See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/267664200

Conventional and slip steering for multi-wheel planetary rovers

Article

CITATIONS READS

9 3,570

1 author:

Giancarlo Genta
Politecnico di Torino
301 PUBLICATIONS   2,421 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Actually, I am a Consultant of the European Community • Deputy Chairman of the Technical Committee TC3 of the IMEKO (Force, Mass and Couples and italian member
of TC-17 for Robotics. View project

Rotordynamics View project

All content following this page was uploaded by Giancarlo Genta on 22 February 2016.

The user has requested enhancement of the downloaded file.

Acc. Sc. Torino
Memorie Sc. Fis. 34 (2010) MECCANICA DEI SOLIDI
MECCANICA DEI SOLIDI

Conventional and slip steering

for multi-wheel planetary rovers
Memoria del Socio corrispondente GIANCARLO GENTA
presentata nell’adunanza del 10 marzo 2010

Abstract. All planetary rovers used up to now and most of those under
study are based on wheels for their mobility. The trajectory control of
wheeled robots is mostly performed by steering some or all wheels, but the al-
ternative of slip steering seems to be quite promising, in particular for low
gravity conditions. The aim of the present paper is to investigate conventional
and slip steering using simplified models suitable for rovers with more than
two axles, moving at low speed on short radius trajectories. Two examples
show that even at low speed the kinematic steering model is unable to predict
the trajectory and that slip steering compares favorably with conventional
steering.
KEYWORDS: planetary rovers, trajectory control, slip steering.

Riassunto. Tutti i robot e i veicoli usati per l’esplorazione planetaria, e la

maggior parte di quelli allo studio, utilizzano ruote per raggiungere la desi-
derata mobilità. Il controllo della traiettoria viene nella maggior parte dei
casi effettuato mediante la sterzatura di alcune o di tutte le ruote, ma l’uso di
una strategia basata sullo scorrimento (slip steering) sembra essere promet-
tente, in particolare nel caso di bassa gravità. Scopo del presente lavoro è lo
studio della sterzatura convenzionale e di quella per scorrimento mediante
l’uso di modelli semplificati, ma adatti a veicoli a molti assali operanti a bas-
sa velocità su traiettorie di piccolo raggio. Due esempi mostrano come il mo-
dello della sterzatura cinematica non permetta di predire la traiettoria nep-
pure a bassa velocità e che la sterzatura per scorrimento permetta di ottenere
prestazioni non inferiori a quelle della sterzatura convenzionale.
PAROLE CHIAVE: veicoli per l’esplorazione planetaria, controllo della traietto-
ria, scorrimento.

1. Introduction
Lunar and planetary exploration depends heavily on devices able to move
on the surface of celestial bodies, carrying scientific instrumentation or
human explorers or performing a number of tasks required to establish
human outposts or to exploit the resources available in situ.
The running gear of all planetary rovers used up to now and most of
those under study is based on a number of wheels going from 4 to 8. The
design of these machines is different from that of most wheeled vehicles and
4 Giancarlo Genta

robots used on Earth, mostly due to the reduced gravity in which they must
operate and to their low speed.
Trajectory control is one of the basic function of any moving machine; in
the case of wheeled vehicles and robots it is performed by exploiting the
lateral forces (cornering forces) exerted by the wheels, that bend the trajec-
tory while also compensating for any lateral force exerted on the vehicle,
like aerodynamic side force due to side wind, gravitational forces due to
lateral inclination of the ground, etc.
However, in the early stages of automotive development, the concept of
kinematic steering evolved, based on the idea that the wheels are in pure
rolling conditions, defined as a condition in which there is no slip between
the wheel and the ground and consequently the velocity of the centre of the
wheel lies in its mid-plane. The same concept is also used in many studies
on the control of wheeled automatic vehicles and robots.
In spite of being well known that this concept holds only for velocities
and lateral forces tending to zero, the design of the steering system is often
based on the so-called Ackerman condition, i.e. the condition that makes
kinematic steering possible [1, 2]1.
The actual control sequence for a vehicle provided of a conventional
steering system is however the following:
The driver, or the automatic trajectory control system, operates the steer-
ing causing some (or all) wheels to work with a sideslip angle and to gener-
ate lateral forces.
v These forces cause a change of attitude of the vehicle2 and then a
sideslip of all wheels.
v The resulting forces bend the trajectory.
The linearity of the behavior of the tires and the very high value of the
cornering stiffness give the driver the impression of a kinematic, not dy-
namic, driving. The wheels seem to be in pure rolling and the trajectory
seems to be determined by the directions of the mid-planes of the wheels.
This impression has influenced the study of the handling of wheeled ve-
hicles and above all of wheeled robots for a long time, originating the very
concept of kinematic steering and in a sense hiding the true meaning of the
phenomena.

