IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 11, NO.

6, DECEMBER 1995 807

A Multisteering Trailer System: Conversion into

Chained Form Using Dynamic Feedback
Dawn Tilbury, Member, IEEE, Ole Jakob S@rdalen,Member, IEEE, Linda Bushnell, Member, IEEE,
and S . Shankar Sastry, Fellow, IEEE

Abstract-This paper examines the kinematic model of an for baggage handling [SI. In all of these applications it is felt
autonomous mobile robot system consistingof a chain of steerable that a multisteering system will be more maneuverable than a
cars and passive trailers, liked together with rigid bars. The single-steering system.
state space and kinematic equations of the system are defined,
and it is shown how these kinematic equations may be converted Such systems are modeled as having one constraint on each
into a multiinput chained form. The advantages of the chained axle: that the wheels are allowed to roll but not to slip. These
form are that many methods are available for the open-loop nonslipping constraints are nonholonomic, or nonintegrable,
steering of such systems as well as for point-stabilization;some and do not reduce the reachable configuration space of the
of these methods are discussed here. Dynamic state feedback is
mobile robot. The existence of differentials in systems of this
used to convert the system to this multiinput chained form. It is
shown how the dynamic state feedback that is used in this paper kind results in the two wheels of a single axle moving through
corresponds to adding, in front of the steerable cars, a chain of different amounts in the course of a turn [l]; this has not been
virtual axles which diverges from the original chain of trailers. explicitly taken into account in this work.
Two different example systems are also presented, along with The present system appears at first glance to be a straightfor-
simulation results for a parallel-parkingmaneuver.
ward extension of the systems considered in previous work [4],
[16], [18], [19], but the main motivation in writing this paper is
I. INTRODUCTION to show how much richer and more complex the current system

I N this paper, the motion planning problem for a car-like is in its structure. As before, the kinematic equations will be
mobile robot pulling a combination of n passive trailers and converted into a chained form. The transformation, however,
m - 1 car-like robots is considered and solved. The controls will require dynamic state feedback. Motivated by the physical
available to the system are the velocity (throttle) of the lead structure of the constraints involved in this particular problem,
car and the steering velocities of all m car-like robots. The the dynamic state feedback that is used consists of adding
system is referred to as a multisteering n-trailer system. It can virtual axles to each of the steerable cars in the system.
be thought of as a generalization of an n-trailer system, in The chained form, which was introduced in [14] and used
which the only two controls available were the driving and in earlier work on the single-steering n-trailer system in [16]
steering velocities of the lead car, and for which the motion and [19], enables a variety of previously developed steering
planning problem was considered and solved in [16] and [19]. and stabilization techniques to be applied.
Mobile robot systems of this kind are of interest in practical A particularly intriguing aspect of this work is its con-
applications; part of the motivation for this work came from nection with an emerging body of literature on differentially
previous work on the fire truck [4], [18]. Also, we have been flat systems by Hiess and his co-workers [6], [15]. They
told anecdotally about the construction of such trailer systems have shown that chained form systems are a special case
with multisteering for use in nuclear environments [5] and also of differentially flat systems: the bottoms of the chains in
Manuscript received December 31, 1993. This work was supported in part the chained form play the role of flat outputs. For the two
by NSF under Grant IRI-904490. D. Tilbury was suppoaed in part by an input case, it was pointed out by Martin [ l l ] and Murray [131
AT&T Ph.D. fellowship. 0. J. S0rdalen was supported in part by the Center that, modulo somewhat different regularity conditions, chained
of Maritime Control Systems, Norwegian Institute of Technology, and by
SINTEF. form systems are equivalent to flat systems for the type of
D. Tilbury was with the Department of Electrical Engineering and drift-free systems that arise in nonholonomic motion planning.
Computer Sciences, University of Califomia, Berkeley CA 94720 USA. The results of the current paper indicate that this is not true
She is now with the Department of Mechanical Engineering and Applied
Mechanics, University of Michigan, Ann Arbor, MI 48109 USA. E-mail: for systems with more than two inputs without allowing for
[email protected]. the possibility of dynamic state feedback. As such this work
0. J. Serdalen was with the Department of Engineering Cybernetics, provides a valuable counterpoint to the results of Gardner and
the Norweigian Institute of Technology, 7034 Trondheim, Norway. He is
now with ABB Corporate Research, N-1361 Billingstad, Norway. E-mail: Shadwick [7] and Bushnell et al. [3].
[email protected].
L. Bushnell was with the Department of Electrical Engineering and U. THE SYSTEM MODEL
Computer Sciences, University of California, Berkeley CA 94720 USA.
She is now with the U.S. Army Research Office, Research Triangle Park, Consider a multisteering trailer system, i.e., a system of
NC 27709 USA. n (passive) trailers and m (steerable) cars linked together
S . S. Sastry is with the Department of Electrical Engineering and Computer
Sciences, University of California, Berkeley CA 94720 USA. by rigid bars, sketched in Fig. 1. Each body (trailer or car)
IEEE Log Number 9414514. is modeled as having only one axle, since, as has been
1042-296W95$04.00 0 1995 IEEE

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 808 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 11, NO. 6, DECEMBER 1995

Fig. 2. Showing the velocity relationships between adjacent bodes when

the rear body is a passive trader.

