Sunteți pe pagina 1din 6

IEEE ISIE 2006, July 9-12, 2006, Montreal, Quebec, Canada

Rest-to-rest Maneuvering of a Nonholonomic Control Moment


Gyroscope
Mahmut Reyhanoglu and Jasper van de Loo

Abstract- This paper studies a restricted control moment technique via both numerical simulations and experimental
gyroscope (CMG), which constitutes a remarkable example of results.
a nonholonomic system that arises as a consequence of its
symmetry properties. A nonholonomic control system formula- II. MATHEMATICAL MODEL
tion is first introduced for the restricted CMG dynamics. The We first briefly describe the control moment gyroscope
geometric phase technique is then used to design a feedback
control algorithm for the control of the CMG system to a (CMG) unit, by Educational Control Products (ECP) [16],
desired equilibrium. The effectiveness of the feedback control shown in Figure 1. The unit consists of a high inertia brass
algorithm is illustrated through both numerical simulations and rotor suspended in an assembly with four angular degrees of
experimental results. freedom. The rotor spin torque is provided by a dc motor
I. INTRODUCTION (motor 1) whose angular position is measured by an optical
In the last few decades, there has been considerable
encoder (encoder 1). The first transverse gimbal assembly
attention paid to control problems for nonholonomic systems. (body C) is driven by another dc motor (motor 2) to affect
motion about axis 2. Another optical encoder (encoder 2)
Numerous examples have been studied in the context of robot
provides feedback of the relative position of bodies C and
manipulation, mobile robots, wheeled vehicles, and space
B. Body B rotates with respect to body A about axis 3.
robotics. An excellent reference that provides a geometric
and control theoretical view of nonholonomic systems is There is no active torque applied about this axis. A brake,
which is actuated via a toggle switch on the Controller box,
the book by Bloch [3]. A few representative control works
may be used to lock the relative position of A and B and
include the study of controllability and stabilizability in [4],
hence reduce the system degrees of freedom. The relative
motion planning in [13], [14], [17], and feedback stabiliza-
tion and tracking in [2], [7]-[10], [19]-[21].
angle between A and B is measured by encoder 3. Finally,
In this paper, we consider a restricted control moment body A rotates without actively applied torque relative to the
gyroscope (CMG) which constitutes a remarkable example
base frame (inertial ground) about axis 4. The axis 4 brake is
of a nonholonomic system that arises as a consequence of controlled similarly to the axis 3 brake and an optical encoder
its symmetry properties. It is well known that if the motion (encoder 4) provides position feedback. Inertial switches are
of a mechanical system exhibits certain symmetry properties, installed on bodies A, B, and C to sense any overspeed.
there exist conserved quantities. If these conserved quantities
are not integrable, then a class of nonholonomic systems
is thereby obtained. The restricted CMG system has a
cyclic rotation coordinate that gives rise to a nonintegrable
generalized angular momentum conservation expression and,
therefore, belongs to this class. Other examples of this class
of nonholonomic systems include internally actuated multi-
body spacecraft ([18], [22]), underactuated rigid spacecraft
([11], [12]), and space robotics systems [6].
This paper uses the geometric phase technique to design a
feedback control algorithm for the rest-to-rest maneuvering
of the CMG system. The key steps in the design of the
control algorithm can also be found in [4]. Our contributions
in this paper are (i) the formulation of the restricted CMG
dynamics as a nonholonomic control system and (ii) the