1
Ackermann introduced in 1818 and patented the idea that the lines perpendicular to the midplanes of
the wheels passing through the contact points on the ground should converge in the instant center of
rotation of the vehicle. This concept was applied to horse driven carts and worked well in connection
with cart wheels provided with steel tires. Only in the 1930s some experiments did show the importance
of the sideslip angle to generate lateral forces and this concept was finally formalized by Olley in 1937.
2
The attitude or sideslip angle is defined as the angle between the longitudinal direction of the vehicle
and the velocity vector.
 Conventional and slip steering for multi-wheel planetary rovers 5

This way of controlling a wheeled vehicle is typical of motor vehicles

used on hard surfaced roads, but there is another possibility. Instead of
steering some of the wheels, the torque causing the vehicle to rotate and to
assume the required attitude can be produced by differential traction on the
wheels of the same axle. In this case no steering wheel in the classical sense
may be present.
This way of controlling the trajectory may be implemented to supplement
the more usual way, as in the case of many VDC (Vehicle Dynamics Con-
trol) system where the driver chooses the trajectory by steering the front
wheels while the device maintains the vehicle in the required trajectory by
differentially braking the left and right wheels using suitable control algo-
rithms. In other cases this may be the main way of controlling the trajectory
like in some wheeled earth moving machines and above all in tracked vehi-
cles. Many three wheeled small robots work in this way too, but some of the
wheels are either omnidirectional or swivelling wheels.
Propulsion of wheeled vehicles and rovers is usually implemented
through the wheel-ground contact. Both propulsion and trajectory control
functions are thus implemented through the forces applied by the wheels on
the ground, which are caused by the deformation of both the bodies in
contact (wheel and ground). As a consequence, the wheels of a vehicle are
always operating with some sideslip and longitudinal slip and are never in
pure rolling.
The steering function can be implemented in the following ways [3, 4]:
v Slip steering3 (Fig. 1a). There is no steering mechanism, but the driv-
ing wheels can produce differential traction on the two sides of the ve-
hicle. This can be achieved by using independent motors in the wheels
or two motors actuating the wheels of the two sides independently, or
by using a differential gear. The brakes can also be used in differential
mode to achieve some slip steering. Very small trajectory curvature ra-
dius can be obtained by rotating backwards the wheels on one side, and
turning on the spot is possible by using equal and opposite rotation
speed. No kinematic steering (see below) is possible.
v Articulated steering (Fig. 1b). One or more axles, carrying each the
wheels at both sides, are articulated and can steer under the control of
one or more actuators. The body can be made of two or more subunits,
each carrying one or more axles, or the wheels can be carried by bo-
gies, pivoted under the body. If each subunit carries only one axle, ki-
nematic steering is possible.

3
Sometimes the term skid steering is used instead of slip steering. It should be avoided, since this way of
steering is due to longitudinal and lateral slip of the wheels, but only in extreme cases it produced actual
skidding (i.e. global slipping) of the wheels on the ground.
6 Giancarlo Genta

Fig. 1. Steering of a wheeled rover or vehicle. a) Slip steering; b) articulated steering; c) coor-
dinated steering, d) independent steering

v Coordinated steering (Fig. c). The two wheels of the same axle steer
about two different steering axes (kingpin axes), but their steering an-
gles are coordinated by a mechanical linkage. If the linkage satisfies
the Ackerman condition (see below), an Ackerman steering, making
kinematic steering possible, is realized. However, no practical steering
linkage, like the much used Jeantaud linkage, realizes an exact Acker-
man steering. If more than one axle is steering, the steering angles of
the various axles may be independent or (seldom) coordinated by an-
other mechanical linkage. Most automotive vehicles have a steering of
this kind.
v Independent steering (Fig. 1d). Each wheel of one axle or of more ax-
les can steer independently about a steering (kingpin) axis, under the
action of an actuator. The possibility of realizing the Ackerman condi-
tions (thus making kinematic steering possible) depends on the steering
control system. If large steering angles are possible, and the control
system is flexible enough, any kind of trajectory can be obtained, in-
cluding turning on the spot, moving sideways, etc. A steering of this
kind was used on the Apollo LRV [5, 6].

The present paper is devoted to the comparison between conventional

and slip steering in multiwheel rovers. The vehicle is modelled as a single
rigid body, and thus does not deal with articulated steering, that will be the
object of a future paper.
The presence of the vehicle suspensions and the compliance of the tires
will be neglected in the study of the vehicle as a whole, while they can be
reintroduced in the computation of the forces exerted by the vehicle on the
ground.
 Conventional and slip steering for multi-wheel planetary rovers 7

