Proceedings of the 18th World Congress

The International Federation of Automatic Control

Milano (Italy) August 28 - September 2, 2011

Control of a Five-axle, Three-steering

Coupled-vehicle System and its
Experimental Verification
Hiroaki Yamaguchi ∗ Masato Mori ∗∗ Atsushi Kawakami ∗∗∗
∗
Department of Integrated Information Technology, Aoyama Gakuin
University, 5-10-1 Fuchinobe, Chuo-ku, Sagamihara-shi, Kanagawa
252-5258, Japan (e-mail: yamaguchi@robotics.it.aoyama.ac.jp).
∗∗
Amada Co., Ltd., 200 Ishida, Isehara-shi, Kanagawa 259-1196,
Japan
∗∗∗
Department of Integrated Information Technology, Aoyama Gakuin
University, 5-10-1 Fuchinobe, Chuo-ku, Sagamihara-shi, Kanagawa
252-5258, Japan (e-mail: kawakami@robotics.it.aoyama.ac.jp).

Abstract: This paper presents a new type of coupled-vehicle system: a five-axle, three-steering
coupled-vehicle system consisting of two vehicles which are car-like mobile robots, two carriers,
and a steering system, and presents its path-following feedback control method. We first
show that, by assuming virtual mechanical elements, it is possible to convert the kinematical
equation of the coupled-vehicle system into a three-chain, single-generator chained form in a
new coordinate system where an arbitrary path whose curvature is two times differentiable is
one coordinate axis and a straight line perpendicularly intersecting the tangent of the path is the
other coordinate axis. Based on the chained form, we secondly derive the path-following feedback
control method which enables the orientations of the two carriers relative to the tangent of the
path to be controllable. By the feedback control method, it is possible to form the two carriers
into a straight-bed carrier configuration or a V-bed carrier configuration and to cause them to
maintain such configurations while performing a path-following operation thus enabling the two
carriers to be configured in accordance with shapes of objects to be transported. The validity
of the mechanical design of the coupled-vehicle system and its path-following feedback control
method has been verified experimentally.
Keywords: nonlinear systems, nonholonomic systems, differential geometry, chained form,
nonlinear control, autonomous vehicles, coupled-vehicles, path-following

1. INTRODUCTION tural object, which increases the likelihood of interference

Automation of transportation operation is an impor-
tant research subject in control engineering. This paper
presents a new type of coupled-vehicle system with five
axles and three steering systems, and a path-following
feedback control method for the system.
Coupled-vehicle systems generally consist of two or
more vehicles each having one or more tractors, such as
mobile robots with two independently driven wheels or
car-like mobile robots, connected to one or more trail-
ers by passive revolute joints. In coupled-vehicle systems,
movement is usually obtained by delivery of driving force
generated in a mobile robot at the front end of the system
via its drive wheels, with the trailers towed behind on
passive wheels. It is also possible to use two mobile robots
with drive wheels, with one at the front end and the other
at the back end of the system. In some cases, moreover,
mobile robots are placed elsewhere in the system.
For coupled-vehicle systems which transport large struc-
tural objects, such object is generally loaded onto a single
trailer, which is towed by a single car-like mobile robot
(Sampei (1995), Samson (1995), Altafini (2001), Lizarraga
et al. (2001), and Deligiannis et al. (2006)). The required
length of the trailer increases with the length of the struc-
with obstacles in the operating environment and thus
complicates the transportation operation, and therefore
there have been proposed various methods to increase
the maneuverability of the coupled-vehicle system by the
addition of a mechanism for actively varying the rolling
direction of the trailer’s passive wheels relative to the ori-
entation of the trailer (Bushnell et al. (1995), Tilbury et al.
(1995), Tilbury and Sastry (1995), Nakamura (2001), and
Lamiraux et al. (2005)). For the same purpose, the authors
have previously proposed the connection of one trailer
to two car-like mobile robots via two passive revolute
joints together with cooperative steering (Yamaguchi et al.
(2010)). In such transportation systems (Sampei (1995),
Samson (1995), Altafini (2001), Lizarraga et al. (2001),
Deligiannis et al. (2006), Bushnell et al. (1995), Tilbury et
al. (1995), Tilbury and Sastry (1995), Nakamura (2001),
Lamiraux et al. (2005), and Yamaguchi et al. (2010)),
however, the load bed configuration cannot be changed
to facilitate the transportation of large structural objects
of differing configurations, and such objects therefore ex-
tend well beyond the trailer bed during the transpiration,
increasing the risk of load and trailer overturning.
To resolve such problem, we propose a modification of
our previously proposed cooperative transportation sys-