demonstration of the effectiveness of the geometric phase
This work was supported in part by Embry-Riddle Aeronautical Univer-
sity
M. Reyhanoglu is with the Department of Physical Sciences,
Embry-Riddle Aeronautical University, Daytona Beach, Florida, USA
reyhanom@erau.edu Fig. 1. Control Moment Gyroscope (CMG) by ECP [16].
J. van de Loo is with the Department of Mechanical Engineering,
Eindhoven University of Technology, 5600 MB Eindhoven, The Netherlands
j.v.d.loo@student.tue.edu We now present the equations of motion for the CMG
1-4244-0497-5/06/$20.00 © 2006 IEEE 160
in a form suitable to the subsequent controls design and the Lagrangian L:
analysis. Figure 2 is a schematic representation of the 4
DOF gyroscope. We follow the notation used in [16]. As L =2 (Ic +ID)q2 + 2(J2 + J1 sin2 q2)4
shown in Figure 2, dextral sets of orthogonal unit vectors + JD 1 4 sin q2,
ai, bi, ci,and di, (i = 1,2,3) are fixed in A, B, C,
and D, respectively. An inertial (or Newtonian) reference where
frame is defined as N, in which a dextral set of orthogonal J1 = JC + JD -ID -KC,
unit vectors Ni, (i = 1, 2, 3) are fixed. Four angles specify J2 = ID + KA + KB + KC.
the configuration of the system. The angular travel of D
(rotor) in C in the direction d2 is defined as ql. Angle The Lagrangian is not dependent on the angular positions
q2 is defined as the rotation about c1 of C relative to B. qi and q4. This is explained by inspection of Figure 2,
Similarly, angle q3 is defined as the angular rotation about where it is seen that the dynamic description of the system
b2 of B relative to A. Finally, q4 is defined as the angular would be identical for any arbitrarily assigned positions of
rotation about a3 of A relative to N. The angular speeds are the rotor and base. Clearly, q4 is an unactuated cyclic rotation
wi = qj, (i = 1, , 4). The configuration shown in Figure coordinate and thus the generalized momentum conjugate to
2 reflects qi 0 (i = 1, , 4). For this system, the mass this coordinate is conserved, i.e.
centers of all the bodies comprising the system are at the
center of the disk (D), which is also the center of all the . (JD sin q2>h1 + (J2 + J1 sin2 q2) 4 = C,
gimbal axes. Thus, only rotational dynamics are considered where c is a constant. In this paper, the constant c is set equal
in the following analysis and the effects of gravity are to zero in order to study rest-to-rest motions. Consequently,
neglected. The scalar moments of inertia about the ith (i using the Lagrangian method, the reduced equations of
1, 2, 3) direction respectively in bodies A, B, C, and D are motion can be written as
given by I, JX, Kx, (x = A, B, C, D). Note that only
moments of inertia are considered while products of inertia JDy1 + (JD sin q2)q4 + q2q4 JD cos q2 = T1, (1)
are set to zero. The CMG unit is such that this simplification (Ic + ID)q2-(JD 4l + Jlj4sinq2>)u4cosq2 =T2, (2)
is valid for most dynamic and control modeling purposes.
Two inputs are considered for this system. The first is a (JD sin q2)ql + (J2 + J1 sin2 q2)q4 = 0- (3)
torque, T1, applied to D by C (via the rotor spin motor). The first two equations represent the directly actuated de-
The second input is a torque, T2, applied to C by B (via grees of freedom and the last equation corresponds to the
a gimbal motor/capstan drive). The equations of motion for unactuated degree of freedom. Note that the equation asso-
the four degree of freedom CMG can be found in [16]. ciated with the locked degree of freedom is discarded.
The equations of motion for the system can be further
A. Reduced Equations of Motion simplified by applying a partial feedback linearization. Let
j = q4 denote the unactuated degree of freedom. Then
equations of motion can be written as
1 = Ui, (4)
f. q2 = U2, (5)
JD sin q2
(6)
0 J2 + J1 sin q2
7T NTZ where
T1 .
q4sinq2 -q4COS q2,
F"-. .. 6 1 .. .
ul =
JD-Q (7)
JD

