Вы находитесь на странице: 1из 11

Automatica 37 (2001) 1941}1951

Hybrid control for global stabilization of the cart}pendulum system夽


Jun Zhao *, Mark W. Spong
School of Information Science and Engineering, Northeastern University, Shenyang, Liaoning 110006, People's Republic of China
Coordinated Science Laboratory, University of Illinois at Urbana-Champaign, 1308 W.Main street, Urbana, IL 61801, USA
Received 2 February 1999; revised 30 March 2001; received in "nal form 8 June 2001

Abstract

A globally stabilizing controller for the cart}pendulum system is designed in this paper. The design procedure consists of three
parts. First, "nd a neighborhood of the origin in which a locally stabilizing controller can be used. Second, construct a linear
controller for the cart subsystem and steer the position and velocity of the cart to the origin. The last part is to design several classes of
bang}bang controllers and a switching strategy which guarantee that the trajectory of the system starting from any initial condition
enters, in "nite times of switching, the neighborhood where the locally stabilizing controller is e!ective.  2001 Elsevier Science Ltd.
All rights reserved.

Keywords: Stabilization; Cart}pendulum system; Hybrid control; Switching; Bang}bang control

1. Introduction Though the dynamics of the cart}pendulum system


appears quite simple, many standard techniques of
The cart}pendulum system is one of the classic mech- nonlinear control (Isidori, 1995; Krstic, Kanelakopoulos,
anical systems that have been studied extensively in con- & Kokotovic, 1995; Marino & Tomei, 1995; Sepulchre,
trol area during the past few decades. In the early stage, Jankovic, & Kokotovic, 1997) are ine!ective. For
this system was used to understand control theory for example, the system is not feedback linearizable or even
educational purpose (see Feng, Yin, & Chen, 1987; has no constant relative degree corresponding to the
Furuta, Okutani, & Stone, 1978). Later on, it was studied swinging energy of the pendulum as an output. Also, the
to demonstrate the validity or usefulness of design tech- controllability distribution of the system does not have
niques and algorithms (Furuta, Yamakita, & Kobayashi constant rank. Many geometric properties of the system
1992; Geva & Sitte, 1993). Recently, particular attention are lost when the pendulum moves through horizontal
has been paid to the study of underactuated mechanical positions. Therefore, the study of the system is much
systems which are systems with fewer actuators than more di$cult than it seems to be.
degrees of freedom. As many underactuated robot sys- One of the most important issues in the study of the
tems have quite similar dynamics to that of the cart} cart}pendulum system is to stabilize the system. The
pendulum system, the cart}pendulum system has been most common approach is the so-called energy-based
studied as a simpli"ed model of underactuated robot method. There are many works in this direction. Spong
systems by Spong (1995, 1998), and Spong and Praly and Praly (1997) gave a stabilization result based on the
(1997). It possesses many interesting challenges from energy assignment. With this controller, the cart}pendu-
standpoint of global nonlinear control. lum system is stabilized for almost all initial states except
the stable manifold of the downward vertical position of
the pendulum. The design consists of three steps. First,
bring the pendulum in a neighborhood of the upward

This paper was not presented at any IFAC meeting. This paper was vertical position with zero angular speed. Then, use for-
recommended for publication in revised form by Associate Editor warding technique to bring the whole system in a neigh-
Heinz unbehauen under the direction of Editor Mituhiko Araki. borhood of the origin. Finally, apply a linear balancing
* Corresponding author. Tel.: #86-24-86815415; fax: #86-24-
23892454.
controller in the neighborhood. A nonlinear controller
E-mail addresses: zdongbo@pub.ln.cninfo.net (J. Zhao), spong@con- was proposed by Cheng and Hauser (1995) to regulate
trol.csl.uiuc.edu (M.W. Spong). the swinging the energy and the cart speed without

0005-1098/01/$ - see front matter  2001 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 5 - 1 0 9 8 ( 0 1 ) 0 0 1 6 4 - 9
1942 J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951

