Journal of Intelligent and Robotic Systems 37: 399414, 2003.

2003 Kluwer Academic Publishers. Printed in the Netherlands.

399

On the Suppression of Off-tracking in

Multi-articulated Vehicles through a Movable
Junction Technique
S. MANESIS, N. T. KOUSSOULAS and G. N. DAVRAZOS
Division of Systems and Automatic Control, Department of Electrical & Computer Engineering,
University of Patras, Rion 26500, Greece; e-mail: stam.manesis@ee.upatras.gr
(Received: 18 November 2002; in final form: 27 May 2003)
Abstract. The motion of a multi-articulated robotic vehicle as well as of a train-like vehicle is
characterized by the deviation of the path of each intermediate vehicle from that of the leading one
(off-tracking phenomenon). In this paper, we propose the use of an innovative movable junction,
which allows the kingpin to slide along the rear axis of the pulling vehicle, a technique that proved to
be efficient in reducing or completely eliminating off-tracking. New kinematic equations are derived
and a nonlinear controller is analytically developed, based on the steady-state off-tracking behavior
of the n-trailer system. Simulations and a comparison study for various cases without/with this
innovative sliding kingpin junction technique showed that its use together with an analytically
derived controller can make possible the elimination of off-tracking.
Key words: n-trailer systems, multi-articulated vehicles, off-tracking, nonlinear controller.

1. Introduction
Multi-articulated (or train-like or multi-body) vehicles or n-trailer systems can be
found in two different research fields: autonomous robotics and transportation systems (referring to vehicles). In autonomous robotics, the goal is to build mobile
multi-body robots that accomplish useful tasks without human intervention while
operating in unknown environments. On the other hand, in intelligent transportation systems the goal is similarly to construct transportation vehicles intelligent
enough to be driven with as less as possible human intervention. In both of the
above areas, one major common problem is the undesired excess in motion due
to off-tracking. This term refers to the deviation of the path of each articulated
vehicle from the paths of preceding vehicles, especially that of the tractors. The
reduction or elimination of off-tracking will result in much improved performance
in terms of safety during turns, cornering, overtaking other especially small cars,
and backtracking.
The motion of the n-trailer system is subject to nonholonomic constraints (rolling without slipping) so it has been studied as a class of nonholonomic systems by
many researchers and has both theoretical and practical interest. The research work

400

S. MANESIS ET AL.

[4] constitutes an excellent survey of the recent advances in control of nonholonomic systems. The main problem that has attracted most of the attention is path
following while only few works consider the off-tracking problem. A closed form
expression for the off-tracking of the rear pivot point of a simple tractortrailer
vehicle can be found in [1] while off-tracking bounds for a car pulling trailers have
been derived in [3]. In [2] the path following problem with reduced off-tracking is
addressed for the n-trailer system. This is achieved by keeping track of the error
distance of each of the middle points of the axles of the vehicle from the path using
different moving frames. In [8] different passive steering mechanisms as well as
control laws are presented for nonholonomic trailer systems. The main focus of
these mechanisms is on reducing the passive tracking error with respect to the
tractors trajectory and little attention was paid to active motion control.
The rest of the paper is organized as follows. Section 2 presents the n-trailer
vehicle model, the off-tracking problem and its consequences in the motion of
such vehicles. In Section 3 the description of the innovative junction technique
called sliding kingpin mechanism is given together with the new kinematic state
equations of the resulting n-trailer system. In Section 4, a controller is analytically
designed based on the compensation for the steady-state off-tracking deviation. In
Section 5, simulation results are presented comparing the performance with and
without sliding using alternately an analytically derived nonlinear controller and
a linear one of heuristic origin. The last section ends the paper with concluding
remarks about the results and some open research problems.
2. The Multi-articulated Vehicle
In this section, we describe the kinematic model of the multi-articulated vehicle,
which is suitable for both transportation and robotic vehicles. Furthermore, the
notion of off-tracking and its consequences in the motion of an n-trailer system
is explained analytically for each domain of application. The n-trailer system is
defined as a long and complex vehicle system consisting of a suitable power tractor
(leading vehicle) pulling a number of passive robot bodies or semi-trailers as shown
in Figure 1.
The kinematic equations of the n-trailer system with a two-axle tractor (of
which the front one provides driving and steering) are:
x0 = U1 cos 0 ,
= U2 ,

y0 = U1 sin 0 ,
U1
tan ,
0 =
L

U1
sin(0 1 ),
L
U1
cos(0 1 ) sin(1 2 ),
2 =
L
..
.