978-3-902661-93-7/11/$20.00 © 2011 IFAC 12976 10.3182/20110828-6-IT-1002.01455

18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

tem employing two car-like mobile robots, which as shown joint (”joint-2”), and a steering system whose steering
in Figure 1: (a) the single bed is replaced by two beds axis is in the same position of the rotating axis of joint-
(two ”carriers”) connected by a passive revolute joint; and 2 is attached to carrier-1, thus enabling active control of
(b) a steering system is added to the forward carrier in the rolling direction of the passive wheels of the steering
the same position of the passive revolute joint coupling system relative to the orientation of carrier-1. Carrier-2
the two carriers. The modified cooperative transportation is coupled to the second car-like mobile robot (”vehicle-
system thus employs five axles and three steering systems. 2”) via a third passive revolute joint (”joint-3”) at the
Key advantages of this coupled-vehicle system over pre- midpoint of the vehicle-2 rear axle.
viously proposed systems include the capabilities for: (1) This paper presents a path-following feedback control
formation of a single straight-bed carrier configuration by method which causes the coupled-vehicle system to move
the two carriers, with the combined length thus enabling along desired paths as the two carriers form into and
the transportation of structural objects as large as those maintain a single straight-bed carrier configuration or a
transported by a conventional carrier of that length; (2) single V-bed carrier configuration by the control inputs,
formation of a single V-bed carrier configuration, thus the moving velocity of vehicle-1 and the angular velocity
enabling the transportation of structural objects for which of its steering system, the angular velocity of the steering
stable transportation is not possible with conventional system between carriers-1 and -2, and the angular velocity
carriers; and (3) cooperative operation by the two car-like of the steering system of vehicle-2.
mobile robots with drive wheels, thus enabling generation
y
and delivery of greater driving force. In summary, the pro-
posed coupled-vehicle system facilitates the transportation
of structural objects with a carrier whose configuration can
be adjusted for the shapes of the objects and increases joint-1
vehicle-1
driving force.
We show that the kinematical equation of the proposed joint-2 carrier-1
coupled-vehicle system can be converted into a chained
form which is a canonical form in a new coordinate system joint-3 steering
where an arbitrary path whose curvature is two times carrier-2
differentiable is one coordinate axis and a straight line
perpendicularly intersecting the tangent of the path is vehicle-2
x
the other coordinate axis. Especially, to permit this con- 0
version, we assume a ”virtual vehicle” as a third car-
like mobile robot in Fig.2 whose virtual rear wheels are Fig. 1. A Five-axle, Three-steering Coupled-vehicle System
equivalent to the passive wheels of the steering system at
the passive revolute joint connecting the two carriers, and 3. KINEMATICAL EQUATION OF
we incorporate the time derivatives of the state variables COUPLED-VEHICLE SYSTEM WITH VIRTUAL
of the virtual vehicle into the kinematical equation of the MECHANICAL ELEMENTS
coupled-vehicle system.
We then propose a path-following feedback control We derive, in this section, the kinematical equation of
method for this coupled-vehicle system, based on a chained the coupled-vehicle system, including the time derivatives
form, to control the orientations of the first and second of the state variables of the virtual mechanical elements
carriers (”carrier-1” and ”carrier-2”) relative to the direc- which are added to the system as shown in Fig.2 to
tion of the tangent of the path as the midpoint on the facilitate the conversion of the kinematical equation into a
rear axle of the leading car-like mobile robot (”vehicle- chained form which is a canonical form. As shown in Fig.2,
1”) moves along the path. With this feedback control we assume a virtual car-like mobile robot (thus, a ”virtual
method, the coupled-vehicle system follows desired paths vehicle”) whose rear wheels are equivalent to the passive
as the relative orientations of carriers-1 and -2 converge wheels of the steering system in the same position of joint-
to and maintain their desired values. Using this method, 2. Then, the kinematical equation of the coupled-vehicle
system is derived as:
it is possible to maintain the constant relative orientation ⎧
between the two carriers in either a straight-bed carrier ⎪ ẋ1 = u1 cos θ1
⎪
⎪
configuration or a V-bed carrier configuration and perform ⎪
⎪ ẏ 1 = u1 sin θ1
⎪
⎪
the path-following operation with the chosen configuration ⎪
⎪ 1 = u2
φ̇
⎪
⎪
in which the two carriers function as a single carrier. The ⎪
⎪ tan φ1
⎪
⎪ θ̇1 = u1
validity of this control method is verified by an experimen- ⎪
⎪ l1
⎪
⎪
tal apparatus of the coupled-vehicle system developed by ⎪
⎪ φ̇2 = u3
⎪
⎪
the authors. ⎨ θ̇ = sin(θ1 − θ3 ) u
2 1
l2 cos(θ2 − θ3 ) (1)
2. STRUCTURE OF COUPLED-VEHICLE SYSTEM ⎪
⎪
⎪
⎪ φ̇ 3 = u 4
⎪
⎪ tan φ2 cos(θ1 − θ2 )
The basic structure of the five-axle, three-steering coupled- ⎪
⎪ θ̇3 =
⎪
⎪ u1
vehicle system is shown in Fig.1. The first car-like mobile ⎪
⎪ l3 cos(θ2 − θ3 )
⎪
⎪ cos(θ1 − θ2 ) sin(θ3 − θ5 )
robot (”vehicle-1”) is coupled via a first passive revolute ⎪
⎪ θ̇4 =
⎪
⎪ u1
joint (”joint-1”) at the midpoint of the vehicle-1 rear axle ⎪
⎪ l4 cos(θ2 − θ3 ) cos(θ4 − θ5 )
⎪
⎪ − tan φ3 cos(θ1 − θ2 ) cos(θ3 − θ4 )
to the first carrier (”carrier-1”). Carrier-1 is coupled to the ⎪
⎩ θ̇5 = u1 .
second carrier (”carrier-2”) via a second passive revolute l5 cos(θ2 − θ3 ) cos(θ4 − θ5 )