regard of the cart position. Furuta et al. (1978) put controller is presented in Section 3. In Section 4 we prove
forward an algorithm of swinging up the pendulum by that the designed hybrid controller indeed asymptotically
moving the cart back and forth and derived the maximal stabilizes the system. A modi"ed algorithm for hybrid
initial angle from which the pendulum can be raised and control is given in Section 5. Section 6 contains conclud-
balanced without "rst letting the pendulum fall to its ing remarks.
stable vertical con"guration. However, no stability for
the cart is considered in the result.
Since there is no continuous feedback of 2-period in 2. Problem statement
angle which can render the vertical upward position of
the pendulum globally asymptotically stable, many We consider the well-known cart}pendulum system
authors have considered discontinuous feedback. For shown in Fig. 1.
example, refer to Wei, Dayawansa, and Levine (1995), The dynamics are given by
Spong and Praly (1997).
On the other hand, there has been considerable inter- (M#m)xK #ml cos $ "mlQ  sin #F,
est in studying hybrid systems in recent years (see, for (1)
example, Branicky, Borkar, & Mitter, 1998; Brockett, xK cos #l$ "g sin ,
1993; Gollu & Varaiya, 1989; and Varaiya, 1993). As where M and x are mass and position of the cart which
simpli"ed models of hybrid systems, switched systems moves along a horizontal axis, m, l,  are mass, length and
were studied by Agrachev and Liberzon (1999), Branicky angular deviation from the upward vertical position of
(1998, 1994), Dayawansa and Martin (1999), Haspanha the bar, which can rotate around a point "xed on the
and Morse (1999), Hassibi and Boyd (1998), Liberzon cart, and F is a horizontal force applied to the cart.
and Morse (1999), Ooba and Funahashi (1997), Peleties Applying the partial feedback linearization technique
and DeCarlo (1991). Hybrid and switching control we can easily rewrite the cart}pendulum system as
has proved powerful for solving global stabilization
problem (De Dona, Reza Moheimani, Goodwin, x "v, v "u, Q "w, w "sin !u cos . (2)
& Feuer, 1999; Frommer, Kulkarni, & Ramadge, 1998;
The total energy of the pendulum is given by
Iftar & Ozguner, 1998; Iglesias, 1994; Kolmanovsky
& McClamroch, 1996; Morse, 1996; Passino, Michel, E(t)"w(t)#cos (t).
& Antsaklis, 1994; Ska"das & Evans, 1999 and Ye, 
Michel, & Hou, 1998). The control goal is to make the origin globally asymp-
In this paper, we focus on the problem of global totically stabilizable.
stabilization of the cart}pendulum system. The main
di$culty is to swing up the pendulum from the down-
ward vertical position and keep the cart stable as well. 3. Hybrid controller
A hybrid controller is given, which globally asymp-
totically stabilizes the system for all initial conditions. The design of the controller consists of three stages.
The design consists of three parts. First, "nd a neighbor- First of all, design a local controller u which locally

hood of the origin in which a locally stabilizing controller asymptotically stabilizes system (2). There are many choi-
can be used. Second, construct a linear controller for the ces of such controllers. Some are given, for example, in
cart-subsystem and steer the position and velocity of the the work by Spong (1995) and Spong and Praly (1997).
cart to the origin. The last part is to design several classes Since the linearized system is controllable, system (2) can
of bang}bang controllers and a switching strategy given be locally stabilized by a linear controller which is a natu-
by an algorithm, which guarantee that the trajectory of ral choice of locally stabilizing controllers.
the system starting from any initial condition enters, in After choosing the local controller u , we determine

"nite times of switching and in "nite time, the neighbor- a local domain of attraction for the closed-loop system
hood where the locally stabilizing controller is e!ective.
Compared with the existing results, this result has two
features. First, we give a global asymptotic stabilization
result. To the best of our knowledge, no global asymp-
totic stabilization results for the cart}pendulum system
have been reported before now. Second, before switching
to a locally stabilizing controller, we switch between
bang}bang controllers. Therefore it is practically easy to
realize the control.
The paper is organized as follows. Section 2 gives the
problem statement. The design procedure of the hybrid Fig. 1. The cart}pendulum system.
J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951 1943

(2) with controller u . Suppose that the domain of attrac- yields



tion is given by
 
 
<Q (t))!y#2y P y#K y
;"(x, v, , w)2  x)a,v)b,)c,w)d (3) 6 2

   
1
with positive constants a, b, c, d, c(/2. "y P #K !1 . (10)
Then, design a controller u for the cart subsystem 3

If we choose a positive constant M satisfying
x "v, v "u, (4)
M (P\(#K)\,
such that the states x and v are steered to the origin in 
"nite time. This can be easily realized because subsystem then on the region
(4) is controllable. ; "y"(x, v, , w)2  )M , yO0
The third stage is to steer any state of the whole system 
(2) to the neighborhood ; so that u can be applied. This <Q (0 holds. Denote the minimal eigenvalue and the
maximal eigenvalue of P by  and  , respectively,
can be realized by a hybrid controller we will give later by  
an algorithm. and let r"M ( / . We claim that the region
 
;"y"(x, v, , w)2  x)a, v)b, )c, w)d
Remark 3.1. Let y"(x, v, , w)2. The linearized system of
system (2) around the origin is expressed as with positive constants a, b, c, and d satisfying
a#b#c#d)r is a domain of attraction for sys-

 
0 1 0 0 0 tem (2) with u .

0 0 0 0 1 In fact, for any y 3;, the contour line
y " y# u"Ay#Bu. (5) 
C  "y  y2Py"y2 Py  of < at y is entirely in ; .
0 0 0 1 0 W    
This is because for any y3C 
0 0 1 0 !1 W
 y)y2Py"y2 Py ) y ) r (11)
     
The linearized system (5) is obviously controllable. holds and thus
Therefore, it can be stabilized by linear controllers. As

well known, any linear controller u "Ky which asymp-
y)  r"M , (12)

totically stabilizes system (5) also locally stabilizes system 
(2). If we choose the controller u "Ky as a locally which implies that )M . By the standard argument of

stabilizing controller for system (2), a local domain of Lyapunov stability we conclude that for any initial point
attraction ; for the closed-loop system (2) with u "Ky in ;, the trajectory of the closed-loop system (2) with

can be determined in the following way. u converges to the origin.
Since A#BK is stable, the Lyapunov equation
Before giving the algorithm, we introduce some
notations.
(A#BK)2P#P(A#BK)"!I (6)
Let h, r be positive constants satisfying 0)h(r)2
has a positive de"nite solution P. and t be the initial time. If (t )"h, de"ne
 