the state of the system. In this paper, the z and y positions of

Fig. 1. A multitrailer system with n (passive) trailers and m stemable axles.
the last axle will be used as configuration variables.
shown in [19], a two-axle car is equivalent (under coordinate Therefore, the configuration of a trailer system consisting
transformation and state feedback) to a one-axle car towing of n trailers and m cars with steering is completely given by
one trailer. [8;, ..., OAl, dl, . . . , Qi2,
q52, ..., Orm,
y, z]' which
is in the space ( s ' ) ~ + x~IR~. ~-'
A. Conjiguration Space
The steering axles are numbered from front to back, starting B. Kinematic Equations
with 1 and going up to m, and the passive axles are numbered The most direct way to construct a kinematic model for this
+
similarly from 1 to n. There are a total of n m axles in system is to write the differential form constraints representing
the system. The angle of each passive axle with respect to the the nonslipping of the wheels in terms of the configuration
horizontal will be represented by 0: where i E { 1,. . . , n } is variables, and then construct the input vector fields as the right
the axle number and j E { 1, . . . , m} is the number of the null space of the constraints. Dualizing the n + m constraints
steerable axle most directly in front of that axle. Each steerable (one for each axle) in a configuration space of n 2m 1 + +
axle together with the passive axles directly behind it is called state variables, a control system with m 1 inputs results. +
a steering train. The input vector fields for the steering inputs are constant
The steerable axles may be interspersed among the passive vector fields, but the driving vector field takes on a much more
axles in any fashion. The subscript of the passive axle directly complex form which would take a good deal of organization
in front of the jth steerable axle is nJ-l for j = 2, .. . , m. and bookkeeping to derive in its general form.
By convention, the first axle is assumed to be steerable, and For this reason, the kinematic model in this paper has been
thus no = 0. The superscripts associate the set of passive constructed in a somewhat different manner. Although even-
trailers behind each steerable car with that car. The angle of tually, the linear velocity of the front car will be considered
the first axle with respect to the horizontal is denoted by 0;. as one of the inputs, it is convenient to define the kinematic
There are n1 passive trailers in the first steering train; their equations in terms of the linear velocity of the last body. The
angles are denoted e:,
.. . , e,&.
The axle directly behind the projections of this velocity onto the horizontal and vertical
first steering train is steerable; its angle with respect to the directions will be the derivatives of the state variables x
horizontal is the superscript representing that it is in the and y, respectively. Proceeding toward the front of the train,
second steering train. For convenience of notation, let n, E n, velocities of each body can be recursively defined in terms
although in general the last axle will not be steerable. If the of the linear and angular velocities of the bodies behind it
last axle is steerable, then n,-1 = n,. in the train. This procedure defines all the derivatives of the
This system as defined is very general, and includes the angles of the passive axles as well as the derivatives of the
following as special cases: hitch angles. All that remains are the derivatives of the steering
1) the standard n trailer system of [IO], [14], [16], 1191 angles, which are defined to be the inputs.
corresponds to m = 1; and Starting at the rear of the train, let the linear velocity of
2) the fire truck of [41, [I81 corresponds to m = 2, n 2 = the last body be denoted wTm.Then the derivatives of x and
n1 = 1. y are the projections of this velocity onto the horizontal and
Let dJ denote the absolute angle (with respect to the vertical directions,
+
horizontal) of the bar connecting the ( j 1)st steerable axle
to the last axle of the 9th steering train (which may be either
y = sin vn", erm
x = cos on", Vn",.
steerable or passive). This can be considered to be the angle of
the bar connecting the ( j +l)st steering train to the jth steering Let vi represent the linear velocity of the axle with angle 0:.
train. The Cartesian position (x,y) of any one of the axles, Consider first the case of a passive trailer; refer to Fig. 2
along with all of the angles described above, will determine for clarification. Although this figure has been drawn for two

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 TILBURY et aL: MULTISTEERING TRAILER SYSTEM 809

passive trailers, these calculations are still valid when the front
body (z - 1) is steerable.
The linear velocity of body i - 1 can be broken into its
two perpendicular components: one is in the direction of the
linear velocity of body i, and the other is along the direction
of the angular velocity of body i. The two linear velocities are
related by the cosine of the angle between them, - y 7n A-"J+' (passive or steerable)

U; = cos (O;-l - 0;) (1)

first body of the
(j+l)ststeering train
and the projection of the angular velocity is by the sine of the (steerable)
difference angle,
L; 4; = sin (0;- - 0; (2) Fig. 3. Showing the velocity relationships between two bodies when the
rear body is a steerable car.
which gives the kinematic equations for the angles of the
passive trailers, 8;. A. Multiinput Chained Form
The equations are only slightly more complicated when the
A multiinput chained form system is defined as
rear body is a steerable car instead of a passive trailer (refer
to Fig. 3). Projecting onto the line connecting the two bodies, .i.; = U0
the relationship between the two linear velocities is ;;=U1
... i0
m -
-um
i: = Z,lUO i? = Z p L O
u;;1 cos (8;;l - $ 3 ) = v i 3 cos ( - $3), (4)
that is, both velocities are multiplied by the cosine of the angle z-,+11 = z:po *..

between the velocity vector and the connecting bar. 'm

&,+1 = Zn", uo
Now, because the velocity of the rear body, vi:', is no
longer perpendicular to the angular velocity vector LA?& ,this The state equations in each chain in (4) are multiplied by uo;
this is the generating input. A more general chained form [141
linear velocity will also contribute to the angular velocity & .
can have more than one generating input, and thus multiple
Adding up the contributions of the two linear velocities, the
chains leading down from each input. In this paper, only the
relationship
single-generator chained form will be of interest.
~ ; ~ + , i 3= sin - q~)v;:~ - sin - $ 3 ) ~ ; ~ Chained form systems were first introduced in [14] as a
class of systems to which one could convert a number of
for j E { 1, .... m - 1) defines the kinematics of the hitch interesting examples, including a car and a car with one trailer,
angles. and for which it was easy to derive steering control laws.
Since the derivatives of the steering wheel angles are free Some sufficient conditions for converting two-input drift-free
variables, that is, they are not constrained by the kinematics, systems into chained form were presented in [14]. In later
they can be considered to be controlled by the inputs, w3. work [4] in the context of steering the fire truck, sufficient
Combining all of these equations, the complete kinematic conditions were given for converting a multiinput system into
model of a trailer system consisting of n trailers and m a multiinput chained form. Necessary and sufficient conditions
steerable cars is given by for converting two-input systems into chained form were given
=w3 in [13], where a connection was made between the chained
form and its dual in the terminology of one-forms, called the
Goursat normal form. In [19] these results were applied to
convert the system of a car with n trailers into Goursat form.
1 The calculations in this context were simplified by the use of
$9 = _ _ [sin';0 :( - $3) - sin - $ 3 ) v;~]
X3+1 a coordinatization of the state space of the car with n trailers
y = sin Orm vTm introduced in [16].
In fact, the techniques of Sgrdalen were a way of system-
x = COS 8Fm (3) atically converting systems of n trailers into chained form by
f o r j E (1,. - a , m} andi E {n3-1+1, n 3 } ,withthetwo noticing that the trajectory of the (x,3) position of the last
..a,

velocity relationships (1) and (2). The inputs are the angular trailer determines the evolution of all the state variables of the
velocities of the steering axles, { w l , .... w"}, and the linear system, since the ratio of their derivatives will give the angle
velocity of the first car, U:. of the trailer through the relationship

tan 8, =. ,Yn
III. CONVERSION
TO CHAINEDFORM Xn
Now that the kinematic behavior of the multisteering system The hitch relationships can then be used to find the position
has been defined, it will be shown how these equations can be of the second-to-last trailer, and its angle, and so forth. This
converted to multiinput chained form. was also noted in [15] and [19].