+2 [T2 + (JD + Jl z4 sin q2)j4 cos q2] (8)


K0o: IC + ID
arethe angular acceleration input variables.
The state equations can now be written as
Fig. 2. Schematic representation [16]. 1 = Wi, (9)
q2 2, (10)
In this paper, we consider a simplified model, where JD sin q2
(1 1)
gimbal axis 3 (see Figure 2) is locked in its initial position J2 + J1 sin2 q2
so that throughout the motion i ui, (12)
W2 U2= (13)
q3 = WJ3 = WJ3 = 0.
It is straightforward to show that the above system is small-
Since the center of mass is fixed, the kinetic energy is also time locally controllable (see e.g. [4]).
161
III. FEEDBACK CONTROL ALGORITHM for the second order system
Geometric phase method can be used to transfer the 1 X2,
system from any initial rest configuration to the origin. Let
(q°, q°, qO) denote an initial configuration, where (q 1, q2) are X2 U
the base space variables and r1 is the fiber space variable. The transfers any initial state (Xl(0),X2(0)) to the final state
next two steps describe the design of a control strategy which (x1, 0) in finite time. Note that this feedback control is the
transfers the initial state to the origin. usual time-optimal bang-bang control law for a pure inertia
The first step is to find a control which brings the base plant with u < k.
space variables to the origin in a finite time, i.e. to (0, 0, rj 1) We use the following control algorithm in our simulations
for some rj1, at time t1 > 0. The next step is to traverse a and experiments:
closed path ty in the base space (or series of closed paths)
to transfer (0, 0,r1l) to (0, 0, 0). This path starts and ends Step 0 : Transfer the system to the origin of the base
at the origin of the base space. The closed path -y can be space in finite time, i.e. to the rest configuration (q 1, q2, TI) =
determined using the geometric phase condition, which is (0, 0, rj1) for some Tll, at time t, > 0; then go to Step 1;
given by Step 1 : Choose (a, b) to achieve the desired geometric
AI =j -T jJDf j2 dql,
+ J, sin 2q2
aJ2
(14) phase and then go to Step 2;
Step 2: Set
The geometric phase is reflected in the fact that traversing a Ul = Uk1 (q -a, wil), U2 -kpq2- kvw2
closed path in the base space yields a non-closed path in the
full configuration as shown in Figure 3. until (ql, q2) = (a, 0); then go to Step 3;
Step 3: Set
u1 =-kp (q -a)- k,w, U2 = Uk2 (q2 b, w2),
until (ql,q2) = (a, b); then go to Step 4;
A: Step 4: Set
Ul = Uk(qi, w), U2 -kp(q2- b)- k2,
until (ql, q2) = (0, b); then go to Step 5;
Step 5 : Set

Fig. 3. Geometric phase.


u1 -k,qi- k,i, U2 = Uk2 (q2, w2),
until (ql, q2) (0, 0); then go to Step 1.
More information about the geometric phase technique
can be found in [4]. In this paper, we will use a Here we assumed that the desired geometric phase can be
rectangular path -y in the qjq2-plane with corner points obtained by a single closed path. Clearly the above algorithm
(0,O), (a,O), (a,b), (O,b), where a and b are design can be modified to account for cases for which more than
parameters determined using the desired geometric phase one closed path is required. Step 0 of the control algorithm
condition can be accomplished in different ways. For instance, one
(JD sinlb can sequentially apply bang-bang control to one of the
J2 + Ji sin b) (15) axes while keeping the other locked in its position to zero
We now describe a feedback algorithm to implement the one base variable at a time, or bang-bang controls can
above ideas. Throughout assume k > 0 and let be simultaneously applied to both axes to drive both base
variables to zero. Note that, in Steps 2-5 of the algorithm,
Sk (Xl, X2) =1 -+ 21x2 X2 while the bang-bang control is applied to one of the axes, a
Define the following feedback function PD-controller with gains kp and k, is used in these steps to
lock the other in its position.
Uk (Xl,X2) { -k sign(sk (Xl X2)) if
-k Sk(Xl, X2) 7 0,
sign(X2) if Sk (X1, X2) = 0, IV. SIMULATION AND EXPERIMENTAL RESULTS
where sign(.) is the signum function defined as It is well-known that friction is a major source of loss of
performance in mechanical systems. The friction torque is
-1 i'f z < O, most simply modeled as the sum of Coulomb and viscous
sign(z) = { O f z = 0, terms. In our simulations, we apply the above control al-
I if z >0. gorithm to the following modified equations of motion that
We use the well-known property that the feedback control include the effects of friction:
U= Uk(X1 X1, X2) JD4l + (JD sinq2)q4 +q2q4JDcosq2 = Tl +.fl,
162
TABLE I 0.2
1
FRICTION COEFFICIENTS 0.1
E