2. Model for a multiwheel rigid rover – kinematic steering

2.1. General model
Often first approximation models in which vehicles are assumed to be
rigid bodies are used in automotive technology. They are usually based on the
assumption that the radius of the trajectory of the vehicle is much larger than
the size of the vehicle, leading to a linearization of the model, with the subse-
quent possibility of defining some gains (e.g., the trajectory curvature gain, the
sideslip angle gain, etc.) and of studying the stability of the vehicle.
What is common in automotive technology, that deals with fast vehicles
mostly operating in a structured environment on roads with large curvature
radii, may be not applicable in the case of planetary rovers, which operate at
low (often extremely low) or moderate speed in unstructured environments.
A model is here developed without any assumption leading to linearization.
A sketch of a multi-wheel rover is shown in Fig. 2a. It consists of a rigid
body with mass m and a moment of inertia Jz about the z axis, that moves on a
flat and terrain sloping at an angle with respect to the horizontal. The inertial
reference frame has its axis laying on the ground; the X axis being horizontal
and Y axis sloping upwards.

Fig. 2. Sketch of a multi-wheel rover. a): full model; b): monotrack model, c) monotrack model
with two axles (bycicle model)

The total number of degrees of freedom of the rigid body is thus 3, and the
coordinates X and Y of the center of mass G and the yaw angle can be taken
as generalized coordinates1.

1
In the studies on the trajectory control of robots the displacements generalized coordinates are usually the X
and Y coordinates of the center of the rear axle. Here the usual praxis common in motor vehicle dynamics of
referring to the center of mass is followed.
 2

8 Giancarlo Genta

Often the monotrack model (Fig. 2b) is used for first approximation studies:
each axle is reduced to a single wheel in the symmetry plane xz of the vehicle.
If a constraint is set that inhibits the rolling motion, the dynamics of the
monotrack vehicle is the same as that of the actual model, the difference being
only the way in which the forces on the ground are accounted for. A further
simplification is that of Fig. 2c, in which the wheels are only two: in this case
kinematics steering is always possible. In Fig. 2c the sideslip angle of the
vehicle and the components u and v of the absolute velocity V are also defined.
The model of Fig. 2a is sometimes called all-wheel dynamic model and was
used for the study of planetary rovers [7, 8]. In the present paper it is used to
study both conventional and slip steering. In the latter case all angles Dij are
equal to 0 and steering is performed by differential longitudinal forces Fxij. In
case of coordinated steering there is a fixed law, due to the kinematics of the
steering linkage, that links angles i1 and i2, while in case of independent
steering that law is due to the control system and can be set with larger flexi-
bility.
The wheels are modelled as non-holonomic constraints, as already stated.
The generic velocity of the of the centre Pij of the contact area of the jth wheel
of the ith axle is
ti
uY
VP ij  VG  Y, (Pij - G )  2 , for i = 1, …, n, j = 1, 2, (1)
v  Yxi
where the upper sign holds for j = 1 (left wheel). Parameter xi may be either
positive (the axle is in front of the center of mass), negative or 0.
If the generic wheel ij is steered by angle Dij, the velocity of the same point
referred to the frame of the wheel is

cos D ij 
sin D ij u
ti
Y
VP ij  

sin D ij 
cos D ij 2
v  Yxi
t (2)
u cos D  v sin D  i cos D  x sin D Y
ij ij 2 ij i ij

t
u sin D  v cos D  p i sin D  x cos D Y
ij ij 2 ij i ij

The condition for pure rolling is that the velocity of point Pij is contained in
the midplane of the wheel
 3

Conventional and slip steering for multi-wheel planetary rovers 9

 
f ij (u, v,Y )  u sin D ij  v cos D ij  p
ti
2
 
sin D ij  xi cos D ij Y  0
(3)
, for i = 1, …, n, j = 1, 2.

It is easy to verify that this constraint is not integrable, and thus is a non-
holonomic constraint.
There are 2n non-holonomic constraint equations, with just 3 degrees of
freedom, which means that the system is overconstrained. The various angles
Dij are not independent, and only two of the steering angles can be stated arbi-
trarily if the vehicle is to be free to move along a trajectory on the ground
plane. Once they are stated, all other ones follow as a consequence.
To simplify the formulation, without introducing any approximation, let the
constraints be stated on the equivalent two-wheeled monotrack model (Fig.
2c). In this way angles 1 and n are stated (they are the control parameter for
the trajectory), while the conditions for all the actual steering angles of the
various wheels follow.
The two constraint equations are thus
f1 (u , v,Y )  u sin D1  v cos D1  aY cos D1  0 (4)
f 2 (u , v,Y )  u sin D n  v cos D n bY cos D n  0.