Now, rewrite the closed-loop system (2) with u "Ky (t ) if w(t )"0, h((t ))r
  
as
for some t 't , and w(t)'0
minr, 'h"  
y "(A#BK)y#(0, 0, 0, sin !#Ky(1!cos ))2. (7) U for t3(t t ),
 
Choose <"y2Py. The derivative of < along the traject- r otherwise.
ory of system (7) is If (t )"r, de"ne



<Q (t)"y2((A#BK)2P#P(A#BK))y (t ) if w(t )"0, h)(t )(r
  
for some t 't , and w(t)(0
#2y2P(0, 0, 0, sin !#Ky(1!cos ))2 maxh, (r"  
U for t3(t t ),
 
)!y#2y P(sin ! h otherwise.
#K y(1!cos )). (8)
Algorithm. Step 0: Apply u to steer x and v to zero.

Applying the inequalities Step 1: Given initial states x "0, v "0,  "(t )
   
and w "w(t ), choose ¹ '0, ¹ '0 and ¹ '0,
      
where ¹ is an upper bound of the time in which the
sin !) , 1!cos ) (9) 
6 2 pendulum moves from " to H"min,  '
 U
1944 J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951

with w(t)*0 under all possible nonnegative constant and apply controller u"C until

control u satisfying u'0 if w(t )"0, ¹ is an upper


   if w '0,
bound of the time in which the pendulum moves from "   (17)
" to H" with E "w #cos  'd#1 and  (or equivalently!) if w (0,
        
w(t)*0 under all possible nonpositive constant controls
u satisfying (cos c!E(0)))u)0, ¹ is an upper then switch the control to u"!C . When v"0, set
  
bound of the time in which the pendulum moves from current states as initial states and then return to Step 2.
"0 to H" with E "w #cos  'd#1 and
      Remark 3.2. In Step 0 we use u to steer x and v to zero,
w(t)*0 under all possible nonnegative constant controls 
u satisfying 0)u)mind, d. The existence of such which will be needed for the stability analysis in Section
  4 (Lemma 4.4). According to di!erent initial energy the
upper bounds will be shown in Lemma 4.1 and possible
choices will be given there. algorithm works in the following way:
Step 2: Calculate the energy E "w #cos  .
    (1) If cos c4E )d#1, then apply zero control
If cos c)E )d#1, set u"0 until (x, v, , w)2 en-  
  until the trajectory enters the domain of attraction ; and
ters ;, then switch to u . Otherwise, if E (cos c, go to
 then apply local controller u .
step 3, and if E 'd#1, go to step 4.
  (2) If E (cos c, then Step 3 is taken. We "rst use u"0
Step 3: If  O or  ", x w '0, set u"0 until 
    until " and xw)0. This is always possible because
" and xw)0, and then set current states as initial
states and restart Step 3. Otherwise calculate E (cos c. Then, we split the condition " and

xw)0 into two cases:

  
a 1 1 Case 1:  ", x )0, w *0.
C "min , 1# d!E . (13)   
 ¹ 3 2  Case 2:  ", x "0 and w (0 hold, or
   
 ", x '0 and w )0 hold, and then apply di!erent
  
Case 1:  ", x )0 and w *0. controllers to the system.
  
Apply u"C until "H"min,  ', and (3) If E 'd#1, then Step 4 is taken. w is obvious-
  
  U ly not zero. We design the controller in two di!erent
then switch the control to u"!C . When v"0, set
 cases.
current states as initial states and then return to Step 2.
Case 2:  ", x "0 and w (0 hold, or Case 1: w x *0.
 
  
 ", x '0 and w )0 hold. Case 2: w x (0. We "rst set control u to zero until
 
   " (in Case 1) or "0 (in Case 2). This is also always
Apply u"!C until "H"max,  (,
  U possible because E 'd#1. Then, apply di!erent
and then switch the control to u"C . When v"0, set  
 controllers to the system.
current states as initial states and then return to Step 2.
Step 4: Case 1: w x *0. If O, set u"0 until ", It is worth while to point out that in the algorithm,
  once the trajectory enters the domain of attraction ;, u
and set current states as initial states and restart Step 4.
Otherwise calculate will be used forever, no matter the trajectory exits ; or
not.
 
a E !cos c


!min,  if w '0, In each iteration of the algorithm, the control is "rst
¹ 3 
C "  (14) set to some constant value C. While  reaches a speci"c

 
a E !cos c value, the control is then switched to !C. When
min ,  if w (0
¹ 3  v reaches 0, the iteration ends. Therefore, the control is of
 the bang}bang form and at the end of each iteration,
and apply controller u"C until
 v must be zero.
 if w '0,
" 
 
 if w (0,
 
(15)
4. Global stabilization
then switch the control to u"!C . When v"0, set
 In this section we show that the hybrid controller given
current states as initial states and then return to Step 2.
in Section 3 indeed globally stabilizes system (2). We
Case 2: w x (0. If O0, set u"0 until "0, and
  begin with the analysis of the validity of the controller.
set current states as initial states and restart Step 4.
Otherwise calculate
Lemma 4.1. (a) Suppose the pendulum starts from (0)"

  
a d d E !cos c with w(0)*0. Then the time ¹"¹(w(0), u) in which
min , , ,  if w '0, the pendulum rotates from " to "H"
¹ 2 2 3 
C "  (16) min,  ' with w(t)*0 is uniformly bounded
  U
 
a d d E !cos c from above for all nonnegative constant controls. One such
!min , , ,  if w (0
¹ 2 2 3  a bound is 4(5.