12977
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

As understood from Fig.2, a vector (x1 , y1 )T represents the
position of the midpoint of the rear axle of vehicle-1; angles
φ1 and θ1 represent the steering angle and orientation
of vehicle-1, respectively; an angle θ2 the orientation
of carrier-1; angles φ2 and θ3 the steering angle and
orientation of the virtual vehicle, respectively; an angle
θ4 the orientation of carrier-2; and angles φ3 and θ5 the
steering angle and orientation of vehicle-2, respectively.
y dθt
φ1
vehicle-1
y1 θ1
virtual-vehicle φ2 l1
θ2
l3
d Ps , as:
θ3 l2
vehicle-2 carrier-1
φ3
θ4 Ps
θ5
l4
l5 carrier-2
0 x1
Fig. 2. A Five-axle, Three-steering Coupled-vehicle System
with Virtual Mechanical Elements
4. PATH-FOLLOWING CONTROL STRATEGY
We define the path-following operation of the coupled-
vehicle system and propose a control strategy to achieve
this operation.
4.1 Path-following Operation
We define the path-following operation of the coupled-
vehicle system as the operation of causing carriers-1 and -2
to move along a path while controlling their orientations
in accordance with the direction of the tangent of the
path. Strictly speaking, we define the operation as that
of causing the midpoint of the rear axle of vehicle-1 to
move along the path while maintaining the orientations of
carriers-1 and -2 relative to the direction of the tangent of
the path at desired values.
When the midpoint of the rear axle of vehicle-1 is off the
path, we define a point Ps on the path which determines
the position of that midpoint relative to the path as
follows. The point Ps is the point on the path which
satisfies the condition of the perpendicular intersection
between the tangent of the path at that point and the
straight line through that point to the midpoint of the
rear axle of vehicle-1. If two or more such points exist on
the path, the one closest to the midpoint of the rear axle
of vehicle-1 is chosen. As indicated in Fig.2, the position
of the point Ps on the coordinate axis defined along the
path is designated by s. The position of the midpoint of the
rear axle of vehicle-1 relative to the point Ps on the other
coordinate axis defined along the straight line through the
point Ps to the midpoint is designated by d. The variable
d is taken as positive in sign when the midpoint of the
rear axle of vehicle-1 is to the left of the point Ps when
looking in the positive direction of the coordinate axis s,
negative in sign when it is to the right of the point Ps ,
and zero when it is on the path. The direction of the
tangent of the path at the point Ps relative to the x-axis is
designated by θt . The orientations of vehicle-1, carrier-1,
the virtual vehicle, carrier-2, and vehicle-2 relative to the
tangent of the path at the point Ps are designated by θpi ,
i = 1, 2, · · · , 5, and these are given as:
θpi = θi − θt , i = 1, 2, · · · , 5. (2)
The derivative of the angle θt with respect to s is given as:
c(s) = . (3)
ds
s
The derivative c(s) is the curvature of the path at the point
Ps and the inverse of its absolute value is the radius of
curvature. The time derivative of the angle θt is therefore
represented in terms of ṡ, the moving velocity of the point
θt
dθt ds
θ̇t = = c(s)ṡ. (4)
ds dt
x
4.2 Control Strategy
We propose, in this subsection, a new control strategy
to achieve the path-following operation of the coupled-
vehicle system. As the coupled-vehicle system is moved
and thus as s is increased or decreased, vehicle-1, vehicle-
2 and the virtual vehicle are steered so that: the position d
of the midpoint of the rear axle of vehicle-1 relative to the
point Ps and the orientation θp1 of vehicle-1 relative to the
direction of the tangent of the path at the point Ps both
converge to zero; the orientation θp2 of carrier-1 relative
to the direction of the tangent of the path at the point Ps
converges to the desired value θp2d ; and the orientation θp4
of carrier-2 relative to the direction of the tangent of the
path at the point Ps converges to the desired value θp4d .
Thereafter, the steering angles of vehicle-1, vehicle-2, and
the virtual vehicle are determined in accordance with the
path curvature c(s) and its first-order derivative c (s), to
maintain d and θp1 at zero and θp2 and θp4 at θp2d and θp4d ,
respectively. In this control strategy, the vector x below is
taken as the state variable vector of the coupled-vehicle
system.
T
x = (s, d, φ1 , θp1 , φ2 , θp2 , φ3 , θp3 , θp4 , θp5 ) . (5)
The time derivatives of s and d, which are the components
of the vector x, are given as:

ṡ = u1 cos θp1 + dθ̇t
(6)
d˙ = u1 sin θp1 .
The time derivatives of s and θt are derived from Eqs.(4)
and (6) as:
⎧
⎪
⎪ cos θt
⎨ ṡ = u1
1 − dc(s)
(7)
⎪
⎪ c(s) cos θp1
⎩ tθ̇ = u 1 .
1 − dc(s)
The time derivative of the state variable vector x of Eq.(5)
is derived from Eqs.(1), (2), (6), and (7) as:
ẋ = g1 (x)u1 + g2 (x)u2 + g3 (x)u3 + g4 (x)u4 , (8)