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 810 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 11, NO. 6, DECEMBER 1995

From the structure of the multiinput chained form q u a -

tion (4), it is clear that the trajectories of the states at the
bottoms of each chain, that is, z $ ( t ) , ~ : ~ + ~. .(. t~,”,+~(t)

relationships
)),
will determine the trajectories of all the states through the
Fig. 4.
ae X(&

The unicycle model The robot is allowed to dnve forward or

backward and to spm about its center =IS.
i?
z? = 2+1 j E (1, . . ‘ , m}, i E ( 0 ) . ’ ) nj}. ( 5 ) B. Extension with “Virtual” Trailers
a
i;
The kinematic equations for the multitrailer system, as
The technique that is used in this paper to convert the derived in Section II, are drift-free, affine in the input, and
multisteering trailer system into chained form is now straight- can be written in the form
forward to explain. From a physical intuition about the system,
those states which determine the trajectories of the system i = G(z)u.
are identified. These states become the bottoms of the chains
of integrators in the chained form, as mentioned above, and Extending the system with dynamic state feedback corresponds
the rest of the coordinate transformation is found through to adding states <,a new input v , and defining a feedback for
differentiation (5). It is then verified that the transformation U . To keep the resulting system drift-free and affine in the
found in this manner is a local diffeomorphism and a valid inputs, it will be required that this feedback has the form:
change of coordinates. There is currently no generalization to
the Goursat normal form for multiinput systems which gives E =4 x , t b
necessary and sufficient conditions for converting systems into U =Y(X, 8..
multiinput chained form using dynamic state feedback.
For the multisteering system of Fig. 1, the (x,y) position of The states E that are added through dynamic feedback will be
the last trailer along with all the hitch angles {$I, . . . , #m-l} interpreted physically as the angles of “virtual” axles in front
between steering trains will determine the entire state of the of the steerable axles, diverging from the chain of trailers. The
system. To see this, consider the last steering train: using new input v will correspond to the driving velocity together
the technique described above for the n-trailer single-steering with the steering velocities of the virtual cars, and the feedback
case, all the angles of the trailers up to and including that of on U will be defined such that the actual steerable cars are
the last (or mth) steerable axle can be found. However, the controlled through the chain of virtual trailers.
hitch angle @”’ ahead of this axle will not be determined For some insight into this formulation of virtual axles,
by anything behind it. This is the reason that its evolution as consider the well-known example of a unicycle, sketched
a function of time is needed to specify the entire state of the in Fig. 4. The body is allowed to drive either forward or
system. The knowledge of the hitch angle $m-l will allow the backward and to spin about its axis.
Cartesian coordinates of the midpoint of the last axle of the The kinematic model takes as inputs the linear velocity ‘U
second-to-last steering train to be found, and this information and the angular velocity w of the body,
can be used to find the angles of all the axles in this steering
train, and so forth until the front of the entire chain is reached. j. =cos Bw
Another possible choice for the states at the bottoms of the y = sin Qv
chains are the ya values of the midpoints of the axles in front 0 =w.
of each of the steerable axles. The resulting chained form is the
same, with the same number of states added through dynamic Since the system is drift-free, the relative degree of any choice
state feedback. The coordinate transformation required to put of outputs will be equal to one (the input appears directly
the system into the chained form, however, will be different. in every state equation). In the sequel, when considering the
In related work (see [6] and [15]), the idea that certain multisteering trailer system, the relative degree of the states
variables determine the entire state of the system has been for- with respect to the steering inputs will be of interest. Note
malized in a more general setting, and these system variables that the relative degree of the body angle 8 with respect to the
have been referred to as$at outputs. y = h(x,U ) are defined to steering input is equal to one.
beflat for the system x = f ( x , U ) if all of the system variables Consider the dynamic feedback
(states and inputs) are differential functions of the outputs y;
that is, x and U are meromorphic functions of the outputs y $=a
and finitely many of their derivatives. Equivalently, the flat w = tan($ - 8)w.
outputs are outputs with respect to which the system has no
zero dynamics [9]. A system is called differentiallyjut if a set The added state has an attractive physical interpretation of
~

of flat outputs can be found. Moreover, there may be many being the angle of another axle added in front of the original
choices for the flat outputs. The multiinput chained form (4) steerable axle, and the new input CII is the steering velocity of
is differentially flat with flat outputs zoo, . . . z,,+~,
)m this “virtual” axle. This is represented in Fig. 5. In addition,
although chained form systems with more than one generator the relative degree of the body angle Q with respect to the new
are not in general flat. (virtual) steering input is now equal to two.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 TILBURY et al.: MULTISTEERING TRAILER SYSTEM 811

9
$ $2 ?'o...
Fig. 5. A unicycle with a "virtual" extension, interpreted as another axle ...........
added in front of the original robot. .........
j first virtual extension

The linear velocity at the front axle can be denoted by wf, e!

............
and it is related to the linear velocity of the rear wheel by the
cosine of the angle between them,

U = COS (IC, - Q ) u u ~ .

Thus, the extended system equations can be written as

x = COS Qw m-1
........ second steering train
y = sin Qw . .