Axis Ic (N m) |, (N m slrad) F-
-0.1
1 0.035 0.0
-0.2
2 0.017 0.0 0 10 15 20 25
4 0.005 0.1
0.2

0.1
E

O-0
(IC + ID)q2- (JD 41 + Jl 14sin q2> j4 cos q2 = T2 + f2,
sin2 q2)q4 + Jl2j4 sin 2q2
-0.1 _
(JD sin q2) 1 + (J2 + Ji 0 10
time [s]
15 20 25

+JDjl12 cosq2 = 4,
Fig. 5. Input torques Ti and T2 (simulation).
where fi = -p,usign(tQ) -,uVA4 ,ci > 0, ,vi > 0,
denotes the Coulomb plus viscous friction torque acting on 501
the ith axis (i = 1, 2, 4). Here pci and ,vi are Coulomb and
viscous friction coefficients, respectively. For the CMG unit,
these friction coefficients are experimentally determined as -50
10 15 20 25

in Table I. o

-1 J
Remark: Here we consider the simplest model of friction. -2
10 15 20 25
A considerably more sophisticated friction model developed
by Dahl [5] includes "stiction", the increased friction found 7~3 _-

when the relative velocity between the sliding surfaces is ,7 2-


1
zero. Following the work of Dahl, several dynamic friction 10
time [s]
15 20 25

models (LuGre model, Leuven model, Maxwell slip model)


have been proposed for accurate friction compensation. The Fig. 6. Configuration variables ql, q2 and r1 (experiment).
modifications of Dahl model for accurate friction compen-
sation can be found in [1] and [15]. 02
012
H-7
50 E

01
-02 L
-50 0 10 15 20 25
0 10 15 20 25

E
z 0.1
-J
0.05-
-2
0 10 15 20 25
4- 005

01
v 3 0 5 10 15 20 25
2 time [s]

0 10 15 20 25
time [s] Fig. 7. Input torques T1 and T2 (experiment).

Fig. 4. Configuration variables ql, q2 and r1 (simulation).


A single path -y with parameters (a, b) = (-23.1169, 7/4)
A. Results for Single Geometric Phase Path is used to achieve the desired geometric phase. The control
gains for the bang-bang and the PD controllers are chosen
The CMG is simulated using the feedback control algo- as
rithm described in the previous section. Moment of inertia
parameters are given as
ki = 5, k2 = 4, kp = 200 k = 40.
Simulation results are shown in Figures 4 and 5. We also
JD = 0.027 kg.m2, J1 = 0.013 kg.m2, J2 = 0.134 kg.m2
perform an experiment using the same control parameters as
The desired CMG configuration is given by (ql,q2,i ) = used for the simulation. Figures 6 and 7 show the results of
(0, 0, 0) and the initial configuration is (q°, qo, r0) = the experiment. It can be seen that the results of the simula-
(127, -1.36, 7) rad. The control algorithm described in tion and the experiment do agree. The torque levels and the
Section III was employed. Step 0 of the control algorithm final values of the configuration variables are approximately
was executed by driving first the base variable q2 and then the same. Although qi and q2 reach the desired zero values,
q1 to zero, thereby keeping the fiber variable r1 constant. rj converges to a nonzero value (of approximately 2 rad).

163
TABLE II 20

GEOMETRIC PHASE PATH PARAMETERS 0

-20
0 5 10 5 20 25 30 35 40
a (rad) b (rad)l
-yl -23.117 -F7/4 -A-F-V
-72 - 13.825
-7.667
-/4
-/4
-1[--
-M -2

-N -3. 198 -/4 0 5 10 5 20 25 30 35 40

-75 -0.5107 -/4


0
0 5 10 5 20 25 30 35 40
time [s]

As confirmed by both the simulation and the experiment,


Fig. 9. Configuration variables ql, q2 and r1 (simulation).
the offset in r1 can be attributed to the friction at the fourth
axis which is not directly actuated. 0.2 2 ]l