12978
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

⎛ cos θp1
⎞
The time derivative of Eq.(12) thus becomes the three-
1−dc(s)
⎜ ⎟ chain, single-generator chained form shown in Eq.(14).
⎜ sin θp1 ⎟
⎜ ⎟
⎜ 0 ⎟ ż11 = w1 , żi1 = wi ,
⎜ tan φ1
−
c(s) cos θp1 ⎟ żi2 = zi1 w1 ,
⎜ l1 1−dc(s) ⎟ (14)
⎜ ⎟ żi3 = zi2 w1 ,
⎜ 0 ⎟
g1 = ⎜
⎜
sin(θp1 −θp3 ) c(s) cos θp1
l2 cos(θp2 −θp3 ) − 1−dc(s)
⎟,
⎟
i = 2, 3, 4.
⎜ ⎟
⎜ 0 ⎟ w1 = ũ1 ,
⎜ tan φ2 cos(θp1 −θp2 ) c(s) cos θp1 ⎟
l3 cos(θp2 −θp3 ) − 1−dc(s) 4
⎜ ⎟
⎜
⎜ cos(θp1 −θp2 ) sin(θp3 −θp5 ) c(s) cos θp1 ⎟
⎟ wj = Lfi L2f1 hj ũi ,
⎝ l4 cos(θp2 −θp3 ) cos(θp4 −θp5 ) − 1−dc(s) ⎠ i=1
− tan φ3 cos(θp1 −θp2 ) cos(θp3 −θp4 ) c(s) cos θp1 j = 2, 3, 4.
l5 cos(θp2 −θp3 ) cos(θp4 −θp5 ) − 1−dc(s)
T In Eq.(14), each of the terms L3f1 h2 ũ1 , L3f1 h3 ũ1 , and
g2 = (0, 0, 1, 0, 0, 0, 0, 0, 0, 0) ,
L3f1 h4 ũ1 , which are the first terms on the right hand sides
T
g3 = (0, 0, 0, 0, 1, 0, 0, 0, 0, 0) , of the equations of w2 , w3 , and w4 respectively, contains
g4 = (0, 0, 0, 0, 0, 0, 1, 0, 0, 0) .
T the second-order derivative c (s) of the curvature of the
path. For calculation of the coupled-vehicle system control
inputs, the curvature of the path to be followed must be
4.3 Conversion into Chained Form
therefore two times differentiable.
This conversion is effective in the following open subset
We show, in this subsection, that Eq.(8), the kinematical U as:
equation of the coupled-vehicle system performing the
path following operation, can be converted into a chained U = {x : φ1 = ±π/2, φ2 = ±π/2, φ3 = ±π/2,
form which is a canonical form, by showing the calculation θp1 = ±π/2, θp2 − θp3 = ±π/2,
results. This conversion is based on differential geometry θp4 − θp5 = ±π/2; dc(s) < 1} . (15)
(Isidori (1989)). The state variable vector x is limited to the open subset
We first convert the four vector fields g1 , g2 , g3 , and g4 U in order to avoid singular attitudes in which the kine-
of Eq.(8) as: matical equation of the coupled-vehicle system performing
⎧ the path-following operation cannot be converted into the
⎪ f = g1 (1 − dc(s)) / cos θp1
⎨ 1 three-chain, single-generator chained form.
f2 = g2
(9)
⎪
⎩ f3 = g3
f4 = g4 . 4.4 Desired Values of Converted Variables in Path-following
Operation
We secondly convert the control inputs u1 , u2 , u3 , and u4
of Eq.(8) as: We show, in this subsection, the desired values for the
⎧
⎪ ũ = u1 cos θp1 / (1 − dc(s)) convergence of the converted variables shown in Eq.(12)
⎨ 1
ũ2 = u2 and thus for the path-following operation of the coupled-
(10)
⎪ ũ3 = u3
⎩ vehicle system.
ũ4 = u4 . The variable z23 is d, the variable z33 is θp2 , and the
The kinematical equation of Eq.(8) is then rewritten as: variable z43 is θp4 , and, as understood from Eqs.(12) and
(14), the following relationships about d, θp2 , and θp4 are
ẋ = f1 ũ1 + f2 ũ2 + f3 ũ3 + f4 ũ4 . (11) satisfied.
⎧
Using the four vector fields f1 , f2 , f3 , and f4 , the variables ⎨ d = z23
of Eq.(8) are converted as: d˙ = z22 ũ1 (16)
⎧ ⎩¨
⎪ z11 = h1 = s d = z21 ũ21 + z22 ũ˙ 1 ,
⎪
⎪ ⎧
⎪
⎪ z21 = L2f1 h2
⎪
⎪ ⎨ θp2 = z33
⎪
⎪ z22 = Lf1 h2
⎪
⎪ θ̇p2 = z32 ũ1 (17)
⎪
⎪ z = h2 = d ⎩
⎨ 23 θ̈p2 = z31 ũ21 + z32 ũ˙ 1 ,
z31 = L2f1 h3 ⎧
(12)
⎪
⎪ z32 = Lf1 h3 ⎨ θp4 = z43
⎪
⎪
⎪
⎪ z33 = h3 = θp2 θ̇p4 = z42 ũ1 (18)
⎪
⎪ ⎩
⎪
⎪ z41 = L2f1 h4 θ̈p4 = z41 ũ21 + z42 ũ˙ 1 .
⎪
⎪
⎪ z = Lf1 h4
⎩ 42 In the path-following operation, while moving the coupled-
z43 = h4 = θp4 . vehicle system, i.e., while maintaining at non-zero the
moving velocity ũ1 of the point Ps which represents the
Here, as shown in Eq.(13), Lp q is the inner product of position of the midpoint of the rear axle of vehicle-1 on
the vector (∂q/∂x1 , ∂q/∂x2 , · · · , ∂q/∂xn )T , which consists the path, it is necessary to achieve the convergence of
of the components obtained by partial differentiation of a the position d of the midpoint of the rear axle of vehicle-
scalar quantity q with state variables xi , i = 1, 2, · · · , n 1 relative to the path to the desired value of zero, the
and a vector p = (p1 , p2 , · · · , pn )T . orientation θp2 of carrier-1 relative to the direction of the
∂q ∂q ∂q tangent of the path at the point Ps to the desired value
Lp q = p1 + p2 + · · · + pn . (13) θp2d (=const), and the orientation θp4 of carrier-2 relative
∂x1 ∂x2 ∂xn

