer to the unit circle, leading to
attitude data are updated, and so the dynamics of the plant are oscillation and possibly instability. The slow motor pole-zero
cancellation is associated with the dynamics of the lithium
ion polymer cells, and sufficient gain causes the pole to close
with the zero.
Fig. 10. Outdoor flight autonomous pitch and roll angle stabilisation. Acknowledgements
conditions during testing. In this test the attitude controller was The authors would like to thank CSIRO ICT Robotics and Ryan
Pope for their ongoing support of this project.
ðz1Þ
C ¼ 400 1þ 0:3 ð51Þ
0:02
The zero integral gain caused the flyer to stabilise at non-zero References
angles. The X-4 lifted itself in ground effect (0.4 m) and regulated
its attitude within 7 11 of equilibrium (see Fig. 7). Ascending Technologies GmbH, (2009). /http://www.asctec.de/main/index.phpS.
Borenstein, J. (1992). The hoverbot—an electrically powered flying robot. /ftp://
For testing beyond ground effect the X-4 was flown tethered ftp.eecs.umich.edu/people/johannb/paper99.pdfS.
indoors. After engaging the attitude controller the suspended flyer Bouabdallah, S., Murrieri, P., & Siegwart, R. (2004). Design and control of an indoor
was hoisted up 1.5 m into the air before bringing the rotors to micro quad-rotor. In Proceedings of the international conference on robotics and
automation.
flight speed. A pilot sent attitude reference commands to the flyer
Bouabdallah, S., Noth, A., & Siegwart, R. (2004). Pid vs lq control techniques applied
to keep it centred in the test area; the pilot did not stabilise the to an indoor micro quadrotor. In Proceedings of the IEEE/RSJ international
vehicle. The X-4 flew at a height of approximately 2 m (see Fig. 8). conference on intelligent robots and systems.
The outdoor test took place on an ANU sports field. A smooth Draganfly Innovations Inc, (2009). /http://www.draganfly.com/S.
Guenard, N., Hamel, T., & Moreau, V. (2005). Dynamic modeling and intuitive
platform was used for take-off to allow the flyer to slide sideways control strategy for an ‘‘x4-flyer’’. In Proceedings of the fifth international
freely rather than catch and flip. To avoid integrator wind-up, the conference on control and automation.
X-4 was brought up to flight speed, then hopped into the air under Hamel, T., Mahony, R., Lozano, R., & Ostrowski, J. (2002). Dynamic modelling and
configuration stabilisation for an x4-flyer. In Proceedings of the 15th triennial
manual mode before switching to autonomous control. During the world congress of the international federation of automatic control.
flight a pilot sent commands to the flyer to control throttle but did Huang, H., Hoffman, G., Waslander, S., & Tomlin, C. (2009). Aerodynamics and
not stabilise the vehicle. The X-4 took off from the ground and control of autonomous quadrotor helicopters in aggressive maneuvering. In
Proceedings of the ninth international conference on robotics and automation.
flew to above 2 m and stayed airborne for 25 s (see Fig. 9). For ten Leishman, J. G. (2006). Principles of helicopter aerodynamics (2nd ed.). Cambridge,
seconds of the flight, the pilot made no stick corrections. In this UK: Cambridge University Press.
time the flyer regulated its attitude within 711 of level for 5 s Nice, E. (2004). Design of a four rotor hovering vehicle. Ph.D. thesis, Cornell
University.
(see Fig. 10). Pounds, P., Mahony, R., & Corke, P. (2006). Modelling and control of a quadrotor
robot. In Proceedings of the Australasian conference on robotics and automation.
Pounds, P., Mahony, R., & Corke, P. (2009). Design of a static thruster for micro air
8. Conclusion vehicle rotorcraft. Journal of Aerospace Engineering, 22.
Prouty, R. W. (2002). Helicopter performance, stability and control (1st ed.). London,
UK: Krieger Publishing Company.
A 4 kg quadrotor with a 1 kg payload was demonstrated in Seron, M., Braslavsky, J., & Goodwin, G. (1997). Fundamental limitations in filtering
flight. From the analysis of flyer attitude dynamics with flapping, and control (1st ed.). London, UK: Springer.
the mechanical design was tuned for best control sensitivity and Waslander, S., Hoffmann, G., Jang, J., & Tomlin, C. (2005). Multi-agent quadrotor
testbed control design: Integral sliding mode vs. reinforcement learning. In
disturbance rejection. A PID controller was designed to stabilise Proceedings of the IEEE/RSJ international conference on intelligent robots and
the dominant decoupled pitch and roll modes. In practice the craft systems.