..............
e l s t e e r i n g train
Remark: Any valid trajectory y = (z, y, 8 , 4) of the
system (7) can be projected down, via the standard projection ...............
T : IR2 x (S1)2+ Et2 x SI,to give a valid trajectory ~(y) =
= (x,y, 6') of (6). Also, for any trajectory C of (6) for which Fig. 6. The multisteering system, showing the virtual axles that must be
O(t) is C1 and for which 8 = 0 whenever ri: = y = 0, there added to convert the system into multiinput chained form.
exists a trajectory y such that ~ ( 7=)C.Trajectories where the
unicycle spins about its axis without moving either forward or After continuing similarly for 42, . , + m - l , y, a total of
a

backward cannot be achieved with the extended model. n3 virtual axles will have been added in front of the jth
This is the motivation for a dynamic state feedback that adds steerable axle, as has been sketched in Fig. 6. Now there are
virtual axles to the system. Each virtual axle that is added in the same number of passive axles between an axle 0: on the
front of a steerable axle increases by one the relative degree chain and any (virtual) steerable axle, and this is the same
of its hitch angle with respect to the steering input at the hitch. as the number of passive axles between the axle Qq and the
front steering wheel 6;. After these virtual axles have been
C. Virtual Extension for the Multisteering System added, the jth steering train now contains n3 axles, of which
It will now be described in detail how the kinematic model only n3 - n3-1 are real (physical). The only axles which are
(3) is locally converted to a multiinput chained form using considered as steerable in this formulation are the first axles
dynamic state feedback and a coordinate transformation. of each virtual extension, or 6; for j E { 1, .... m}.
As described in Section III-A, the states at the bottoms of The state variables that have been introduced, which corre-
the chains are chosen to be 2, 4l,.... 4m-1, y. Consider the spond to the angles of these virtual trailers, are denoted by Q:
front-most hitch angle 4'; this will become the state at the f o r j E (2, + . e , m } , i E (0, - " , nJ-1-1}.Theirderivatives
bottom of the first chain, or z:,+~. Its relative degree with are defined as if they were actual axles,
respect to the first steering input u 1is equal to n1+ 2, or one 0; = p J ,
more than the number of axles in the first steering train. In . 1
order to define all the states z1 in the first chain by (5), q5l Q: = - sin (Q:-l - 0:) (8)
will need to be differentiated a total of n1+ 2 times. However, LI
since 4' depends on all the angles behind it in the trailer for j E (2, .... m} and i E (1, .... n3-1 - l}, where LI
system according to (3), the relative degree of $1with respect is an arbitrarily chosen positive parameter (usually chosen to
to any of the other steering inputs u2,.... um will be equal be equal to one for simplicity). The velocities of the virtual
to two. axles are defined in the same manner as the velocities of the
The derivatives of these inputs should not appear in the real axles, (1) and (2).
coordinate transformation. In order to increase the relative The new (fictitious) inputs are denoted by p3. for j E
degree of $1with respect to the other steering inputs, n1 virtual (1, . - a , m } , and as a notational convenience let p1 := ul,
axles will be added in front of each steering axle for since no virtual axles need be added in front of the first car.
j E (2, e . ., m}. The virtual inputs are temporarily denoted These inputs p3 represent the angular velocity of the front car
by 123, the angular velocities of the axles at the front of each in each virtual extension. In effect, the angular velocities of
virtual chain. the steering wheels which are in the middle of the chain are no

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 -

812 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 11, NO. 6, DECEMBER 1995

longer directly controlled; rather, they are controlled indirectly An extended kinematic model can now be derived in the
through this virtual steering train. Therefore the states in the subset D. To this end, the input w is introduced as the linear
extended system will be functions of the (true) inputs w3 and velocity in the 3: direction of the last body in the last steering
their derivatives. The true kinematic inputs W J are of course train,
the derivatives of the actual steering angles, A
w = cos erm
Wn”,. (11)
U3 =O’ nJ-1
- The linear velocities of the other bodies w,” can then be
- sin(8i3-l-1 - Q;J-J42-l-17 (9) expressed as multiples of this generating velocity w,
for j E (2, . . . , m}. The complete configuration space of the
extended system can now be defined as
* 1 [z, o;, . . . , 81
nl’
41, 8 01
2 . ..,.. , The linear velocity at a wheel 0,” will depend on all the
difference angles behind it, fii, e;:‘,
.. ., ~ 9 z ~ -~.
Although
67, . . , Cm,YIT
‘
the vector @: contains more angles than this, the velocity v,”
E (sl)N+2m-l ~2
will be written as a function of 63:. Using (1) and (2), it can
where N = C,”=,n, is the number of passive axles, both real be shown that the function s ; ( s ) will have the form
and virtual, in the extended system.
In this paper, it is assumed that the configuration of the
system with the dynamic feedback is in a subset D of the
extended configuration space, defined as the set where all of
the relative angles between adjacent axles and hitches are less
than 7r/2,
- 4‘) cos (0:;’ - &}.
D 2 {* E (Sl)N+2”-1 x Et2: 18; - e:+ll < -,
ll
2
x sec (12)
This function is smooth in D .
It is also possible to write the derivative of the states in the
The kinematic equations are well-defined on this set. compact form

D. Kinematics of the Extended System

In an effort to write the kinematic model in a compact form Consider first the derivatives of the angles which have e:,
and also to show the triangular structure of the coordinate the same form for both actual (3) and virtual (8) axles, and
transformation, the following vectors are introduced: define the function f,” to be the derivative of e: divided by
the velocity w
e,”a [e:, ‘ .., e;J, 4 3 1
a
- [e,-, ’ “ ,
= Yl erm,
3a 0’
4i?, - L% , Q3+1 . . .
4, > irl. (10)
In general terms, the superscripts G)of the vectors refer to the
jth steering train for j 6 {I, . , m}. The subscripts (i) refer
e
for j E (1, m}, i 6 (1, - . . , n31. Now define the
. e - ,

to the tails of the steering train starting from the ith trailer functions f;ZJ+l to be the derivatives of the hitch angles, 4 3 ,
(which is real if z 2 n3-1 and virtual if 0 5 i < n3-l).Thus, divided by the velocity ‘U. Recall that the equation for the
we have the following. kinematics of 4 was given in (3), so that the functions f A J + ,
Thevector@:,forj E (1, ..., m-l}, i E (0, ” - , n3},
0
are
refers to the angles of the axles in the jth steering train
behlnd (and including) the level of the zth axle. The hitch
angle 4’ is also included.
The vector
0 e:,
for z E (0, ..., nm}, refers to the
angles of the axles in the mth steering train behind (and
including) the level of the ith axle. Since there is no hitch
angle behind the last steenng train, the position y of the
midpoint of the last axle has been included.
a The vector 63,” groups all of the angles of the steering for j E { 1, . . . , m - l}. The final function that is defined
trains from j to m for the axles from i to the end of the here is the derivative of the y coordinate of the last trailer.
jth train n3.The vector 63; contains all of the angles of
all the axles, both real and virtual, in the entire extended
From (3) it can be seen that equating y = fzm+l
‘U will give

system.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 TILBURY et aL: MULTISTEERING TRAILER SYSTEM 813

The kinematic model (3) with the dynamic feedback (8) can where & is defined such that
then be rewritten locally in the following manner:
- 03 "3 T T
5203 - [ 0, (00) I
4'
-0--
P j> j E (1, e * . , m}
*j
gi = f;(&l)?l, j E (1, * . a , m } ,i E (1, a - . , nj} and is found from (10) and (17).
j E (1, * * . , m - l}
Theorem 1: Let the coordinates 2," , j E (0, . . . , m}, i E
= f:J+l(@:,)u,
$j
(0, +
, nj l},be given by (18)-(22) and the inputs u3,j E
Y = frm+dfgm) U (0, m}, be given by (23) and (24). Then the kinematic
. a . ,

x =U. (16) equations have the form of (4).