The conditions for obtaining the narrowest radius of curvature of the trajec-
tory is
D n  D1 .
The constraint equations can be written in a simpler way by solving Equa-
tions (4) in v and Y :

v  ;b tan D1  a tan D n =

u
l (5)
Y  ;tan D1 tanD n = .
u
l

These equations link the yaw and lateral velocities with the longitudinal ve-
locity.
The steering angles of the various wheels are linked with the trajectory con-
trol parameters by the relationships

D ij  atan
b  xi tanD1  a xi tanD n
l i ;tan D1 tan D n =
t , for i = 1, ..., n, j = 1, (6)
2 2.
 4

10 Giancarlo Genta

The Lagrangian function including the kinetic energy and the gravitational
potential energy of the system is

L  T U  m  u 2  v 2  rz2Y 2 mgY sin( A ),

1
2 (7)

where rz is the radius of inertia

Jz
rz  .
m

The rotational kinetic energy of the wheels has been neglected: no gyro-
scopic effect of the wheels will be obtained in this way.
The velocities in the body-fixed frame are linked to the derivatives of the
generalized coordinates by the relationship
u cos Y sin Y X
 . (8)
v -sin Y cos Y Y

The constraints are non-holonomic and, once only two of them are consid-
ered, the Jacobian matrix of functions fj has rank 2: the non-holonomic con-
straints are independent. They can be dealt with by writing the Lagrange
equation in the form [9, 10]
d uL uL 2 uf j
 Lj  Qi . (9)
dt uqi uqi j 1 uqi

The Lagrange multipliers i are 2, as expected.

The forces applied to the system are the net driving forces, i.e. the 'drawbar
pulls' referred to the single wheels, that are directed in the direction of their
midplanes, i.e. forces Fxij, with i = 1, …, n and j = 1, 2, act in the direction of
the xij.axis fixed to the ijth wheel.
The virtual displacement of the center of the contact area of the ijth wheel
in the reference frame of the wheel is expressed in terms of virtual displace-
ments of the vehicle in the inertial frame as

D X cosY Dij DY sinY Dij DY cosDij  xi sinDij

ti
2 (10)
D qPij  .
D X sinY Dij DY cosY Dij DY p i sinDij  xi cosDij
t
2
 5

Conventional and slip steering for multi-wheel planetary rovers 11

The virtual work due to the driving force at the wheel ij is thus

DWij  Fxij DX cosY  D ij  DY sinY  Dij  DY  

1
ti cos D ij  xi sin D ij
2
(11)
In performing the relevant derivatives entering Lagrange equations, it must
be remembered that, since each linear velocity of the center of mass expressed
in the body-fixed frame depends on the velocity expressed in the inertial frame
through Eq. (8), a set of relationships like, for instance,
uL uL uu uL uv uL uL
   cos Y  sin Y . (12)
uX uu uX uv uY uu uv
can be written.
The equations of motion are thus
;mu vY  L1 sin D11  L2 sin D n1 =cosY ;mv  uY 

n 2
 L1 cosD1  L2 cosD n =sin Y  Fxij cos Y  D ij
i 1 j 1
;mu vY  L1 sin D11  L2 sin D n1 =sin Y  ;mv  uY  L1 cosD1 
Fxij sin Y  D ij
n 2
 L2 cosD n =cosY  mg sin A 
(13)
i 1 j 1
n
J zY  L1a1 cosD1 L2b cosD n  ti ; Fxi1 cosD i1 
1
i 1 2
 Fxi 2 cosD i 2 =  xi ;Fxi1 sin D i1  Fxi 2 sin D i 2 = ]

By premultiplying the right and left sides of the first two equations by the
matrix in Eq. (8), the equations of motion become
mu vY  L1 sin D11  L2 sin D n1 

n 2
 mg sin A sin Y  Fxij cos D ij
i 1 j 1
mv  uY  L1 cosD1  L2 cosD n 

n 2
 mg sin A cosY 
(14)
Fxij sin D ij
i 1 j 1
n
J zY  L1a1 cosD1 L2b cosD n  ti ; Fxi1 cosD i1 
1
i 1 2
 Fxi 2 cosD i 2 =  xi ;Fxi1 sinD i1  Fxi 2 sinD i 2 = ]
 6

12 Giancarlo Genta

The 2 constraint equations are equations (5), repeated below

v
u
;b tanD1  a tanD n =
l
(5a)
Y  ;tan D1 tan D n =
u
l
The total number of unknown is now 5, which coincides with the number of
equations. All the equations are nonlinear, owing to the presence of trigono-
metric functions and of products of unknowns.
From the equations it is clear that the meaning of the Lagrange multipliers
is (apart from its sign) the force the constraint applies to the rigid body in
correspondence to the front and rear axles in model of Fig. 1c.
Velocities u and v, angle , its corresponding yaw velocity r  Y and the
Lagrange multiplers 1 and 2 can be considered as the state variables. Working
in this way, there are 6 state variables, and 6 equations (2 differential and 4
algebraic):
L L
cosD ij
n 2
Fxij
u  vr 1 sin D1 2 sin D n g sin A sin Y 
m m i 1 j 1 m