J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951 1945

(b) Suppose E(0)'d#1. If the pendulum starts from for all w(0)*2 and u*0. It is obvious that

(0)" with w(0)'0, then the time ¹"¹(w(0), u) in w(¹(2,0))"0. Therefore, we only have to estimate
which the pendulum rotates from " to H" with


w(t)*0 is uniformly bounded from above for all nonposi- ? d
¹(w(0), u)" (23)
tive constant controls u satisfying (cos c!E(0)))u)0. L (w(0)!2u sin !2(1#cos )

One such a bound is L(E(0)#cos c)\.
  with w(0))2, w(¹)"w "0,( ) and u*0.
(c) Suppose E(0)'d#1. If the pendulum starts from
 F? 
(0)"0 with w(0)'0, then the time ¹"¹(w(0), u) Now, "x u and w(0) and let
in which the pendulum rotates from "0 to H" I()"w(0)!2u sin !2cos !2.
 (24)
with w(t)*0 is uniformly bounded from above for all
nonnegative constant controls u satisfying 0)u) Substituting t "0, t"¹," and w(¹)"0 into (19)

mind, d. One such a bound is (2/2d. we know I( )"0.
  Since
Proof. First of all, we derive a formula for calculating the
dI
time in which the pendulum rotates from (t ) to (t ) "2u sin #2 cos )0, (25)
  d
with w(t) without changing sign and under constant con-
trol u. According to system (2) we have I() is a convex function with respect to 3[, ]. There-
1 d fore I()*y() given by
ww " (w)"Q (sin !u cos ).


2 dt
w(0)#2((1#u!1)
Integrating it from t to t results in !

; , ))arctan(u)#,
1 arctan(u)
(w(t)!w(t ))
2  y()" (26)
(w(0)#2((1#u!1))
"u(sin (t )!sin (t))#cos (t )!cos (t). (18)
  !arctan(u)!
Let ;( !), arctan(u)#)) .
(, u) Hence,
"(w(t )#2u(sin (t )!sin )#2(cos (t )!cos ).
 
   ? d  S>L
¹(w(0), u)) "
Then, we have L (y() L
d d
dt
"w(t)"$((t), u). (19) ;

Thus, w(0)#2((1#u!1)(!)/arctan(u)



FR  d

if w(t)*0, ?
FR  (, u)

R #
t !t "
 
dt" (20)  S>L
R d FR 
! if w(t))0. d
FR  (, u) ;
(a) Let t "0,(t )", t "¹"¹(w(0), u). Then,
  
((w(0)#2((1#u!1))( !)/( !arctan(u)!)
(t )"(¹)"H"min,  '. Noticing
  U (27)
w(t)*0, from (20) we have
2 arctan (u)

FH d "
¹(w(0), u)" . (21)
L (w(0)!2u sin !2(1#cos ) w(0)#2(1#u!2#w(0)
A simple calculation shows that
!arctan(u)!
EQ "!wu cos , (22) #2

which indicates that the energy E of the pendulum does w(0)#2(1#u!2


not decrease if u*0, w(t)*0 and ))H. Therefore,
if w(0)*2, H" must hold for all nonnegative con- !
 )2 .
w(0)#2((1#u!1)
stant control u. In this case, since ¹(w(0), u) decreases as
either w(0) or u increases, ¹(w(0), u))¹(2, u))¹(2,0)
1946 J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951

To make further estimate, we express in terms of w(0) 2 2


and u. It follows from (22) that ) (4#2u# (u#5
(u#1) u#1


R
E(t)!E(t )" EQ dt"u(sin (t )!sin (t)). (28) 4
  ) (u#5)4(5. (33)
R (u#1)
Recalling I( )"0 and noticing (24) we have
(b) From E(0)*d#1 and (28) we know that the

w(0)!2u sin !2"2 cos . (29) pendulum can indeed reach H" under all non-

positive constant controls u satisfying (cos c!E(0)))
Substituting
"sin into (29) and then squaring it give 
u)0. Therefore,
rise to
w(0)!2u sin !2 cos !2
(4u#4)
#4(2!w)u
#w!4w"0. (30)
Since
(0, the solution to Eq. (30) is "2E(0)!2u sin !2 cos 

(w(0)!2)u!(4u!w(0)#4w(0) 4 2

" (31) * E(0)# cos c!2 cos . (34)


2(u#1) 3 3

and thus "!arcsin


. One can easily show that Note that cos )0, the conclusion then follows by ap-
arcsin x)2x for all x3[0,1]. From (27) we have plying (21) with H".