Pro08 The chained form follows directly from the defi-
By way of notation, define
nitions of the coordinates and input transformation along with
the local kinematic model (17).
F{
- k [ii,
-f?',
z - . e , fr]' To show that this coordinate transformation is a local
diffeomorphism, it will be shown that it has a triangular
so that the local kinematic model with dynamic feedback (16) structure and that its Jacobian is nonsingular at the origin.
can be written compactly as Let the state of the extended system be

&pj, j € {l,..*,m} q = [z, g,; !g, . . * , e21

0 1 -
-1
1
- Fl(O0) -lJ
1 = [ ( I 0 , q l , q 2 , * - . ,Q"1
x =U. (17) where the states have been partitioned as follows:
qo = x
E. Conversion to Chained Form
In this section, a coordinate transformation to a multiinput
q1 =&
chained form will be found. The first chain has only one =[e:, e:, *..I Cl,
$7
coordinate
20" 2, (18) qm-l - e,
"-I
whose derivative will act as the generator (remember that the
generating input is U = cos e'; ucm
= i).The mth chain will
= [e,"-l, e;.-', . .,l.:,::e +"-lI

be the longest, and its last coordinate is the other Cartesian

q" =e:
coordinate =[e,-, e;", * .. , e:, y].
m
%,+1
a
= Y. (19) The dimension of the state space is 1+ (nl + 2) .+ (n, + + a -
2) = N + 2m + 1, as before.
The other chains have the hitch angles at the bottom The new states z are partitioned in an analogous way,
' A
j E (1, m - l}, (20) z = [zo,z', . . , z"] where
e

z+
,;l = $3, * e ' ,

and the remaining coordinates are found through the relation- z j = [ z i , * .. , Z t , , , ] .

ship (5).
This can be written more specifically as follows. Recalling Not only do q and x have the same dimension, but q j and z3
that the derivatives of $ l , . . , $"-', y, were defined as also have equivalent dimensions (equal to n3 2 for j # 0
- +
f A J t l . for j = 1, . . . , m, it can be seen that the second-to-last and 1 for j = 0).
coordinate in each chain will be The Jacobian of the coordinate transformation
az
j
zn, 5 f n,+l
j (q.
-n (21) J = -
89
The general form of the new coordinates is
is block-upper triangular and the elements on the diagonal are
nonzero on the open set D of interest. This will be shown
most easily by considering J block-by-block
where LEh denotes the Lie derivative of the function h along
the vector F. Each coordinate zf will be a function of the
state variables e:.
The input transformation is defined by taking the derivatives Since z i is a function of @, it is a function only of
of the first states in the chains for j E { 1, . . . , m} q j , qj+l, , 4". This implies that the blocks J3,k = 0
whenever j > IC; i.e., all blocks below the diagonal of J
are zero.
Next, it will be shown that each diagonal block Jj,j is
upper-triangular with nonzero diagonal entries. Since the first

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 814 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 11, NO. 6, DECEMBER 1995

block is only one-by-one, consider it first. By definition 2: = parameterize the input space with at least as many parameters
II: = go, thus JO,O= 1 which is nonzero. as there are states, integrate the chained form equations
Now consider the other diagonal blocks, J3,?, and note first symbolically, and finally, solve for the input parameters in
of all that the lower-right entry in each block is equal to 1, terms of the desired initial and final states.
since by definition Various approaches for feedback stabilization of chained
3 3
form systems are also mentioned. Although most of the work
%,+1 = 4n,+l in these areas have concentrated on two-input systems, the

=i $3
y
j E (I,
j=m.
m - 1) decoupled form of the multiinput chained form system will
allow the techniques to be generalized in a straightforward
manner.
To calculate the next diagonal entry, note that zA3,which was In this section no particular system of trailers or nonlinear
defined in (21), is a function of BA,, and the dependence on equations will be considered. The problem that is solved in
q i J = 6)& is through a tangent function. Therefore, this section is as follows: Given a system of equations in
multiinput chained form (4) and a desired initial and final
state, find inputs which will steer the system from the initial
state to the final state.
1
sec2 - 4 ~cos
) (0;;' -q ~)si:~(~iT~)
A. Polynomial Inputs
G,+1
j~ (1,...,m-l} One approach to the point-to-point steering problem is to
sec2 Qcmj = m hold the first input uo constant and identically equal to one
over the entire trajectory. The time needed to steer is then
where the function s: is the velocity function defined in (12)
determined from the change in the 20" coordinate,
and is nonzero in D.
The other diagonal entries are found similarly. Since each (25)
x: is a function of only Bi, and depends on 8: = q: through a
tangent function, each diagonal element of the Jacobian matrix The parameters for the remaining inputs are chosen to be
will be a product of secants and cosines of difference angles coefficients of a Taylor polynomial,
and will be nonzero on D (indeed it will be equal to one at
the origin).
The input transformation can also be seen to be nonsingular.
Defining -p to be the vector of the virtual inputs,