Y r


n 2
L1 cosD1  L2 cosD n  mv  ur mg sin A cosY  Fxij sin D ij
i 1 j 1
n
L1a1 cosD1 L2b cosD n  J z r  ti ; Fxi1 cosD i1  Fxi 2 cosD i 2 = 
1
i 1 2
 xi ;Fxi1 sin D i1  Fxi 2 sin D i 2 = ]
(15)
u
v;b tanD1  a tanD n =
l

r  ;tan D1 tanD n =

u
l
The trajectory can be easily computed by integrating the absolute velocities
X cosY - sin Y u
 (16)
Y sin Y cosY v
.
2.2. Steady state steering response
A steady state solution with u, v and r constant is readily obtained. Given
the speed V on the trajectory, that is constant, from the fifth equation (15) it is
possible to obtain the lateral and longitudinal velocities v and u
 7

Conventional and slip steering for multi-wheel planetary rovers 13

b tan D1  a tan D n

v V
l  ;b tan D1  a tanD n =
2 2
(17)
l
u V
l 2  ;b tan D1  a tan D n =
2

The angular velocity r is thus

tan D1 tan D n (18)
r V
l  ;b tan D1  a tan D n =
2 2

From Eq. (18) the radius of the trajectory is obtained

(19)
l 2  ;b tan D1  a tanD n =
2
V
R 
r tanD1 tan D n
The other equations (15) can be considered as a set of equations allowing to
compute the Lagrange multipliers 1 and 2 and the longitudinal forces Fxij that
allow the vehicle to proceed on a circular, steady state trajectory. There are
more unknowns than equations, since other relationships linking the various
longitudinal forces must be written. If the controller insures that the longitudi-
nal forces are all equal, the equations become


n 2
L1 sin D1  L2 sin D n Fx cos D ij  mvr mg sin A sin Y
i 1 j 1


n 2
L1 cosD1  L2 cosD n Fx sin D ij  mur mg sin A cosY (20)
i 1 j 1
n
L1a1 cosD1 L2b cosD n Fx ti ; cosD i1  cosD i 2 = 
1
i 1 2
]
 xi ;sin D i1  sin D i 2 =  0
They are linear in 1, 2 and Fxij and are easily solved. If the ground is slop-
ing ( 0), these unknowns are functions of time, i.e. all forces, including
Lagrange multipliers, must change in time to keep the velocity constant.

2.3. Response to a lateral force

Assume that a constant lateral force acts on the vehicle travelling on a
straight trajectory, as in the case of traversing a slope with angle in a direc-
tion perpendicular to the slope ( = 0). In the more general case of arbitrary
angle there is also a longitudinal force.
Since the required trajectory is straight, r is equal to zero. Also all steering
angles are equal to zero, as needed to travel on a straight trajectory.
 8

14 Giancarlo Genta

The first second and third Equation (20) yield the values of the Lagrange
multipliers
b a
L1  mg sin A cos Y , L2  mg sin A cos Y .
l l
From the first equation it follows
Fx  mg sin A sin Y .
The physical meaning of this solution is obvious: the lateral force is dealt
with by the Lagrange multipliers, i.e. by the forces the ‘guide’ is exerting on
the vehicle. It is clear that this solution has no meaning, since there is no
physical guide but the lateral forces exerted by the wheels, which imply non-
zero sideslip angles.

2.4. Linearization of the model

If the radius of curvature of the trajectory is large with respect to the length
of the vehicle, some simplifications can be performed.
The components u and v of the velocity V in the body fixed frame are ob-
tained from the small sideslip angle of the vehicle by the relationships

u  VcosB z V v  V sin B z VB . (21)

The equations of motion simplify as

L1 L2 n 2
Fxij
V  vr D11 D n1 g sin A sin Y 
m m i 1 j 1 m
Y r


n 2
L1  L2  mv  Vr mg sin A cosY  Fxij sin D ij
i 1 j 1
n
(22)
L1a1 L2b  J z r 
ti
; Fxi1  Fxi 2 =  xi ;Fxi1 sin D i1  Fxi 2 sinD i 2 =
i 1 2

v
V
;bD1  aD n =
l

r  ;D1 D n =
V
l

The linearized expression for the steering angles is

 9

Conventional and slip steering for multi-wheel planetary rovers 15

D ij 
b  xi D1  a xi D n , for i = 1, …, n, j = 1, 2.
(23)
l i ;D1 D n =
t
2
In case of steady stare running, the usual relationships for kinematic steer-
ing are obtained:
bD  aD n
v V 1
l (24)
D D
r V 1 n
l
yielding the usual value of the trajectory curvature gain
l 1 (25)

RD1 D n l .