12979
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

to the direction of the tangent of the path at the point
Ps to the desired value θp4d (=const). As understood from
Eqs.(16), (17), and (18), it is therefore necessary to achieve
the convergence of the variables z21 , z22 and z23 to zero,
the variables z31 and z32 to zero and the variable z33 to
θp2d , and also the convergence of the variables z41 and
z42 to zero and the variable z43 to θp4d . In other words,
the desired values of the variables z21 , z22 , and z23 are
respectively zero, zero, and zero, those of the variables z31 ,
z32 , and z33 are respectively zero, zero, and θp2d , and those
of the variables z41 , z42 , and z43 are respectively zero, zero,
and θp4d .
5. PATH-FOLLOWING FEEDBACK CONTROL
METHOD
6. TRANSPORTATION BY STRAIGHT-BED AND
V-BED CARRIERS
We propose, in this section, methods of operation of the
coupled-vehicle system for transportation with carriers-1
and -2 for causing the two carriers to be (a) in a straight-
line-shaped configuration as a straight-bed carrier; and (b)
in a V-shaped configuration as a V-bed carrier. Figs.3
and 4 illustrate transportation by the straight-bed car-
rier formed by causing the relative orientation between
carriers-1 and -2 to be an angle of zero degrees and by the
V-bed carrier formed by causing the relative orientation
between them to be an obtuse angle. Of course, it is
possible to make it an acute angle.
path
joint-1

We propose, in this section, a method of feedback control joint-2 vehicle-1

for the path-following operation of the coupled-vehicle carrier-1
system based on the chained form. With this method, joint-3 steering
we achieve the path-following operation while moving the carrier-2
coupled-vehicle system, i.e., while increasing or decreasing
vehicle-2
the variable z11 (which is the position s of the point Ps
on the path), by obtaining the convergence of the variable Fig. 3. A Straight-bed Carrier
z33 (which is the orientation θp2 of carrier-1 relative to the
y
direction of the tangent of the path at the point Ps ) to
its desired value θp2d , the convergence of the variable z43
(which is the orientation θp4 of carrier-2 relative to the joint-3 vehicle-2
direction of the tangent of the path at the point Ps ) to its carrier-2
desired value θp4d , and the convergence of all of the other
joint-2
variables (which are z21 , z22 , z23 , z31 , z32 , z41 , and z42 )
steering circle path path
to zero. We thus propose the following feedback control
method as:
⎧ x
⎪ w1 = a0 carrier-1
⎪
⎪ z22 z23 joint-1 vehicle-1
⎪
⎪
⎪
⎪ w2 = p21 z21 + p22 + p23 2
⎨ a0 a0
z32 z33 − θp2d (19) Fig. 4. A V-bed Carrier
⎪ w3 = p31 z31 + p32 + p33
⎪
⎪ a 0 a 2
⎪
⎪
0
z43 − θp4d
⎪
⎪ z42 6.1 Transportation by Straight-bed Carrier
⎩ w4 = p41 z41 + p42 + p43 .
a0 a20
For formation of carriers-1 and -2 into a straight-bed
In Eq.(19), a0 is a non-zero constant value, and physically carrier configuration and transportation of an object in
means the moving velocity ṡ of the point Ps on the path. that configuration, as illustrated in Fig.3, we propose the
Here, when we define the three-dimensional vectors ζi , following procedures.
i = 2, 3, 4 as shown in Eq.(20), their time derivatives ζ̇i , Procedure 1-1: Select an arbitrary curve, e.g., a Bezier
i = 2, 3, 4 are given as shown in Eq.(21). curve, as a path to be followed.
⎧  T Procedure 1-2: Set θp2d =θp4d , i.e., set θp2d and θp4d to a
⎪
⎪ z22 z23
⎪ ζ2 = z21 ,
⎪ , single value, as the desired value of both θp2 and θp4 repre-
⎪
⎪ a0 a20 senting the orientations of carriers-1 and -2, respectively,
⎪
⎨  T
z32 z33 − θp2d relative to the direction of the tangent of the path at the
ζ3 = z31 , , (20)
⎪
⎪ a0 a20 point Ps .
⎪
⎪   Procedure 1-3: While moving the coupled-vehicle sys-
⎪
⎪ z42 z43 − θp4d
T
⎪
⎩ ζ4 = z41 , . . tem in forth-and-back motion along the path, obtain the
a0 a20 convergence of d to zero, θp2 to θp2d , and θp4 to θp4d by
  performing a path-following operation. More specifically,
pi1 pi2 pi3
ζ̇i = 1 0 0 ζi = Λi ζi , i = 2, 3, 4. (21) move the point Ps (which represents the position of the
0 1 0 midpoint of the rear axle of vehicle-1 on the path) through
the distance sL and then through the distance −sL . In
When pi1 , pi2 , pi3 , i = 2, 3, 4 are given so as to obtain other words, move the point Ps in forth-and-back motion
negative values for the real parts of the eigenvalues of the along the path through the distance 2sL .
matrices Λi , i = 2, 3, 4 of Eq.(21), the vectors ζi , i = 2, 3, 4 Procedure 1-4: Repeat Procedures 1-1 to 1-3 sufficiently
exponentially converge to zero, resulting in achieving the to ensure the convergence of d to zero, θp2 to θp2d , and θp4
path-following operation. to θp4d .