P = [Pl,
- " ' 7 P"1, um = vo + v1t + . . . + vnm+ltn,+l (26)
and g to be the vector of the transformed inputs, with the number of parameters on each input chosen to
be equal to the number of states in its chain. The chained
form equations can be integrated symbolically and the input
the Jacobian matrix of this transformation can be seen to be parameters a 3 ,,B3, . . v3 can be found in terms of the initial
)

and final states. This is a fairly simple procedure since all of

the equations that need to be solved are linear.
Of course, if the time needed for steering is zero from (25),
This matrix is upper-triangular as well, and its diagonal then this method will not work. This case corresponds in the
elements are nonzero. physical system to the "parallel-parking'' dn-ection, or no net
Each of the inputs uJ for j E { 1, . . , m } depends only on change in the x coordinate. The easiest way to remedy this
the derivatives of the states E);, giving J;,k = 0 whenever situation is to first choose an intermediate point and then plan
j > k . The diagonal entries are equal to the path in two pieces.
du' -_32;_ -
- -- -
32;
dP3 ae; aq; B. Piecewise Constant Inputs
This steering method was originally inspired by multirate
by (23). This is the same as one of the diagonal entries of J
digital control [12], but is most easily understood in terms of
above, and is nonzero by the same argument. 0
motion planning simply as piecewise constant inputs. The first
input uo is chosen to be constant over the entire trajectory.
IV. STEERING CHAINED FORM SYSTEMS This choice will ensure the linearity of the equations that
Once a system is in multiinput chained form, many different need to be solved for the other input parameters, as well as
algorithms can be used to steer it. Three methods are briefly generate "nice" trajectories (since this input is related to the
described in this section; all three of them were presented driving input of the multitrailer system, a constant uo will
in [14] and 1191 for steering two-input systems in chained usually transform to a uni-directional velocity, or equivalently
form. The basic idea behind each of the three methods is to no back-ups).

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 TILBURY et al.: MULTISTEERING TRAILER SYSTEM 815

The other inputs are chosen to be piecewise constant, and

to guarantee that the resulting equations have ,a solution, each
input should have at least as many switches as there are states
in its chain. There will need to be the largest number of
switches on the mth input since it has the longest chain (by
definition).
The time for the trajectory can be chosen arbitrarily as T.
The first input is constant over the entire trajectory, uo(t) =
U& for t E [0, T), where the magnitude is such that the first
chained form state will move from its initial to its final position
over the time period,
Fig. 7. A sketch of the fire truck system showing @e virtual extension that
U$ = [(4)f
- (4?1
(27) is added in front of the rear steering wheel. These vehicles are equipped with
T a long ladder on the trailer and are used by lire departments in large cities in
the United States. The extra steering wheel at the rear of the trailer is used
The other inputs are piecewise constant. Let the switching for improved on nmow
times be
0 = tt < ti < < ti3+2 = T, case, is proposed here. This method required only one step, and
all of the necessary frequencies are applied simultaneously to
+
where n3 2 switching times are needed for each input since the inputs
there are n3+2 states in the jth chain. There are many different
methods available for choosing these times. They are most uo =a0 crsin wt+
commonly chosen so that for the mth input, which has the u1 = P O +
~1 cos w t + +
. . . ,&,+I cos (nl 1)wt +
most switching times, the holding times will be equal. The
switching times for the other inputs are then chosen to be some
subset of the switching times for the mth input. The jth input + + +
um = vo vl cos wt ' . . vnn+1 cos ( n , 1)wt. +
will be of the following form: u"t) = ukkfort E [ t k , tk+l).
When the chained form equations are integrated using these The input parameters are found in the same manner as in
input values, the final state can be expressed in terms of the the other methods: the chained form equations are integrated
inputs and the initial state as symbolically, evaluated at time T, and the parameters are
solved for as a function of the initial and final states.
The main drawback to this approach is that there will be
some interference between the levels (although not between
chains) and solving for the input parameters will require
solving nonlinear algebraic equations. In the simple cases that
we have explored, this has not been a problem for a symbolic
where the matrices A3 are assured to be nonsingular whenever mathematical software package.
the first input U& is nonzero [12]. When this method is implemented on a multisteering trailer
Similarly to the previous section, if the first input does come system, the first input, which always goes through one period,
out to be zero from (27), then a slight modification of this will transform back to the driving input, which will usually
method is necessary. A multirate input can also be added on change direction (at least one back-up). This seems to work
uo, using at least two time periods, or an intermediate point well when parallel-parking type maneuvers are desired. The
can be chosen and the path can be planned in two steps. free parameter Q can be adjusted to change the distance that
the trailer system drives forward before it backs up.
C. Sinusoidal Inputs
A method for steering multichained systems with sinusoids D. Stabilization for Multiinput Chained Forin Systems
was proposed in [4]. This method is step-by-step and uses one Chained form systems are open-loop controllable, as shown
step to steer each level of the chain (although the states of all above by the various point-to-point steering algorithms, but
chains at the same level can be steered simultaneously). Since are not stabilizable by pure smooth static-state feedback [2].
the longest chain has length n, + 2, this is the number of Bearing this result in mind, various researchers have tied to
steps that will be needed. stabilize such systems by time-varying or nonsmooth state
Although this method works perfectly well, and the magni- feedback.
tudes of the sinusoids are simple to solve for, the algorithms Many of the algorithms for point stabilization require the
can be tedious in practice because of the many steps that are system to be in chained form. For two-input systems, a class
needed. The trajectories that are generated consist of many of smooth, time-varying control laws for local and global
segments and do not always follow a very direct path between asymptotic stabilization to a point was given in [17]. This
the start and goal. procedure was extended in [20] to locally asymptotically sta-
+
Therefore, an "all-at-once'' sinusoids method, which is an bilizing the origin of (m 1)-input, m-chain, single-generator
extension of that detailed in [191 for the two input single chain chained form systems. This method consists of taking the

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 816 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 11, NO. 6, DECEMBER 1995

Fig. S. The five-axle, two-steering system showing the virtual exhension which is added in front of the second steering wheel. Such a system could be
envisioned as being used in a nuclear power plant or in a dangerous area where maneuverability around narrow passageways is of utmost importance.

chained form system and converting it into power form, which kinematics of the second steering axle 6’: are thus altered: it
is diffeomorphic to chained form. The control laws, which are is no longer steered directly, but through the virtual steering
time-varying functions of the state, will stabilize the system wheel.
in power f o m . The bottoms of the chains in the multiinput chained form are
Many of the other results which have been presented in the the ( 5 , y) coordinates of the rear axle and the hitch angle &,
literature for two-input chained form systems could also be and the rest of the coordinates are found through mfferentiation
extended to multiinput single-generator chained form systems according to (5),
in a straightforward manner.