1 =

(1)

OFF-TRACKING SUPPRESSION IN MULTI-ARTICULATED VEHICLES

401

Figure 1. Illustration of the multi-articulated vehicle coordinates.

i1


U
1
cos(j 1 j ) sin(i1 i ),
i =
L j =1
where x0 , y0 are the Cartesian coordinates of the leading vehicle and U1 , U2 are the
two control inputs, the linear velocity and the steering angle rate, respectively, [7].
The other state variables represent the orientation angles for each vehicle i, as
shown in Figure 1. The above state equations are derived from the algebraic manipulation of the 2n holonomic constraints:
xi+1 = xi Li cos i+1 ,
yi+1 = yi Li sin i+1

(2)

and the n + 1 nonholonomic constraints

xi sin i yi cos i = 0.

(3)

Off-tracking relates to the question of how much road is needed for the rear
wheels of a vehicle during a turn and can be defined for cars as well as for multiarticulated vehicles, where the interest focuses on the last vehicle where the phenomenon achieves its strongest demonstration. There is a close connection between
off-tracking and the curvature of the followed path, and the sharper the curvature
the more intensive it is. Off-tracking can be easily visualized as the deviation of the
trajectory of a semi-trailers axle center (where the kingpin hitch assumedly lies)
from the path of the axle center of the leading vehicle during a turn. However, when
it comes to modelling and exact quantification for each time instant, the definition
is not unique. We may define the minimal distance between the two trajectories
or the distance from the ideal position of the semi-trailer axle center. This issue is
dealt with in Section 4 below.
In the research field of mobile robotics, the major problems are to find an
obstacle free path and a path following control law. When finding an obstaclefree path for multi-articulated robotic vehicles, we must take into consideration

402

S. MANESIS ET AL.

the presence of off-tracking. The reason is that the last trailer may collide with
obstacles if the vehicle attempts to follow the designed path for the leading vehicle
with off-tracking neglected. One efficient way to solve this problem is to find an
obstacle-free path for the leading vehicle, add a controller for path following and
use another kingpin controller for off-tracking elimination.
In the case of truck-trains it is imperative that the last semi-trailer follows
exactly the path of the (lead) tractor during a turn for lane change or a turn due
to the curvature of the highway. Otherwise, it will be possible for at least the
last semi-trailer to violate the outer boundary of the highway or to crash with an
adjacent car during a lane change even when both tractor and the car keep their
relative velocity within safe limits. It is known that the driver of any long tractor
trailer vehicle because of the off-tracking of the rear wheels turns the tractor far
away into the desired path in order to avoid the unpleasant consequences of this
phenomenon.

3. The Sliding Kingpin System

We will present here the sliding kingpin technique, the realization of which requires a hydraulic (or pneumatic) mechanism. The principle of its operation is as
follows. Off-tracking is eliminated by actively displacing each semi-trailer with
respect to the previous one, a technique first described in [5]. This is realized by
sliding the kingpin hitch in each semi-trailer in a direction perpendicular to the
longitudinal axle (i.e., along the rear axle) of the trailer by a distance Si and in a
direction opposite to the curvature of the path. The sliding distance is expected
to be a function of some quantities related to the motion of the vehicle, most
eminently the steering angle of the leading vehicle. Furthermore, a proportional
type of relation is expected, i.e., the sliding distance will increase at sharp turns
performed by the driver to compensate for expected higher off-tracking. Of course,
the question remains how to determine the exact required magnitude and direction
of the sliding distance. The investigation of this matter and the derivation of a
closed-form equation for the controller can be found in the next section.
A basic assumption in this line of thought is that the leading vehicle does not
perform sharp turns so as the required sliding distance can be kept within reasonable levels. In mobile robotics, this assumption doesnt seem very restrictive
according to our experience while in the case of truck-trains, it is generally satisfied
because of the limited radius of highways curvatures which is not permitted to be
less than 700 m. The introduction of the sliding kingpin system completely relieves
the driver either human or computer from having to perform correcting maneuvers such as forcing the leading vehicle to perform a wider turn than necessary in
an effort to ensure minimal off-tracking for the last trailer.
To proceed we are going to develop a kinematic model for the n-trailer system
after the introduction of the sliding kingpin mechanism. Consider two intermediate
semi-trailers of an n-trailer system as shown in Figure 2.