12980
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011
Procedure 1-5: Load an object onto the straight-bed
carrier, and perform the path-following operation.
6.2 Transportation by V-bed Carrier
For formation of carriers-1 and -2 into a V-bed carrier
configuration and transportation of an object in that con-
figuration, as illustrated in Fig.4, we propose the following
procedures.
Procedure 2-1: Set a circular path of a radius R with a
center at (0, R)T such that when the midpoints of the rear
axles of vehicles-1 and -2 lie on the path and when the
straight line between these two midpoints passes through
the center of the circular path the relative orientation
between carriers-1 and -2 becomes the desired value θpc .
Place the starting point of the circular path at the origin
(0, 0)T of the Cartesian coordinate system, and take the
counterclockwise direction on the path as the positive
direction of the coordinate axis s.
Procedure 2-2: With the midpoints of the rear axles of
vehicles-1 and -2 on the circular path and with the line
between these midpoints passing through the center of the
circular path, set the desired orientations θp2d and θp4d of
carriers-1 and -2 relative to the direction of the tangent
of the path at the point Ps so as to cause the relative
orientation between the two carriers to be θpc , i.e., so as
to satisfy the geometrical condition θpc =θp2d −θp4d .
Procedure 2-3: With the origin (0, 0)T of the Cartesian
coordinate system as a starting point, set a path to be
followed as a free-form curve (e.g., a Bezier curve) such
that the direction of the tangent of the path at the origin
corresponds to the direction of the x-axis and the path
curvature as well as its first- and second-order derivatives
at the origin are all equal to zero.
Procedure 2-4: Obtain the convergence of d to zero, θp2
to θp2d , and θp4 to θp4d while moving the coupled-vehicle
system along the circular path by the path-following
operation. More specifically, move the point Ps which
represents the position of the midpoint of the rear axle
of vehicle-1 on the path from the origin (0, 0)T of the
Cartesian coordinate system through the distance 2πR
thus through one round on the circular path in either
the clockwise or the counterclockwise direction. When the
point Ps is not initially at the origin, move it to the origin
before beginning the single round on the circular path.
Procedure 2-5: Repeat Procedures 2-1 to 2-4 sufficiently
to ensure the convergence of d to zero, θp2 to θp2d , and θp4
to θp4d . The orientations of vehicle-1 and vehicle-2 then
correspond to the directions of the tangents of the path at
the midpoints of their rear axles, and they are parallel to
each other.
Procedure 2-6: To cause the coupled-vehicle system to
follow the path which is a free-form curve while main-
taining the relationships d=0, θp2 =θp2d , and θp4 =θp4d , set
all the steering angles so that the steering directions of
vehicle-1, the steering system between carriers-1 and -2,
and vehicle-2 all correspond to the direction of the x-axis,
since the direction of the tangent of the path at its starting
point corresponds to the direction of the x-axis, the path
curvature and its first-order and second-order derivatives
are all equal to zero.
Procedure 2-7: Load an object onto the V-bed carrier,
and perform the path-following operation.
12981
7. EXPERIMENTAL SYSTEM
We describe, in this section, the experimental system
developed to verify the validity of the path-following
feedback control method for the coupled-vehicle system.
Figs.5 and 6 show the experimental coupled-vehicle system
and the layout of the visual feedback control system
developed for this purpose, respectively.
Vehicles-1 and -2 are each operated by two-wheel rear
drive and two-wheel front steering. The two front wheels of
each vehicle are independently steered and the extensions
of the axles of its two front wheels and that of its rear
wheels intersect at a single point which is the center of
its revolution. The two rear wheels are driven through a
differential gear so that in rounding a curve the difference
in rotational speed between the inner and outer wheels
can be absorbed mechanically. These mechanical features
enable operation of each of the vehicles to be in strict
accordance with its kinematical equation. In each vehicle,
each of the two front wheels is steered by a DC servomotor
with a zero-backlash gear head, and the two rear wheels
are driven by a DC servomotor with a gear head. The
rotational angle of each motor is measured by a rotary
encoder.
Rotary encoders are also used for measurement of the
rotational angles of the three passive revolute joints: joint-
1 between vehicle-1 and carrier-1, joint-2 between carriers-
1 and -2, and joint-3 between carrier-2 and vehicle-2. In
the steering system between the two carriers, the rolling
direction of its passive wheels can be actively changed
relative to the orientation of carrier-1. The passive wheels
are steered by a DC servomotor with a zero-backlash
gear head. The rotational angle of this servomotor is also
measured by a rotary encoder.
The weight of the experimental coupled-vehicle system
is 9.1[kg]. As shown in Fig.6, the positions of the two black
circular marks on the top of vehicle-1 are measured by
a CCD camera installed on the ceiling 3.0[m] above the
experimental field 3.0[m] in diameter, and together with
the measurement of the rotational angles of joints-1, -2,
and -3, which makes it possible to determine the position
and orientation of the coupled-vehicle system. The two
identical marks are 120[mm] in diameter and located
219[mm] above the experimental field. The measurement
error for the position of the center of mass of each circular
mark is 3.0[mm] or less. The control period for the path-
following feedback control system in this experiment is
33[ms].
8. EXPERIMENTAL VERIFICATION
We describe, in this section, the verification of the validity
of the path-following feedback control method for the
coupled-vehicle system, by transportation of an object
by the experimental coupled-vehicle system in the V-bed
carrier configuration. The design parameters determining
the size of the experimental coupled-vehicle system are as:

l1 = 0.27[m], l2 = 0.405[m], l3 = 0.27[m],
l4 = 0.405[m], l5 = 0.27[m].
In this experiment, as shown in Fig.7, a circle of radius
R=0.2025[m] with a center at (0, R)T is taken as the path
for formation of the V-bed carrier by carriers-1 and -2,
and a 10th-order Bezier curve with a starting point at
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