3. Model for high-speed cornering

3.1. General model
A model for high speed cornering can be derived from the previous one by
excluding the kinematic steering condition, i.e. the set of non-holonomic
constraints stating that the velocities of the centers of the wheels lie in their
midplanes, and introducing the cornering forces of the wheels.
Since this model is explicitly based on the dynamic equations of the system,
it is important to evaluate correctly the forces the wheel exchange with the
ground. Longitudinal and lateral forces depend on the normal forces the wheels
exert on the ground and thus can be computed only after the longitudinal and
transversal load shift have been defined and depend also on the characteristics
of the suspensions that distribute the vertical load on the ground.
It is a common opinion that transversal load shift can be neglected if the
lateral acceleration is below 0.5 g, which in case of ideal steering corresponds
to a lateral force coefficient y = 0.5 [1]. Considering that a larger value is
obtained in actual steering, since the vehicle is unable to fully exploit the
cornering forces of the tires, and that the force coefficient on regolith can
hardly be higher than 0.6, neglecting load shift is acceptable, at least as a first
approximation, in planetary vehicles.
The effect of the suspensions is easily predictable if articulated suspensions
like rocker bogies are used, but depend strictly on the ground irregularities if
automotive type elastic suspensions are used. Also the compliance of the
wheels can have some effect in the latter case.
 10

16 Giancarlo Genta

For a first approximation study that does not consider the exact profile of
the ground, all these effect can be neglected.
The cornering forces can be computed from the sideslip angle. From Eq.
(2), the tangent of the sideslip angle is obtained as the ratio between the out-of
plane and the in-plane velocity of the wheels:

  1

u sin D ij  v cos D ij  p ti sin D ij  xi cos D ij Y
2
(26) 
A ij  artg ,
 
u cos D ij  v sin D ij 
1
2

ti cos D ij  xi sin D ij Y 
for i = 1, …, n, j = 1, 2,
where as usual the upper sign holds for j = 1 and the lower one for j = 2.
The steering angles of the wheels are determined by the control system in
case of independent steering or by the kinematic characteristics of the me-
chanical linkages in case of coordinated steering. The same formulae used for
kinematic steering can be used, with the difference that in this case a perfect
Ackerman control is not needed.
The equations of motion can be obtained exactly as those seen above for the
kinematic condions, with the obvious difference of cancelling the non-
holonomic constraint (and thus cancelling the multipliers Li) and introducing
the forces Fyij acting on the centers of the contact areas in the direction perpen-
dicular to their midplanes. Moreover, since the wheels now work with a side-
slip angle, also the aligning torques Mzij should be added.
The equations of motion are thus

;F   =
n 2
mu vY  mg sin A sin Y  xij cos D ij Fyij sin D ij
i 1 j 1

;F   =
n 2
mv  uY  mg sin A cosY  xij sin D ij  Fyij cos D ij
i 1 j 1 (27)

;   =
n 2
1
J zY  ti Fxij cos D ij Fyij sin D ij 
2
]
i 1 j 1

;   =
 xi Fxij sin D ij  Fyij cos D ij  M zij .
There are now no constraint equations, but only the equations linking an-
gles ij to the cornering forces. In general they can be written in the form
Fyij  f1ij A ij , A ij  f '1ij u , v, r , u , v, r (28)
M zij  f 2ij A ij , A ij  f '2ij u, v, r , u , v, r ,
even if other parameters, like the camber angle, may enter the equations. For
hard surfaced ground and pneumatic tires today the most common approach is
the use of the so-called magic formula, an empirical formulation based on
 11

Conventional and slip steering for multi-wheel planetary rovers 17

experimental data. In case of soft terrain other formulas, based on terramechan-

ics and experimental results, are often used [11 - 15]. Reliable data on the non
pneumatic tires developed for both terrestrial [16] and planetary use are still
missing.

3.2. Linearized model

A very common simplification is to assume that the radius of the trajectory
is much larger than the wheelbase and, consequently, all the relevant angles are
small enough to linearize their trigonometric functions.
If the steer and sideslip angles are small, the latter are
1
uD ij  v  p tiD ij  xi Y
2 uD ij  v  xiY
A ij z z (29)
1 1
u  vD ij  ti  xiD ij Y u tiY
2 2
where the second simplification is obtained by neglecting the products of small
quantities that are of the same order of magnitude of the terms that are ne-
glected in the series for the cosine.
At least at a high enough speed, also the second term at the denominator can be
neglected; remembering Eq. (21) the sideslip angles can be approximated as
xY
A ij z B  i D ij (30)
V
If the two steering angles of the wheels of any given axle are supposed to be
equal (or are substituted by their average value), their sideslip angles are equal.
This justifies the usual praxis of speaking of sideslip angles of a given axle and
not of that of each wheel.
Linearizing also the cornering forces and aligning torques as functions of the side-
slip angle, the equations of motion (e.g. Eq. (25.108) in [2]) simplify as
n 2
mV  Fxij
i 1 j 1


mV B  r  mVB  YB B  Yr r  YD D  Fye (31)
J z r  N B B  N r r  N D D  M ze .
Where, neglecting aerodynamic forces, the so-called derivatives of stability
reduce to
n n
YB  Ci YB  Ci
i 1
, i 1
,