OFF-TRACKING SUPPRESSION IN MULTI-ARTICULATED VEHICLES

403

Figure 2. The kingpin slides along the axle when the i semi-trailer turns.

The position of each semi-trailer Pi , is taken to be the middle point of the ith
semi-trailers rear axle. Position Pi is defined by the pair (xi , yi ) in the Cartesian
coordinates system while i is the orientation of the ith semi-trailer with respect to
the horizontal axis. It has been pointed out [3] that when the lead car of a single
trailer system is travelling along a circle of radius Rl , then the trailer is travelling
along a circle of radius Rt , with the same center, where Rt < Rl . In order to compensate for this path deviation of the trailer, we suppose that the kingpin hitching
point slides from the point Pi to the point PSi , by a distance Si .
The determination of this distance is crucial for the performance of the configuration and will be considered below. The following assumptions are made for
deriving the mathematical model:
(a) All trailers have the same length L. This assumption is not restrictive at all and
its main purpose here is to obtain a simpler model.
(b) Each trailer is modelled as a semi-trailer having only one rear axle.
(c) Each trailer is hooked up to the midpoint of the rear axle of the preceding
trailer. In reality this is only approximately true as drivers in an effort to balance vehicle stability and ease of driving, select the distance of the hook-up
point from the rear axle center depending on the load of the trailer.
(d) By sliding the location of the kingpin, the weight of the trailer shifts toward
an outer direction, which does not affect the kinematic behavior of the train.
(e) The unbalanced pulling point (when the kingpin sliding is nonzero) does not
cause skidding of the whole axle. Basically, it is hypothesized that the loss
of symmetry inflicted by the sliding does not produce moments that are high
enough to destabilize the vehicle system.

404

S. MANESIS ET AL.

(f) The sliding of the kingpin can be performed with the trailer fully loaded via
a hydraulic, pneumatic, or even electric mechanism. The hydraulic way is
probably the one that will most favorably balance power, efficiency, weight,
and cost.
(g) Sliding is always applied in the opposite direction to the steering input.
As in the case of the classical n-trailer system, we have the classical n + 1 nonholonomic constraints imposed by the rolling and non-slipping condition given by
Equation (3) and 2n holonomic equations introduced by the corresponding links,
which, because of the sliding kingpin mechanism are of the form:
xi+1 = xi L cos i+1 + Si sin i ,
yi+1 = yi L sin i+1 Si cos i .

(4)

The method for deriving the state equations for the modified n-trailer system
is the same one as for the classical case. Taking the derivatives of the holonomic
equations (4), combining with Equation (3) and eliminating variables xi , yi for
i = 0 leads to a system of (n + 1) equations.
Due to the complex nature of the above equations the solution is computationally tedious and time consuming dependent on the number of trailers. For
the algebraic solution we have used Mathematica v4.01, a software package for
symbolic manipulation. Until now and using technologically advanced computers, it has been possible to derive kinematic state equations for only five trailers
(n = 5), the computation time lasting about 20 hours. The derivation of kinematic
state equations for larger combinations is possible in the future and depends on the
computational power that will be available in the market or the willingness of the
designer to wait for the necessary calculations. Table I presents the time used for
deriving the state equations for the cases of three and five trailers with Mathematica
4.01 on a Pentium III, 1.2 GHz.
The kinematic equations for a multi-articulated vehicle with five trailers and a
sliding kingpin mechanism combined with equations of the leading vehicle yields:
x0 = U1 cos 0 ,
= U2 ,
1
0 = U1 tan ,
L

y0 = U1 sin 0 ,

Table I. Computational time for deriving

kinematic state equations
Number of trailers (n)
3
5

Time used (s)

322
65891

OFF-TRACKING SUPPRESSION IN MULTI-ARTICULATED VEHICLES

1 =
2 =

3 =

4 =

5 =

405


1 
1 (t) cos(0 1 ) ,
sin(

)[L
+
S
(t)
tan
]