(c) Since 0)u)mind, d, we know from (28) that
 
2arcsin(!
) the pendulum can indeed reach ". Now, for "xed u,
¹(w(0), u)) 
the minimum of the function w(0)!2u sin #2(1!cos )
w(0)#2((1#u!1) restricted to 3[0, ] occurs at M "arctan(u), and the

minimum satis"es
4

)!
w(0)!2u sin M #2(1!cos M )
 w(0)#2((1#u!1)
"w(0)#2(1!sec M )"w(0)#2(1!(1#u)
(2!w(0))u#(4u!w(0)#4w(0)
"2 . 1
*d!2u* d. (35)

(u#1) w(0)#2((1#u!1) 2

(32) The conclusion then follows from (20). 䊐

Applying the inequalities (1#u!1*u/(2#u) and Remark 4.1. The same bounds can be given, respectively,
u#5*4#2u we have for the following cases.
¹(w(0), u)
(a) The pendulum rotates from " to H"
maxL,  ( with w(t))0 and under all possible
(2!w(0))u#(4u!w(0)#4w(0)  U
)2 nonpositive controls.
(u#1)(w(0)#2u/(2#u) (b) The pendulum rotates from " to H"L with

E(0)'d#1, w(t)(0 and under all nonnegative con-
(2!w(0))u 
stant controls u satisfying 0)u)(E(0)!cos c).
)2 ((2#u)/2u 
(u#1) (c) The pendulum rotates from "0 to H" (or

equivalently !) with E(0)'d#1, w(t)(0 and un-
  
2 4u!w(0)#4w(0)
# der all nonpositive constant controls u satisfying
u#1 w(0)#2u/(2#u)
!mind, d)u)0.
 


2(4#2u 2 4u!w(0)#4w(0) Lemma 4.2. After each iteration of Step 3 or 4, the change
) #
(u#1) u#1 w(0)#4u/(u#5) of the energy satisxes  E)3u.

2 Proof. It is a straightforward corollary of (28). 䊐


) (4#2u
(u#1)


2 (u#5)(4u!w(0)#4w(0)) Remark 4.2. By Lemma 4.2 and taking (13), (14) and (16)
#
u#1 uw(0)#5w(0)#4u into account, we know that the algorithm guarantees
J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951 1947

that t"¹ and lasts until v"0. One can easily calculate that
v(2¹)"0 and
1

  
E(t)) d#1 if E(0)(cos c, and E(t)*cos c if 2 2 R
2 x(2¹)"x(0)# v(t) dt"x(0)# u () d dt
1   
E(0)' d#1.
2
"x(0)#C ¹, (37)

Lemma 4.3. After each iteration of Step 3 or 4, the energy where
E increases at least by u sin H in the case E(0)(cos c, or


decreases at least by u in the case E(0)'d#1. C if 0)t)¹,
 u " 
!C if ¹(t)2¹.

Proof. We only give the proof in the case E(0)(cos c.
Noticing !a)x(0))0 and (13) we know
The proof in the case E(0)'d#1 is similar.
 x(2¹))a. 䊐

Without loss of generality, we suppose x(0))0. Also Lemma 4.5. If E(0)(cos c, then there is a constant q'0,
suppose that the pendulum moves from " at t"0 such that sin H*q for all H in each iteration.
with w*0 and u'0, and at t"¹ reaches "H"
min,  '. According to the algorithm, control
 U Proof. It is obvious that sin H increases with the
is switched to !C at t"¹ and lasts until t"2¹ when
 iteration because the energy increases according to
v reaches zero. We now show that during the time inter-
val [0,2¹] the energy increases at least by usin H. To do Lemma 4.3.
this, it is su$cient to show that E(2¹) is no less than E(¹)
because E(¹)!E(0)"usinH follows from (28). At the beginning of any iteration, we know from (2)
If (¹)"H", (28) gives that that w (0)"u'0. The existence of q is then evident.

Now, we give the main theorem. 䊐
E(2¹)!E(¹)"!u(sin (¹)!sin (2¹))*0. (36)
Theorem 4.1. System (2) is globally asymptotically stabil-
If (¹)"H(, w(¹)"0 must hold. Let "!H,
  ized by the hybrid controller given in Section 3.
then 0( (. We claim that (2¹) must satisfy

0)(2¹))H or # )(2¹))2 (equivalently
 Proof. Global attractiveness follows from Lemmas 4.1}
!)(2¹))0), which still guarantees (36).
 4.5. More precisely, the trajectory of system (2) starting
Now suppose that the claim is false, that is, (2¹),
[0, H] [# , 2]. For any tM 3[¹, 2¹], if (tM )"(¹)"H, from any initial point enters, in "nite switches and "nite
 time, the neighborhood ;, and then is steered to the
(28) gives E(tM )"E(¹), which implies w(tM )"w(¹)"0.
Since w (t)"sin (t)!(!C ) cos (t)(0, whenever origin by local controller u . Local asymptotic stabiliz-

3[, H], we know that the pendulum cannot reach ability is guaranteed by local controller u . 䊐

(2¹) upward from (¹). Therefore, the pendulum must
arrive at (2¹) through "0. In particular, there exists
t 3(¹, 2¹) such that (t )"# . Applying (28) once 5. A modi5ed algorithm
  
again we get E(t )!E(¹)"!u(sin (¹)!sin (t ))"0.
 