To illustrate the procedure presented in this paper for 22” 6y

converting multisteering trailer systems into chained form,
the algorithms given in Section ID will be followed for two
example systems.

A. Fire Truck
Although the fire truck example has been examined exten-
sively in previous work, it can also be considered in terms
of the algorithms described in this paper, and in fact, the
formulation is somewhat different than in [4]. In that paper,
the bottoms of the chains in the multiinput chained form were (the chains are written upside down here to show the order in
chosen to be the (5, y) position of the passive axle along which the coordinates are calculated: starting at the bottom).
with the angle of the trailer (see Fig. 7). Because of the The resulting equations are in multichained form.
relative simplicity of the three-axle system, that choice allowed
kinematic equations to be put into multiinput chained form 3. A Five-Axle System
without using dynamic state feedback. The fire truck fits into Consider a five-axle system with two steering wheels,
the class of multisteering trailer systems, thus the kinematic depicted in Fig. 8. In effect, this system is a fire truck with two
equations can also be converted into multiinput chained form passive trailers. With these extra trailers, the (2, y) position
using a virtual extension (and a different choice of states at of the first passive axle, along with the trailer angle $Iwill ,
the bottoms of the chains). Although this extension is not no longer define the entire state of the system.
necessary for this particular system, no systematic procedure is The kinematic equations for this system can be found from
known for transforming a general multisteering trailer system (3). The system consists of two steering trains, the first has
into multiinput chained form without using the sort of virtual length two and the second has length three. One virtual axle
extension that is proposed in this paper. will need to be added to the second steering train since there
The kinematic equations for the fire truck can be obtained is one passive axle in front of its steering car; see Fig. 8.
from (3) and will not be repeated here. This new state corresponding to the angle of the virtual axle
The system has two steering trains, the first has length two is denoted e:, and the kinematics of must be changed to
and the second has length one. Since there is one passive represent that the angular velocity of the second steering wheel
axle in front of the second steering train, this train will be is no longer controlled directly by the input w 2 but indirectly
augmented by the addition of one virtual axle as described through the virtual steering wheel Qg.
The input p2 will now
in Section III-C. The angle of this virtual axle is denoted control the virtual steering velocity 6’;. When needed, the real
0;. A sketch of the extended system is shown in Fig. 7. The input w2 can be calculated as the derivative of the angle 0:.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 TILBURY e l al.: MULTISTEERING TRAILER SYSTEM 817

the original coordinates. A movie animation was made of this

trajectory; scenes from this movie are shown in Fig. 9. The
path taken by the virtual axle is not shown.

REFERENCES
[l] J. C. Alexander and J. H. Maddocks, “On the maneuvering of vehicles,”
SIAMJ. Appl. Math., vol. 48, no. 1, pp. 38-51, 1988.
[2] R. W. Brockett, “Asymptotic stability and feedback stabilization,” in
Differential Geometric Control Theory, R. W. Brockett, R. S. Millman,
and H. J. Sussman, Eds. Boston: Birkhauser, 1983, pp. 181-191.
[3] L. Bushnell, D. Tilbury, and S. S. Sastry, “Extended Goursat nor-
mal forms with applications to nonholonomic motion planning,” in
Proc. IEEE Con$ Decision and Control, San Antonio, TX, 1993, pp.
3447-3452.
Fig. 9. A parallel-parking trajectory for the five-axle, two-steering system. [4] -, “Steering three-input chained form nonholonomic systems using
The planning algorithm as described in this paper does not account for sinusoids: The fire truck example,” Int. J. Robotics Res., vol. 14, no. 4,
obstacle avoidance; however, it does plan “nice” paths which may be used pp. 366-381, Aug. 1995.
in conjunction with an obstacle-avoidance algorithms to achieve a complete [5] C. Canudas de Wit, personal communication, 1993.
[6] M. Fliess, J. Uvine, P. Martin, and P. Ronchon, “Flatness and defect
solution to the path-planning problem.
of nonlinear systems: Introductory theory and examples,” Int. J. Contr.,
vol. 61, no. 6, pp. 1327-1361, June 1995.
The velocities of all the other bodies can be calculated [7] R. B. Gardner and W. F. Shadwick, “The GS algorithm for exact
in terms of the U , the linear velocity of the last trailer in linearization to Brunovsky normal form,” IEEE Trans. Automat. Contr.,
the horizontal direction, from (1) and (2). The bottoms of vol. 37, no. 2, pp. 224-230, 1992.
[8] G. Giralt, personal communication, 1993.
the chains will be the (2,y) position of the last trailer and [9] A. Isidori, Nonlinear Control Systems, 2nd ed. New York Springer-
the angle r$l of the hitch connected to the second steering Verlag, 1989.
wheel. The other coordinates in the chained form are found [lo] J.-P. Laumond, “Controllability of a multibody mobile robot,” IEEE
Trans. Robotics Automat., vol. 9, no. 6, pp. 755-163, Dec. 1993.
by differentiation. The complete coordinate transformation is [Ill P. Martin and P. Rouchon, “Systems without drift and flatness,” in Math.
too complicated to include here but may be obtained from the Theory Networks Syst., to appear.
[12] S. Monaco and D. Normand-Cyrot, “An introduction to motion planning
first author via electronic mail. under multirate digital control,” in Proc. IEEE Con5 Decision and
Control, Tucson, AZ, 1992, pp. 1780-1785.
C. A Parallel-Parking Trajectory [13] R. M. Murray, “Nilpotent bases for a class of nonintegrable distributions
with applications to trajectory generation for nonholonomic systems,”
Once the kinematic equations are in multiinput chained Math. Contr., Signals, Syst.: MCSS, vol. 7, no. 1, pp. 58-75, 1994.
form, the system can be steered by one of the algorithms [14] R. M. Murray and S. S. Sastry, “Nonholonomic motion planning:
discussed in Section IV. As an illustration, consider the Steenng using sinusoids,” IEEE Trans. Automat. Contr., vol. 38, no.
5, pp. 70CL716, May 1993.
parallel parking maneuver shown in Fig. 9 for the five-axle, [15] P. Rouchon, M. Fliess, J. Levine, and P. Martin, “Flatness and motion
two-steering system described in Section V-B. The system planning: The car with n trailers,” in Proc. European Control Con$,
parameters have been chosen as n = 3 (three passive axles), Groningen, The Netherlands, 1993, pp. 1518-1522.
[I61 0. J. S~rdalen,“Conversion of the kinematics of a car with N trailers
m = 2 (two steering wheels), and the lengths of the hitches into a chained form,” in Proc. IEEE Int. Con$ Robotics and Automation,
as Li = Lg = Li = 5, and Li = 3. Atlanta, GA, 1993, pp. 382-387.
[17] A. Teel, R. M. Murray, and G. Walsh, “Nonholonomic control systems:
Polynomial inputs in the chained form equations are used to From steering to stabilization with sinusoids,” in Proc. IEEE Con$
steer the system from an initial point of (2,y) = ( 0 , 20) to a Decision and Control, Tucson, AZ, 1992, pp. 1603-1609. Also, to
final point of (z, y) = (0, 0), where (z, y) are the coordinates appear in Int. J. Contr.
[I81 D. Tilbury and A. Chelouah, “Steering a three-input nonholonomic
of the midpoint of the last axle. All of the body angles are system using multirate controls,” in Proc. European Control Con$,
aligned with the horizontal axis in both the initial and final Groningen, The Netherlands, 1993, pp. 1428-1431.
configurations. [19] D. Tilbury, R. Murray, and S. Sastry, “Trajectory generation for the N
problem using Goursat normal form,” IEEE Trans.Automat. Contr., vol.
As noted in Section IV-A, polynomial inputs are not imme- 40, no. 5, pp. 802-819, May 1995.
diately suited to this type of trajectory since the time needed [20] G. C. Walsh and L. G. Busbnell, “Stabilization of multiple input chained
to steer the system, computed from (25),would come out to be form control systems,” Syst. Contr. Lett., vol. 25, no. 3, pp. 227-234,
June 1995.
zero and the algorithm would fail. Therefore the trajectory was
planned in two steps, choosing an intermediate point (z, y) =
(30, 10). The virtual angles were chosen equal to zero in both
the initial and final states, and the virtual hitch length was Dawn Tilbury (S’91-M’95) received the Bach-
chosen as L: = 1. The procedure is first to transform the initial elor’s degree in electrical engineering from the
and final states into the chained form coordinates. Then, using University of Minnesota in 1989. From 1989 to
1994, she was a graduate student at the University of
the polynomial inputs methods discussed in Section IV-A, the California at Berkeley, where she received the M.S.
chained form inputs needed to steer the system are found. and Ph.D. degrees in 1992 and 1994, respectively.
These inputs can then be transformed back to the original She has been a visiting scholar at LAAS, Toulouse,
France; LSS, Paris, France; M.I.T., Cambndge; and
coordinates to find the virtual inputs, and the real inputs can Harvard University, Cambridge.
finally be calculated using the relationship (9). She is currently with the Department of Mechani-
The simulation was performed on the system in the chained cal Engineering and Applied Mechanics, University
of Michigan, Ann Arbor, as an Assistant Professor. Her research interests
form coordinates, then the inverse coordinate transformation are in the area of nonlinear control with applications to mobile robots and
was used on the simulation data to obtain the trajectory in transportation.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.
 818 IEEE TRANSACTIONS ON ROBOTICS AND AUTOMATION, VOL. 11, NO. 6, DECEMBER 1995