mark
carrier-2 carrier-1

mark

steering

vehicle-2 vehicle-1

Fig. 5. An Experimental Five-axle, Three-steering Coupled-vehicle System

y [m]
CCD camera
1.5
analog image signal

P6 P7 P8 P9 P10
1

PC
circular path
0.5 Bezier-curve path
optical signal
P5

0
P0 P1 P2 P3 P4
y

x
-0.5 x [m]
-0.5 0 0.5 1 1.5 2 2.5

Fig. 6. A Visual Feedback Control System Fig. 7. Circular and Bezier-curve Paths
the origin (0, 0)T of the Cartesian coordinate system is with the eigenvalues of the matrices Λi , i = 2, 3, 4 of
taken as the transportation path. The control points of Eq.(21) all equal to −0.14.
the Bezier curve are located as: For the Bezier-curve path-following, to obtain the
⎧
⎪ P0 =(0.0[m], 0.0[m])T , P1 =(0.2[m], 0.0[m])T , movement of the point Ps on the path through the distance
⎪
⎪
⎪
⎪ T T of Lb =2.3513[m] from its starting point P0 to its ending
⎪
⎪ P2 =(0.4[m], 0.0[m]) , P3 =(0.6[m], 0.0[m]) ,
⎨ T T point P10 in the time of T2 =45[s], the moving velocity ṡ
P4 =(0.8[m], 0.0[m]) , P5 =(1.0[m], 0.5[m]) , of the point Ps is given as:
⎪
⎪
T T
P6 =(1.2[m], 1.0[m]) , P7 =(1.4[m], 1.0[m]) ,
⎪
⎪ ṡ = a0 = Lb /T2 = 0.05225[m/s].
⎪
⎪
T T
P8 =(1.6[m], 1.0[m]) , P9 =(1.8[m], 1.0[m]) ,
⎪
⎩ T For this Bezier-curve path-following, the coefficients in the
P10 =(2.0[m], 1.0[m]) .
control inputs of Eq.(19) are given as:
The desired values θp2d and θp4d of the orientations θp2 
and θp4 of carriers-1 and -2, respectively, relative to the p21 = −1.2, p22 = −0.48, p23 = −0.064
direction of the tangent of the path at the point Ps are p31 = −1.2, p32 = −0.48, p33 = −0.064
given as: p41 = −1.2, p42 = −0.48, p43 = −0.064.
 with the eigenvalues of the matrices Λi , i = 2, 3, 4 of
π 5π
θp2d = − [rad], θp4d = − [rad]. Eq.(21) all equal to −0.4.
6 6
The initial state variables of the coupled-vehicle system
To obtain one complete round of the point Ps on the are given as:
circular path of the length Lc =1.2723[m] in the time ⎧
T1 =60.0[s], the moving velocity of the point Ps on the ⎪
⎪ x1 |t=0 = 0.0[m], y1 |t=0 = 0.0[m],
⎪
⎨ φ1 |t=0 = 0.0[rad], θ1 |t=0 = 0.0[rad],
path is given as:
φ2 |t=0 = 0.0[rad], θ2 |t=0 = 0.0[rad],
⎪
⎪
⎩ φ3 |t=0 = 0.0[rad], θ3 |t=0 = 0.0[rad],
ṡ = a0 = Lc /T1 = 0.02121[m/s]. ⎪
For the circular path-following, the coefficients in the θ4 |t=0 = 0.0[rad], θ5 |t=0 = 0.0[rad].
control inputs of Eq.(19) are given as: The initial position of the point Ps on the Bezier-curve

p21 = −0.42, p22 = −0.0588, p23 = −0.002744 path is thus the same position of the starting point of
p31 = −0.42, p32 = −0.0588, p33 = −0.002744 the circular path at the origin (0, 0)T of the Cartesian
p41 = −0.42, p42 = −0.0588, p43 = −0.002744. coordinate system, and in the s-d coordinate system the