L
S
U
1
0
1
1
L2

1 
sin(1 2 ) U1 [L cos(0 1 ) + S2 (t) sin(0 1 )]
3
L

[L + S1 (t) tan ] + LS 1 (t)[S2 (t) cos(0 1 ) + L sin(0 1 )]

L2 S2 (t) cos(1 2 ) ,
1 
S2 (t)[L cos(1 2 ) + S3 (t) sin(1 2 )] sin(2 3 )
L4 

U1 sin(0 1 )[L + S1 (t) tan ] LS 1 (t) cos(0 1 )

+ L sin(2 3 ) [L cos(1 2 ) + S3 (t) sin(1 2 )]


U1 cos(0 1 )[L + S1 (t) tan ] + LS 1 (t) sin(0 1 )

+ LS 2 (t)[S3 (t) cos(1 2 ) + L sin(1 2 )]

L3 S3 (t) cos(2 3 ) ,
1 
S2 (t)[L cos(1 2 ) + S3 (t) sin(1 2 )][L cos(2 3 )
L5

+ S4 (t) sin(2 3 )] sin(3 4 ) U1 sin(0 1 )[L + S1 (t) tan ]


LS 1 (t) cos(0 1 ) + L sin(3 4 ) [L cos(2 3 )

(5)
+ S4 (t) sin(2 3 )] [L cos(1 2 ) + S3 (t) sin(1 2 )]


U1 cos(0 1 )[L + S1 (t) tan ] + LS1 sin(0 1 )

+ LS 2 (t)[S3 (t) cos(1 2 ) + L sin(1 2 )]

+ L2 S3 (t)[S4 (t) cos(2 3 ) + L sin(2 3 )]

L4 S4 (t) cos(3 4 ) ,
1 
S2 (t)[L cos(1 2 ) + S3 (t) sin(1 2 )]
L6
[L cos(2 3 ) + S4 (t) sin(2 3 )][L cos(3 4 )
+ S5 (t) sin(3 4 )] sin(3 4 ) sin(4 5 )


U1 sin(0 1 )[L + S1 (t) tan ] LS 1 (t) cos(0 1 )

+ L sin(4 5 ) [L cos(3 4 ) + S5 (t) sin(3 4 )]


[L cos(2 3 ) + S4 (t) sin(2 3 )] [L cos(1 2 )
+ S3 (t) sin(1 2 )][U1 cos(0 1 )[L + S1 (t) tan ]
+ LS 1 (t) sin(0 1 )] + LS2 (t)[S3 (t) cos(1 2 )

+ L sin(1 2 )]

+ L2 S3 (t)[S4 (t) cos(2 3 ) + L sin(2 3 )]

+ L3 S4 (t)[S5 (t) cos(3 4 ) + L sin(3 4 )]

L5 S5 (t) cos(4 5 ) .

406

S. MANESIS ET AL.

Now if we assume that Si (t) remains equal to zero for all times (which is a rather
strong assumption), the equations are simplified significantly and it is possible to
derive a recursive form for the general case of n trailers:
y0 = U1 sin 0 ,
= U2 ,
x0 = U1 cos 0 ,
U1
1 = 2 sin(0 1 )[L + S1 (t) tan ],
L
1
2 = 3 U1 [L + S1 (t) tan ] sin(1 2 )[L cos(0 1 )
L
+ S2 (t) sin(0 1 )],
1
3 = 4 U1 [L + S1 (t) tan ] sin(2 3 )[L cos(0 1 )
L
+ S2 (t) sin(0 1 )][L cos(1 2 ) + S3 (t) sin(1 2 )],
..
.
1
n = n+1 U1 [L + S1 (t) tan ] sin(n1 n )
L
n1

[L cos(i1 i ) + Si+1 (t) sin(i1 i )].

(6)

i=1

4. Nonlinear Controller Design

In a multi-articulated vehicle operating with the sliding kingpin mechanism, two
different controllers need be derived: one for path planning and another that will
regulate the sliding distance in the sliding kingpin mechanism for off-tracking
elimination. For path following issues, the linear velocity and the steering angle
rate are the control inputs. In the transportation domain, the human driver regulates the above control inputs in such a way as to achieve kinematic stability
and the desirable trajectory tracking. For an autonomous multi-body robot moving
inside a limited laboratory or in an industrial environment, the embedded controller
regulates the control inputs based on a control algorithm for path following and obstacle avoidance. The overall structure of the control system for a multi-articulated
vehicle either robotic or transportation is depicted in Figure 3.
To automatically determine the proper sliding distance, a number of heuristic
control policies have been tried. A good balance between simplicity and effectiveness is provided by the following control policy
Si (t) = Ki 0 (t),

i = 1, . . . , n.