B. Results for Multiple Geometric Phase Paths


Ef
o irL -h h

As shown above, in the case of a single geometric path, -0.2


0 5 10 15 20 25 30 35 40

while the directly actuated variables qi and q2 reach the 0.15 _

desired zero values, due to friction the unactuated variable rj


does not approach the desired value zero. A solution to this 2
F-,
0.05

I
0
problem is to repeat the five steps described in the previous -0.05
u 0 u
section until r1 reaches its desired value. This process is -0
5
LJ

10 15
LJ

20
Lu
time [s]
Ll

25
LJ

30 35 40

implemented both in the simulation and the experiment.


For this case, the desired CMG configuration is the zero Fig. I10. Input torques T, and T2 (simulation).
configuration and the initial configuration is (q °, qo, fq°)
(67, -1.36, 7) rad. The control algorithm described in 20

Section III was employed using multiple geometric phase


paths. Five rectangular paths were used to achieve the desired 7-20

configuration. The path parameters are given in Table II. The 0 5 10 15 20 25 30 35 40

results of the simulation are shown in Figures 8, 9, and 10.


The base space motion is shown in Figure 8. As in the single -lor
-2
geometric phase path case, Step 0 of the control algorithm 4
0 10 15 20 25 30 35 40

was executed by driving first the base variable q2 and then 5'
:z 2
qi to zero. Clearly, during this step, the fiber variable r1
remains constant. Figures 11 and 12 show the results of 0
0 5 10 15 20
time [s
25 30 35 40

the experiment using the same controller parameters as in


the simulation. Again, it can be seen that the results of Fig. 11. Configuration variables ql, q2 and r1 (experiment).
the simulation and the experiment do agree. The torque
levels and the final values of the configuration variables r 1 1

are approximately the same. It is clear that the geometric


phase control technique can be used to perform rest-to-rest
,f
1-2
0.1
.I
maneuvers with a high accuracy. -0.1 H
-0.21
0 5 10 5 20 25 30 35 40

02y Y Y Y Y? i
-5
0

-0.2-

-0.4-

-0.6- Fig. 12. Input torques Ti and T2 (experiment).