However, E(t )"w(t )#cos(# )*cos(# )'0. In the algorithm given in Section 3, the control is set to
    
This is a contradiction. 䊐 zero between adjacent bang}bang controllers. Energy
remains unchanged during these periods. This is a kind of
`wastea of time and as a result the algorithm may con-
Lemma 4.4. At the end of each iteration of Step 3 or Step 4,
verge slowly. To overcome this drawback we modify the
x)a and v"0 must hold.
algorithm as follows.
Step 0: Apply u to steer x and v to zero.

Proof. At the end of Step 0, x and v have already been set Step 1: Given initial states x "0, v "0,  "(t )
   
to zero. Suppose that at the beginning of an iteration and w "w(t ). Choose '0, ¹ '0, ¹ '0 and
   
x(0))a and v(0)"0 hold, we now show that at the end ¹ '0 (see Remark 5.1 below).

of the iteration x)a and v"0 still hold. Without loss Step 2: Calculate the energy E "w #cos  .
   
of generality, suppose that Step 3 with x(0))0 is carried If cos c)E )d#1, set u"0 until (x, v, , w)2 en-
 
out. Let ¹ denote the time when  reaches H. According ters ;, and then switch to u . Otherwise if E (cos c, go

to the algorithm, the control is switched to !C at to step 3. If E 'd#1, go to Step 4.
  
1948 J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951

Step 3: Case 1: One of the following conditions, Calculate


(38)}(41), is satis"ed:

  
a#x 1 1
min  , 1# d!E
 3 ¹ 4 2 
w '0, ) ( . (38) 
 2  2 if (44) or (45) holds
CM " (48)
 
  a#x 1!E
w "0, ) ). (39) min , 
 2  ¹

4
if (46) or (47) holds.
3
w )0, ( )2!c. (40)
 2 
Apply controller



w (0,

c( ) .
 2
(41) !CM if CM *,
u"C "   (49)
 0 if CM (,

Calculate
until


 

(a!x ) 1 max, 
min  , (1#d!E )  U
( if (44) or (45) holds,
¹ 4  
 "H" min2!c,  ' if (46) holds,
U 
if (38) or (39) holds,  if (47) holds.
CM " (42) 

 
 (a!x ) 1!E
min  , 
¹

4 If CM (, set current states as initial states and restart

Step 3. If CM *, switch the control to u"!C . When
if (40) or (41) holds.  
v"0, set current states as initial states and then return to
Step 2.
Apply the controller Step 4: Case 1:
 3

CM if CM *,
u"C "   (43) w (0,
 2
( ) .
 2
(50)
 0 if CM (

Calculate
until

 
a!x E !cos c
CM "min , 


. (51)
min,  ' if (38) or (39) holds,  4
 U ¹

"H"  if (40) holds,
 Apply controller
maxc,  ( if (41) holds.
U 

CM if CM *,
u"C "   (52)
If CM (, set current states as initial states and restart  0 if CM (
 
Step 3. If CM *, switch the control to u"!C . When
  until "/2. If CM (, set current states as initial states
v"0, set current states as initial states and then return to 
and restart Step 4. If CM *, switch the control to
Step 2. 
u"!C . When v"0, set current states as initial states
Case 2: One of the following conditions, (44)}(47), is 
satis"ed. and then return to Step 2.
Case 2:
 3  3
w (0, ( ) . (44)
 2  2 w '0,

) ( .

(53)
2 2
3 Calculate
w "0, ( ) . (45)
  2
 
a#x E !cos c
CM "min ,  . (54)
 ¹ 4
w '0,
3
) . (46) 
 2  Apply controller


 !CM if CM *,
w *0,  ( . (47) u"C "   (55)
  2  0 if CM (,

J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951 1949

until "3/2. If CM (, set current states as the new



initial states and restart Step 4. If CM *, switch the

control to u"!C . When v"0, set current states as

initial states and then return to Step 2.
Case 3: Either


w '0,

(56)
3 
) (2 or 0) (
2   2

or


w (0,

(57)
3 
( )2 or 0) ) Fig. 2.
2   2
holds.
If  "0, calculate


 
a!x d d E !cos c
CM "min , , ,  (58)
 ¹ 2 2 3

and

 
a#x d d E !cos c
CM "min , , ,  , (59)
 ¹ 2 2 3

where ¹ is given in Section 3. Apply controller



CM if w '0 and CM *,
  
u"C " !CM if w (0 and CM *, (60)
   
Fig. 3.
0 otherwise

until
initial energy E (cos c, w(t))0, c( ) and un-
   

if w '0,
"   (61) der all possible nonnegative constant control u satisfying
 (or equivalently!) if w (0. 0)u)(1!E )/4,  is chosen as
   

 
a 1!E 
Then, if u"0, set current states as initial states and ) min ,  .
restart Step 4. Otherwise, switch the control to ¹ 4
WGW G
u"!C . When v"0, set current states as initial states
 Remark 5.2. The algorithm works in the same way as the
and then return to Step 2.
If  O0, set u"0 until "0 (or equivalently, 2) or algorithm in Section 3 and similar discussion can be
 made.
" or ", whichever holds "rst, and then set
 
current states as initial states and restart Step 4.
Simulation results are presented in Fig. 2 and Fig. 3 by
using Matlab. Fig. 2 shows the evolution of the state
Remark 5.1. In the algorithm ¹ is an upper bound of trajectory starting from x"0, v"0," and w"0

the time in which the pendulum moves from  to which is the most di$cult con"guration to swing up the

H"min, ' with w(t)*0 for all ) ( pendulum and stabilize the system. Fig. 3 gives the con-
 U  
 and under all possible nonnegative constant controls, trol switching.