(7)

That is, the sliding distance is proportional to the rate of change of the tractor
orientation. Although the above controller performs in an acceptable manner, as
can be seen from the simulation results in the next section, its heuristic origin may
make it unsuitable for actual applications. It is desirable and from an engineering

OFF-TRACKING SUPPRESSION IN MULTI-ARTICULATED VEHICLES

407

Figure 3. The overall control system for the n-trailer vehicle.

point of view imperative to derive analytically a controller that guarantees theoretically the off-tracking elimination in an n-trailer vehicle. For that reason, we
proceed to the derivation of a nonlinear controller: It is known that the curve radius
So in general and for the ith trailer this
for a vehicle is given by r = U/ = U/.
radius will be given by
ri =

Ui
.
i

(8)

From the set of Equations (1) we have that

i1


U
1
cos(j 1 j ) sin(i1 i ).
i =
L j =1

(9)

By combining (8), (9) and taking into consideration the relation

Un = U1

n1


cos(j j +1 )

(10)

j =0

and after some algebraic manipulation this yields that

ri = L cot(i1 i ).

(11)

It stems from the last relation that the curve radii for different trailers are different so it is logical to introduce different sliding for each trailer. In [3] it was
proven that if the leading vehicle travels along a circular trajectory with radius r
(where
r > L) then the trailer converges to a circular trajectory with radius R =

r 2 L2 . The following lemma provides the condition for the leading vehicle
and the semi-trailer of a simple tractortrailer system to follow the same circular
trajectory.

408

S. MANESIS ET AL.

Figure 4. A simple tractortrailer system depicting the trajectories of the trailer, tractor, and
kingpin point.

LEMMA 1. If the kingpin sliding is given by S = r 2 + L2 r then the trailer in

the steady state follows the circular trajectory with radius r of the leading vehicle.
Proof. According to [3]the trailer in steady-state will travel along a circular
trajectory of radius Rss = rz2 L2 (Figure 4), whereas rz = r + S. So we have
that


Rss = rz2 L2 = (r + S)2 L2 = .
After some algebraic manipulations, we conclude that Rss = r.
The above lemma can be extended for the n trailers case following the same
procedure, so the sliding for the ith trailer will be given by

(12)
Si = ri2 + L2i ri
and must be realized on the ith 1 trailer.
By combining (11) and (12) we find that the different sliding distances that we
must apply to the different trailers are given by the relation
Si = L

1 cos(i1 i )
sin(i1 i )

(13)

OFF-TRACKING SUPPRESSION IN MULTI-ARTICULATED VEHICLES

409

so that
Si = L

i1 i
.
1 + cos(i1 i )

(14)

5. Simulation Results
To test the controllers derived in the last section, the Matlab/Simulink environment
was used. Although the main difficulty in simulations is to find the appropriate
that correspond to real conditions, after many trials we
driver inputs (U1 , )
managed to find a pair of inputs that according to our opinion is close to real
driving conditions. We must note here that the controller, which was analytically
derived, performs exceptionally well under any combination of inputs. The driver
inputs U1 , that we have used for simulation purposes are shown in Figure 5. In
Figure 6 simulation results are shown for a multi-articulated vehicle with three
trailers without/with using both controllers while the same simulation results for a
multi-articulated vehicle with five trailers are shown in Figure 7. In Figure 8 are
shown the outputs of the analytically derived controller (sliding distance), for both
train-like vehicles with three and five trailers.
Without loss of generality, we assume that all trailers and the leading vehicle
have a unit length. We must note that for our simulation purposes we use the
kinematic equations (5), which fully describe the kinematic behavior of n-trailer
system with sliding kingpin mechanism. Simulation results using the simplified
kinematic equations (6) have been presented in [6].
Although the inputs U1 , shown in Figure 5 do not correspond exactly to the
real driver inputs that they are applied to real transportation or robotic n-trailer
vehicles they are used here because the maneuver that the n-trailer vehicle performs
as a result of these two inputs is close to reality.