12982
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011
initial position (s, d)T |t=0 of the midpoint of the rear axle
of vehicle-1 and the initial orientations θp1 |t=0 of vehicle-1,
θp2 |t=0 of carrier-1, θp3 |t=0 of the virtual vehicle, θp4 |t=0 of
carrier-2, and θp5 |t=0 of vehicle-2 relative to the direction
of the tangent of the path at the point Ps are given as:
⎧ of objects to be transported. The validity of the path-
⎪ s|
⎨ t=0
= 0.0[m], d|t=0 = 0.0[m],
θp1 |t=0 = 0.0[rad], θp2 |t=0 = 0.0[rad],
⎩ θp3 |t=0 = 0.0[rad], θp4 |t=0 = 0.0[rad],
⎪
θp5 |t=0 = 0.0[rad].
Fig.8 shows the formation of the V-bed carrier config-
uration as the coupled-vehicle system follows the circu-
lar path. As vehicle-1 performs the two counterclockwise
rounds of the circular path while maintaining d at zero, the
orientations θp2 of carrier-1 and θp4 of carrier-2 converge
to θp2d and θp4d , respectively, thus together forming the
V-bed carrier configuration with the relative orientation
between carriers-1 and -2 converging to the desired value
θpc (=θp2d −θp4d ). After the two counterclockwise rounds
of the circular path, the orientations of vehicles-1 and -2
correspond to the direction of the x-axis of the Cartesian
coordinate system, and the direction of the steering system
between carriers-1 and -2 corresponds to the direction of
the y-axis.
Following the formation of the V-bed carrier config-
uration by carriers-1 and -2, the coupled-vehicle system
performs the transportation of an object by following the
10th-order Bezier-curve path with the origin (0, 0)T of
the Cartesian coordinate system as the starting point of
the path. In preparation for the transportation operation,
the directions of the steering systems of vehicle-1, -2,
and between carriers-1 and -2 are all initially caused to
correspond to the direction of the x-axis of the Cartesian
coordinate system. This is because the 10th-order Bezier-
curve path is planned so that the direction of the tangent
of the path at the origin (0, 0)T of the Cartesian coordinate
system which is the starting point of the path corresponds
to the direction of the x-axis and so that the path cur-
vature and its first- and second-order derivatives at the
origin are all equal to zero. An object weighting 3.0[kg] is
next loaded onto the V-bed carrier, and the experimental
coupled-vehicle system is then ready to begin the trans-
portation operation. As shown in Fig.9, the experimental
coupled-vehicle system performs the transportation of the
object while maintaining the V-bed carrier configuration
and following the 10th-order Bezier-curve path, thus suc-
cessfully executing entry into a garage with very tight
clearances and verifying the validity of the path-following
feedback control method for the coupled-vehicle system.
9. CONCLUSIONS
We have proposed, in this paper, a five-axle, three-steering
coupled-vehicle system and a path-following feedback con-
trol method for the system. We have shown that the
kinematical equation of the coupled-vehicle system can be
converted into a chained form which is a canonical form
by assuming the existence of virtual mechanical elements
in a new coordinate system where an arbitrary path whose
curvature is two times differentiable is one coordinate axis
and a straight line perpendicularly intersecting the tangent
of the path is the other coordinate axis. We have proposed
a path-following feedback control method based on the
chained form. With this method it is possible to form the
12983
two carriers into either a straight-bed carrier configuration
or a V-bed carrier configuration, and to perform the path-
following operation with either configuration as though
the two carriers constitue a single carrier, thus enabling
the carriers to be configured in accordance with shapes
following feedback control method is verified by the exper-
imental coupled-vehicle system performing transportation
of an object in a garage entry operation with very tight
clearances.
REFERENCES
Sampei, M., Tamura, T., Kobayashi, T., and Shibui, N.
(1995). Arbitrary Path Tracking Control of Articulated
Vehicles using Nonlinear Control Theory. IEEE Trans-
actions on Control Systems Technology, 3(1), 125–131.
Samson, C. (1995). Control of Chained Systems: Ap-
plication to Path Following and Time-varying Point-
stabilization of Mobile Robots. IEEE Transactions on
Automatic Control, 40(1), 64–77.
Altafini, C. (2001). Some Properties of The General N-
Trailer. International Journal of Control, 74(4), 409–
424.
Lizarraga, D.A., Morin, P., and Samson, C. (2001).
Chained Form Approximation of a Driftless System,
Application to the Exponential Stabilization of the Gen-
eral N-Trailer System. International Journal of Control,
74(16), 1612–1629.
Deligiannis, V., Davrazos, G., Menesis, S., and Aram-
patzis, T. (2006). Flatness Conservation in the N-Trailer
System Equipped with a Sliding Kingpin Mechanism.
Journal of Intelligent and Robotic Systems, 46(2), 151–
162.
Bushnell, L.G., Tilbury, D.M., and Sastry, S.S. (1995).
Steering Three-Input Nonholonomic Systems: the Fire
Truck Example. The International Journal of Robotics
Research, 14(4), 366–381.
Tilbury, D.M., Sordalen, O.J., Bushnell, L.G., and Sastry,
S.S. (1995). A Multisteering Trailer System: Conversion
into Chained Form Using Dynamic Feedback. IEEE
Transactions on Robotics and Automation, 11(6), 807–
818.
Tilbury, D.M. and Sastry, S.S. (1995). The Multi-Steering
N-Trailer System: A Case Study of Goursat Normal
Forms and Prolongations. International Journal of
Robust and Nonlinear Control, 5(4), 343–364.
Nakamura, Y., Ezaki, H., Yuegang Tan, and Woojin
Chung (2001). Design of Steering Mechanism and Con-
trol of Nonholonomic Trailer Systems. IEEE Transac-
tions on Robotics and Automation, 17(3), 367–374.
Lamiraux, F., Laumond, J.-P., Van Geem, C., Boutonnet,
D., and Raust, G. (2005). Trailer Truck Trajectory
Optimization: the Transportation of Components for
the Airbus A380. IEEE Robotics and Automation
Magazine, 12(1), 14–21.
Yamaguchi, H., Mori, M., and Kawakami, A. (2010). A
Path Following Feedback Control Method Using Para-
metric Curves for a Cooperative Transportation System
with Two Car-like Mobile Robots. Proceedings of the 7th
IFAC Symposium on Intelligent Autonomous Vehicles,
Lecce, Italy, September.
Isidori, A. (1989). Nonlinear Control Systems. New York:
Springer-Verlag, Second Edition.
18th IFAC World Congress (IFAC'11)
Milano (Italy) August 28 - September 2, 2011

t=0.0[s] garage t=40.0[s] garage

(a) (b)

t=80.0[s] garage t=120.0[s] garage

(c) (d)
Fig. 8. Formation of a V-bed Carrier Configuration

t=0.0-45.0[s] garage

t=45.0[s]
object

t=0.0[s]
Fig. 9. Transportation of an Object by a V-bed Carrier

12984