¹ is an upper bound of the time in which the pendulum

moves from  to H" with the initial energy
 
E 'd#1, w(t)*0 for all ) ( under all 6. Concluding remarks
    
possible nonpositive constant controls u*(cos c!
E )/4, ¹ is an upper bound of the time in which the A global stabilization controller for the cart}pendulum
 
pendulum moves from  to H"maxc,  with the system has been designed in the paper. It is practically
 U
1950 J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951

easy to realize the controller because we only switch Feng, Z., Yin, Z., & Chen, H. (1987). Microprocessor-based controller
between bang}bang controllers before the trajectory en- for double inverted pendulum. Proceedings of IFAC 10th world
ters the neighborhood of the origin, where a locally congress (pp. 237}240).
Frommer, J. F., Kulkarni, S. R., & Ramadge, P. J. (1998). Controller
balancing controller is e!ective. switching based on output prediction errors. IEEE Transactions on
Although the method is based on the energy, we Automatic Control, 43, 596}607.
do not limit ourselves to keeping the energy increasing Furuta, K., Yamakita, M., & Kobayashi, S. (1992). Swing-up
(in the case of E (cos c) or decreasing (in the control of inverted pendulum using pseudo-state feedback. Proceed-
 ings of Institute of Mechanical Engineers, Vol. 206 (pp. 263}269).
case of E 'd#1) all the time during an iteration.
  Furuta, K., Okutani, T., & Stone, H. (1978). Computer control of
However, the controller is designed in such a way a double inverted pendulum. Computers in Electronic Engineering, 5,
that at the end of each iteration the energy increases 67}84.
(in the case of E (cos c) or decreases (in the case Geva, S., & Sitte, J. (1993). A Cart-Pole experiment benchmark for
 trainable controllers. IEEE Control Systems, 13(5), 40}51.
of E 'd#1) comparing with the initial energy
  Gollu, A., & Varaiya, P. P. (1989). Hybrid Dynamical Systems. Pro-
of the iteration. This gives us a big `freedoma of
ceedings of the 28th IEEE conference decision and control
choices of controls, which allows us to steer the (pp. 2708}2712).
position and velocity of the cart to ; at the end of each Haspanha, J. P., & Morse, A. S. (1999). Stability of switched systems
iteration. with average dwell-time. Proceedings of 38th Conference on decision
It may be possible to extend the method to other and control (pp. 2655}2660).
mechanical systems, such as underactuated robot Hassibi, A., & Boyd, S. (1998). Quadratic stabilization and control of
piecewise-linear systems. Proceedings of the American Control
systems. Conference (pp. 3659}3664).
Iftar, A., & Ozguner, U. (1998). Overlapping decompositions, expan-
sions, contractions, and stability of hybrid systems. IEEE Transac-
tions on Automatic Control, 43(8), 1040}1055.
Acknowledgements Iglesias, P. A. (1994). On the stability of sampled-data linear
time-varying feedback systems. Proceedings of 33rd conference on
The authors would like to thank the anonymous re- decision and control (pp. 219}224).
Isidori, A. (1995). Nonlinear control systems. Berlin: Springer.
viewers for their valuable comments. This work was Kolmanovsky, I., & McClamroch, N. H. (1996). Hybrid feedback laws
supported in part by the National Key Project of China, for a class of cascade nonlinear control systems. IEEE Transactions
the National Science Foundation of China under Grant on Automatic Control, 41, 1271}1282.
79970114, the Key Teacher Foundation of China, and Krstic, M., Kanelakopoulos, I., & Kokotovic, P. V. (1995). Nonlinear
the National Science Foundation of the USA under and adaptive control design. New York: Wiley.
Liberzon, D., & Morse, A. S. (1999). Basic problems in stability and
Grants CMS-9712170 and ECS-9812591. design of switched systems. IEEE Control Systems Magazine, 37(3),
117}122.
Marino, R., & Tomei, P. (1995). Nonlinear control design. Englewood
Cli!s, NJ: Prentice Hall.
References Morse, A. S. (1996). Supervisory control of families of linear set-point
controller-part 1: Exact matching. IEEE Transactions on Automatic
Agrachev, A. A., & Liberzon, D. (1999). Lie-algebraic conditions Control, 41, 1413}1431.
for exponential stability of switched systems. Proceedings of 38th Ooba, T., & Funahashi, Y. (1997). On a common quadratic Lyapunov
IEEE conference on decision and control. USA. (pp. 2679}2684). function for widely distant systems. IEEE Transactions on Automatic
Branicky, M. S. (1998). Multiple Lyapunov functions and other analysis Control, 42, 1697}1699.
tools for switched and hybrid systems. IEEE Transactions on Auto- Passino, K. M., Michel, A. N., & Antsaklis, P. J. (1994). Lyapunov
matic Control, 43, 475}482. stability of a class of discrete event systems. IEEE Transactions on
Branicky, M. S., Borkar, V. S., & Mitter, S. K. (1998). A uni"ed Automatic Control, 39, 269}279.
framework for hybrid control: Model and optimal control theory. Peleties, P., & DeCarlo, R. (1991). Asymptotic stability of m-switched
IEEE Transactions on Automatic Control, 43, 31}45. systems using Lyapunov-like functions. Proceedings of the American
Branicky, M. S. (1994). Stability of switched and hybrid systems. Control Conference (pp. 1679}1684).
Proceedings of 33rd IEEE Conference on decision and control Sepulchre, R., Jankovic, M., & Kokotovic, P. V. (1997). Constructive
(pp. 3498}3503). nonlinear control. Berlin: Springer.
Brockett, R. W. (1993). Hybrid models for motion control systems. In Ska"das, E., & Evans, R. J. (1999). Savkin, A.V. & Peterson, I.R.
H. L. Trentelman., & J. C. Willems (Eds.), Essays on control prospec- Stability results for switched controller systems. Automatica, 35,
tives in the theory and its applications (pp. 29}53). Boston, MA: 553}564.
Birkhauser. Spong, M. W. (1995). The swing up control problem for the acrobot.
Cheng, C. C., & Hauser, J. (1995). Nonlinear control of a switching IEEE Control Systems Magazine, 15, 49}55.
pendulum. Automatica, 31(6), 851}862. Spong, M. W. (1998). Underactuated Mechanical Systems. In B.
Dayawansa, W. P., & Martin, C. F. (1999). A converse Lyapunov Siciliana., & K. P. Valavanis (Eds.), Control problems in robotics
theorem for a class of dynamical systems with undergo switching. and automation (pp. 135}150). Berlin: Springer.
IEEE Transactions on Automatic Control, 44, 751}760. Spong, M. W., & Praly, L. (1997). Control of underactuated mechanical
De Dona, J. A., Reza Moheimani, S. O., Goodwin,G. C., & Feuer, A. systems using switching and saturation. In A. Stephen Morse (Ed.),
(1999). Allowing for over-saturation in robust switching control of Control using logic-based switching (pp. 162}172). London: Springer.
a class of uncertain systems. Proceedings of IEEE Conference on Varaiya, P. P. (1993). Smart cars on smart roads: problems of control.
Decision and control (pp. 3053}3058). IEEE Transactions on Automatic, 38, 195}207.
J. Zhao, M.W. Spong / Automatica 37 (2001) 1941}1951 1951