Figure 5. Driver inputs U1 in [m/s] and U2 = in [rad/s].

410

S. MANESIS ET AL.

Figure 6. Trajectories for a multi-articulated vehicle with three trailers without/with linear
and nonlinear controller.

Observing Figures 6 and 7, we may easily notice that the simulation results are
much better when we use the nonlinear controller of Equation (13) instead of the
linear controller of Equation (7), in the sense that when the former (analytically
derived) controller is in action, the off-tracking is practically zero in all phases of
motion (sections of the trajectory). Observing Figure 8, we may notice that the

OFF-TRACKING SUPPRESSION IN MULTI-ARTICULATED VEHICLES

411

Figure 6. (Continued.)

analytically derived nonlinear controller (13) has inherently the notion of delay in
applying the sliding distance and that stems from the fact that the sliding distance
is a function of both i 1 and i .
6. Conclusions
Off-tracking is one of the most significant problems with potentially dangerous
consequences occurring in multi-articulated vehicles. This problem creates difficulties or even renders prohibitive the wide application of multi-articulated vehicles
in both robotic and transportation domains. The sliding kingpin mechanism is a
mechanism whose principle of operation allows correcting such deviations. In this
paper, we derive the complete kinematic equations for multi-articulated vehicle
equipped with the novel sliding kingpin mechanism. We concentrate on the three
and five semi-trailer cases without any additional assumption other than what nonholonomy requires. The complex nature of these equations makes very difficult and
time consuming but not impossible in principle the derivation of kinematic equations for more semi-trailers. Based on the analysis, we derive a nonlinear controller
for adjusting the sliding distance of the sliding kingpin mechanism based on the
theoretical steady-state off-tracking when the leading vehicle moves along a circular trajectory. Its response has been compared to another controller, which is linear
and has been heuristically developed. Both controller designs have been validated
through simulations and they both showed satisfactory performance. However, the
analytically derived nonlinear controller has proved to be superior in all cases,
eliminating off-tracking in all phases of the vehicle motion.

412

S. MANESIS ET AL.

Figure 7. Trajectories for a multi-articulated vehicle with five trailers without/with linear and
nonlinear controller.

Acknowledgement
This research work has been partially supported by the Caratheodory Program of
the Research Commission of the University of Patras.

OFF-TRACKING SUPPRESSION IN MULTI-ARTICULATED VEHICLES

Figure 7. (Continued.)

Figure 8. Sliding distances when the nonlinear controller of Equation (13) is used.

413

414

S. MANESIS ET AL.

References
1.
2.
3.

4.
5.

6.

7.
8.

Alexander, J. C. and Maddocks, J. H.: On the maneuvering of vehicles, SIAM J. Appl. Math.
48(1) (1988), 3851.
Altafini, C. and Gutman, P.-O.: Path following with reduced off-tracking for the n-trailer
system, in: Proc. of the 37th IEEE Conf. on Decision and Control, Florida, December 1998.
Bushnell, L., Mirtich, B., Sahai, A., and Secor, M.: Off-tracking bounds for a car pulling trailers
with kingpin hitching, in: Proc. of the 33rd Conf. on Decision and Control, Lake Buena Vista,
FL, December 1994, pp. 29442949.
Kolmanovsky, I. and McClamroch, H. N.: Develompents in nonholonomic control problems,
IEEE Control Systems Mag. 15(6) (1995).
Manesis, S.: Off-tracking elimination in road-trains of heavy duty trucks with multiple semitrailers, in: Proc. of the 8th IFAC/IFIP/IFORS/IMACS on Large Scale Systems: Theory and
Applications (LSS 98), Patras, Greece, 1998.
Manesis, S., Davrazos, G. N., and Koussoulas, N. T.: Controller design for off-tracking elimination in multi-articulated vehicles, in: Proc. of IFAC World Congress, Barcelona, Spain, July
2002.
Murray, R. M., Li, Z., and Shastry, S. S.: A Mathematical Introduction to Robotic Manipulation,
CRC Press, Boca Raton, FL, 1997.
Nakamura, Y., Ezaki, H., Tan, Y., and Chung, W.: Design of steering mechanism and control of
nonholonomic trailer systems, in: Proc. of IEEE Internat. Conf. on Robotics and Automation,
2000.