 xiCi  M zi,A  x C  x M (32)

n n
1
NB  Nr  2
i i i zi ,A
, V i 1
,
i 1
 12

18 Giancarlo Genta

K 'i xi Ci M zi ,A
n n
YD  K 'i Ci ND 
, i 1
,
i 1

Ci and Mzi, are the cornering and the aligning stiffness referred to the whole
axle instead of the single wheel and K’i are coefficients stating how the various
steering angles of the axles are linked together.
Notice that the longitudinal behaviour uncouples from the lateral (usually
referred to as handling) behaviour and only two (linear) first order differential
equations are needed to study the latter. They are usually written in a state
space formulation, using the lateral v and yaw r velocities (or angle B and the
yaw velocity r) as state variables.

3.3. Steady-state steering response

Also for the case of dynamic steering it is possible to obtain a steady-
state solution. The sideslip angles are still given by Eq. (26), and the equations
of motion reduce to

;F   =
n 2
mvr  mg sin A sin Y  xij cos D ij Fyij sin D ij
i 1 j 1

;F   =
n 2
mur  mg sin A cosY  xij sin D ij  Fyij cos D ij
i 1 j 1
(33)

;   =
n 2
1
ti Fxij cos D ij Fyij sin D ij 
2
] 0
i 1 j 1

 x ;F sin D  F
i xij ij yij cosD =  M
ij zij

The 3 equations of motion must be supplemented by the 2n equations link-

ing angles ij to the cornering forces and by other 2n equations linking the
longitudinal forces with each other.
Assume that the ground is flat ( = 0), all wheel exert the same longitudinal
force Fx and that the functions that express the side forces and aligning torques
are, in steady state condition,
Fyij  f1ij A ij  f '1ij u, v, r
M zij  f 2ij A ij  f '2ij u, v, r .

The set of 3+2n algebraic equation can be written in the form

pv, r , Fx , Fyij  0 , (34)
where vector p is
 13

Conventional and slip steering for multi-wheel planetary rovers 19

 
n 2 n 2
mvr  Fx cos D ij Fyij sin D ij
i 1 j 1 i 1 j 1

 
n 2 n 2
mur  Fx sin D ij  Fyij cos D ij
i 1 j 1 i 1 j 1
(
 
n 2
1
p  Fxij ti cos D ij  xi sin D ij  . 35)
i 1 j 1 2

 
n 2
1
 Fyij p ti sin D ij  xi cos D ij  M zij
i 1 2
Fy11 f '111 u, v, r
.......... ....
If the unknowns are ordered in vector
X v r ; Fx Fy11 Fy12 ... Fyn 2 = T
, (36)
the Newton-Raphson algorithm to compute the solution at the (k+1)th itera-
tion from that at the kth iteration is
X k 1  X k S 1p , (37)

where the Jacobian matrix at the ith iteration can be computed partially ana-
lytically and partially numerically.
The first column, up / uv can be computed numerically obtaining the values
of functions pi for two values of v separated by a very small quantity. The
second column regarding the derivatives with respect to r can be computed in
the same way. The first 3 lines of all other columns contain sums of sines and
cosines of angles ij. The last 2n terms of the third column are zero, while the
2n × 2n matrix at the lower right corner is an identity matrix.
Starting the computation from the kinematic values of the parameters, the
convergence on the solution is straightforward.
If a linearized solution is acceptable, from the second Equation (31) it is
possible to obtain
mVr  YB B  Yr r  YD D  Fye
(38)
0  N B B  N r r  N D D  M ze .
yielding the various gains, defined as the ratios between the outputs (curva-
ture of the trajectory, sideslip angle, etc.) and the inputs (steering angle, lateral
forces and moments). They are constant with respect to the inputs (the system
is linearized) but in general are functions of the speed.
The trajectory curvature gain (ratio between the curvature 1/R and the steer-
ing angle when all external forces and moment vanish) is
 14

20 Giancarlo Genta

1 YD N B  N D YB

;
RD V N B Yr mV N rYB = (39)

The sideslip angle gain is

B N D Yr mV  N r YD

D N B Yr mV N r YB (40)

3.4. Response to a lateral force

Assume again that a constant lateral force acts on the vehicle, as in the case
of traversing an incline with angle . If the direction of motion is perpendicular
to the slope ( = 0), the external force is only lateral, otherwise also a longitu-
dinal force acts on the vehicle.
The solution can be performed as for the previous case, and, if the steering
angles are locked to zero, the various functions pi for the Newton-Raphson
algorithm are
mvr  mg sin A sin Y 2nFx
n 2
mur  mg sin A cosY Fyij
i 1 j 1

x F
n 2
p i yij  M zij .
i 1
(41)

Fy11 f111 u, v, r

.......... ....