Ole Jakob S6rdalen (S’92-M93) was born in S. Shankar Sastry (S’79-M”MM’90-F’95)

Krager0, Norway, in 1965. He received the Diploma received the Ph.D. degree in electrical engineering
Engineer (Siv.Ing.) degree in 1988 and the Dr.Ing. in 1981 from the University of California at
degree in 1993 in electrical engineering and com- Berkeley.
puter sciences from the Norwegian Institute of Tech- He was on the faculty of Massachusetts Institute
nology, Trondheim, Norway. of Technology, Cambridge, from 1980 to 1982.
He is currently with ABB Corporate Research He is currently with the Department of Electrical
Norway. He has held visiting appointments at Lab- Engineering and Computer Sciences, University
oratoire d’Automatique de Grenoble, France, the of California at Berkeley, as a Professor. He
University of California at Berkeley, and the Uni- was with Harvard University, Cambridge, as the
versity of Tokyo. His research interests include Gordon McKay Professor of Electrical Engineering
nonlinear control of mechanical systems and nonholonomic systems. om 1994. He has held visiting appointments at the Australian National
University, Canberra; the Umversity of Rome; LAAS, Toulouse; and the
Center for InteUgent Control Systems at M.I.T. His areas of research are
nonlinear control. nonholonomc motion Dlanmnn, robohc remote surgery,
I I .

venfication of hybnd control systems, and biological motor control. He is

Linda Bushnell (S’85-M’95) received the B.S. de- a coauthor (wth M. Bodson) of Adaptive Control. Stability, Convergence
gree in electrical engineering and computer sciences and Robustness, (Prenhce Hall, 1989) and (with R. Murray and Z. Li) of A
in 1985, and the M.S. degree in electrical and sys- Mathematical Introductzon to Robotic Manzpulatzon (CRC Press, 1994).
tems engineering in 1987, both from the University Dr. Sastry was an Associate Edtor of the IEEE TRANSACTIONSON
of Connecticut at Storrs. She received the M.A. AUTOMATICCONTROL,IEEE Control Magazine, IEEE TRANSACTIONS ON
degree in mathematics in 1989 and the W.D. degree O ~ mAND s SYSTEMS, and the Journal of Mathematical Systems, Estimation,
in electncal engineering and computer sciences in and Control and is an Associate Editor of the IMA Journal of Control
1994, both from the University of California at and Information, the International Journal of Adaptive Control and Signal
Berkeley. Processing, and the Journal of Biomimetic Systems and Materials. He received
She is currently with the U.S. Army Research the President of India Gold Medal in 1977, the lBM Faculty Development
Office, Research Triangle Park, NC. Her general award for 1983-1985, the NSF Presidenhal Young Investigator Award in
research interests include &stributed interactive simulation, motion planning 1985, and the Eckman Award of the of the Amencan Automahc Control
for nonholonomic systems, intelligent control, hybnd systems, and exterior Council in 1990.
differential systems.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY KANPUR. Downloaded on July 24,2023 at 11:30:14 UTC from IEEE Xplore. Restrictions apply.