-0.8-

-1 .2
-25 -20 -15 -10 -5
qt [rad]
0 5 10 15 20 V. CONCLUSIONS
In this paper, a restricted control moment gyroscope
Fig. 8. Motion in base space (simulation). (CMG) has been studied in the context of nonholonomic
control systems. The geometric phase technique has been
164
used to design a feedback algorithm for the rest-to-rest [10] I. Kolmanovsky, M. Reyhanoglu, and N.H. McClamroch,"Switched
control of the system. The effectiveness of the algorithm Mode Feedback Control Laws for Nonholonomic Systems in Extended
Power Form," Systems and Control Letters, vol. 27, no. 1, 1996, pp.
has been illustrated through both numerical simulations and 29-36.
experimental results. Our current research includes design of [11] H. Krishnan, M. Reyhanoglu, and N.H. McClamroch,"Attitude Stabi-
state feedback and output feedback tracking algorithms for lization of a Rigid Spacecraft Using two Control Torques: A Nonlinear
Control Approach Based on the Spacecraft Attitude Dynamics," Au-
the CMG unit. tomatica, vol. 30, no. 6, 1994, pp. 1023-1027.
[12] H. Krishnan, M. Reyhanoglu, and N.H. McClamroch,"Attitude Stabi-
VI. ACKNOWLEDGMENTS lization of a Rigid Spacecraft Using Two Momentum Wheel Actua-
M. Reyhanoglu wishes to acknowledge the support pro- tors," AIAA Journal of Guidance, Control and Dynamics, vol. 18, no.
2, 1995, pp. 256-263.
vided by Embry-Riddle Aeronautical University. J. van de [13] R.M. Murray and S.S. Sastry,"Nonholonomic Motion Planning: Steer-
Loo acknowledges the support provided by Eindhoven Uni- ing Using Sinusoids," IEEE Transactions on Automatic Control, vol.
versity of Technology. 38, no. 5, 1993, pp. 700-716.
[14] Y. Nakamura and R. Mukherjee,"Nonholonomic Motion Planning of
Space Robots via a Bi-directional Approach," IEEE Transactions on
REFERENCES Robotics and Automation, vol. 7, no. 4, 1991, pp. 500-514.
[1] B. Armstrong-Helouvry, P. Dupont, and C.C. de Wit,"A Survey of [15] H. Olsson, K.J. Astrom, C.C. de Wit, M. Gafvert, and P. Lischinsky,
Models, Analysis Tools and Compensation Methods for the Control "Friction Models and Friction Compensation," European Journal of
of Machines with Friction," Automatica, vol. 30, no. 7, 1994, pp.1083- Control, vol. 4, no. 3, 1998, pp.176-195.
1138. [16] T. R. Parks, Manual For Model 750 Control Moment Gyroscope,
[2] A. Astolfi,"Discontinuous Control of Nonholonomic Systems," Sys- Educational Control Products, Bell Canyon, CA; 1999.
tems and Control Letters, vol. 27, 1996, pp. 37-45. [17] M. Reyhanoglu and E. Al-Regib,"Nonholonomic Motion Planning for
[3] A. M. Bloch, Nonholonomic Mechanics and Control, Springer, New Wheeled Mobile Systems Using Geometric Phase," Proceedings of
York, NY; 2003. IEEE International Symposium on Intelligent Control, 1994, pp.135-
[4] A. M. Bloch, M. Reyhanoglu, and N. Harris McClamroch," Control 140.
and Stabilization of Nonholonomic Dynamic Systems," IEEE Trans- [18] M. Reyhanoglu and N.H. McClamroch,"Planar Reorientation Maneu-
actions on Automatic Control, vol. 37, no. 11, 1992, pp.1746-1757. vers of Space Multibody Spacecraft Using Internal Controls," AIAA
[5] P. Dahl,"A Solid Friction Model," The Aerospace Corporation, CA, Journal of Guidance, Control and Dynamics, vol. 15, no. 6, 1992, pp.
Tech. Rep. TOR-0158(3107-18), 1968. 1475-1480.
[6] S. Dubowsky and E. Papadopoulos,"The Kinematics, Dynamics and [19] O.J. Sordalen and 0. Egeland,"Exponential Stabilization of Nonholo-
Control of Free-flying and Free-floating Space Robotic Systems," nomic Chained Systems," IEEE Transactions on Automatic Control,
IEEE Transactions on Robotics and Automation, vol. 9, no. 5, 1993, vol. 40, no. 1, 1995, pp. 35-49.
pp. 531-543. [20] G.C. Walsh, D. Tillbnry, S. Sastry, R. Murray, and J.P. Laumond,
[7] J.M. Godhavn and 0. Egeland,"A Lyapunov Approach to Exponential "Stabilization of Trajectories for Systems with Nonholonomic Con-
Stabilization of Nonholonomic Systems in Power Form," IEEE Trans- straints," IEEE Transactions on Automatic Control, vol. 39, no. 1,
actions on Automatic Control, vol. 42, no. 7, 1997, pp. 1028-1032. 1994, pp. 216-222.
[8] Z.P. Jiang and H. Nijmeijer,"A Recursive Technique for Tracking Con- [21] G.C. Walsh and L.G. Bushnell,"Stabilization of Multiple Input
trol of Nonholonomic Systems in Chained Form," IEEE Transactions Chained Form Control Systems," Systems and Control Letters, vol.
on Automatic Control, vol. 44, no. 2, 1999, pp. 265-279. 25, 1995, pp. 227-234.
[9] Z.P. Jiang and H. Nijmeijer, "Tracking Control of Mobile Robots: A [22] G.C. Walsh and S.S. Sastry,"On Reorienting Linked Rigid Bodies Us-
Case Study in Backstepping,", Automatica, vol. 33, no. 7, 1997, pp. ing Internal Motions," IEEE Transactions on Robotics and Automation,
1393-1399. vol. 11, no. 1, 1995, pp. 139-145.

165

S-ar putea să vă placă și