Wei, Q., Dayawansa, W. P., & Levine, W. S. (1995). Nonlinear control- Mark W. Spong received the D.Sc. degree in
ler for an inverted pendulum having restricted travel. Automatica, systems science and mathematics in 1981 from
31(6), 841}850. Washington University in St. Louis. Since
Ye, H., Michel, N., & Hou, L. (1998). Stability theory for hybrid 1984 he has been at the University of Illinois
at Urbana-Champaign, where he is currently
dynamical systems. IEEE Transactions on Automatic Control, 43,
Professor of General Engineering and Re-
461}474. search Professor in the Coordinated Science
Laboratory. He has held visiting positions at
the University of Waterloo, Canada, the CIN-
VESTAV del IPN, Mexico City, The Lund
Dr. Jun Zhao was born in 1957. He received Institute of Technology, Sweden, The Labora-
the B.S. and M.S. degrees in mathematics in toire d'Automatique de Grenoble, France, The
1982 and 1984, respectively, both from Liaon- Universite de Tecnologie de Compiegne, France, the Katholiek Univer-
ing University, China. He completed his Ph.D. sitet, Leuven, Belgium, The National University of Singapore, and the
in Control Theory and Applications in 1991 at Technical University of Munich. Dr. Spong's main research interests
Northeastern University, China. From 1992 are in robotics, mechatronics, and nonlinear control theory. He has
to 1993 he was a postdoctoral fellow at the published over 150 technical articles in control and robotics and is
same University. Since 1994 he has been with co-author of two books. In the past he was Editor-in-Chief of the IEEE
School of Information Science and Engineer- Transactions on Control Systems Technology and is currently Vice
ing, Northeastern University, China, where he President for Publication Activities of the IEEE Control Systems So-
is currently a professor. From February 1998 ciety and a member of its Board of Governors. He is a Fellow of the
to February 1999, he was a visiting scholar at IEEE and a member of Phi Beta Kappa. He received the Senior
the Coordinated Science Laboratory, University of Illinois at Urbana- Scientist Research Award from the Alexander von Humboldt Founda-
Champaign. tion, Bonn, Germany, in 1999 and the IEEE 3rd Millennium Medal in
His main research interests include hybrid systems, nonlinear sys- 2000. In addition, he is President of Mechatronic Systems, Inc., a com-
tems, geometric control theory, and robust control. pany that he founded in 1996.

Вам также может понравиться