It must be expressly noted that this solution is a steady state one, and hence
it does not allow to study a change of slope, and in particular, if the solution
yields r 0, the vehicle turns about its vertical axis. Angle may thus not
being constant for a time long enough to allow this solution to have an actual
meaning.
4. Slip steering
4.1. Equations of motion
Since no kinematic steering is possible in this condition, only dynamic
steering will be considered.
The wheels do not steer (all ij = 0), and thus the sideslip angles of the
wheels are:
 15

Conventional and slip steering for multi-wheel planetary rovers 21

v  Yxi
A ij  artg , for i = 1, …, n, j = 1, 2, (42)
t
u iY
2
where as usual the upper sign holds for j = 1 and the lower one for j = 2.
The equations of motion are
n 2
mu vY  mg sin A sin Y  Fxij
i 1 j 1

n 2 (43)
mv  uY  mg sin A cosY  Fyij
i 1 j 1

n 2
1
J zY  ti Fxij  xi Fyij  M zij
i 1 j 1 2
The steering control is now provided by the longitudinal forces: the sim-
plest control law is stating that the longitudinal forces are
Fxij  Fx $Fx (44)

where the upper sign holds for j =1 and Fx is the control parameter for direc-
tional control.
The equations of motion reduce to
mu vY  mg sin A sin Y  2nFx
n 2
mv  uY  mg sin A cosY  Fyij (45)
i 1 j 1

;x F =
n n 2
J zY  ti $Fx  i yij  M zij
i 1 i 1 j 1

In the present case it is not important to state that the longitudinal forces
exerted by the wheels are all equal, since the longitudinal force simply add to
each other and what matters is their sum, not their individual value.
The unknown Fx appears only in the first equation, which can be separated
from the other ones and solved in Fx by itself. Notice that if the interation of
the forces in x and y directions is accounted for, this uncoupling is not com-
plete, but can be nevertheless used inside each iteration loop.

4.2. Steady state response; directional behavior.

The steady state equations of motion, can be written in the usual form
p0 ,
 16

22 Giancarlo Genta

where, if the uncoupling is accounted for, the size of vector p is 2 + 2n and not
3 + 2n. In this case it is advisable to take as unknowns the lateral velocity v,
the differential force Fx (instead of the yaw velocity r or the radius of the
trajectory R = V /r) and the sideslip angles ij instead of the lateral force Fij.
The use of Fx is advisable since in the present case there is no relationship
linking the control variable with the radius of the trajectory at low speed (there
is no kinematic steering condition) and thus once stated Fx a non realistic
trajectory can be obtained even at low speed. It is thus better to state the re-
quired trajectory (i.e. a value of R or a value of r) and then computing the
differential force that allows to obtain it.
The vector of the unknowns is thus
X  ;v $Fx A11 A12 A 21 ... A n 2 = .
T (46)

Vector p is
n 2
mur  Fyij
i 1 j 1

;x F =
n n 2
ti $Fx  i yij  M zij (47)
i 1 i 1 j 1
p .
v  rxi
A y11 artg
t
u ir
2
.......... ...................
The equation
mvr (48)
Fx 
2n
must also be solved at each iteration step to account for the interaction between
longitudinal and lateral forces on the ground.

4.3 Linearized solution

Also in this case it is possible to obtain a linearized solution, based on the
assumption that the wheelbase is much smaller than the radius of the trajectory
and all angles are small. Since in slip steering the sideslip angles are usually
larger than in the case of conventional steering, the linearized solution holds
only for very large radii. The equations of motion are still Equations (31),
where the steering angle is = 0 and a control yaw torque due to the differen-
tial longitudinal forces is added. From equation (25.129) and (25.134) in [2] it
follows
 17

Conventional and slip steering for multi-wheel planetary rovers 23

1 YB

R
n
ti $Fx
;
V N B Yr mV N r YB =
i 1

B

Yr mV (49)
n
N B Yr mV N rYB
ti $Fx
i 1

where the derivatives of stability are given by Eq. (32).

.3.1. Response to a lateral force. A particularly simple solution is obtained

for the case of a vehicle travelling on a straight trajectory on a laterally sloping
ground. In this case it is possible to assume that the yaw velocity r = 0 and the
yaw angle is constant. The equations of motion reduce to

mg sin A sin Y  2nFx

n 2 (50)
mg sin A cosY  Fyij
i 1 j 1

;x F =
n n 2
ti $Fx  i yij  M zij  0
i 1 i 1 j 1

Since the wheels are not steering, all sideslip angles are equal
v
A ij  artg (51)
u

which causes also all cornering forces and aligning moments to be equal. The
equations of motion thus yield

mg sin A sin Y
Fx 
2n
mg sin A cosY (52)
Fyij 
2n
n
2 Fyij xi  2nM zij
i 1
$Fx  n
.
ti
i 1
 18

24 Giancarlo Genta

From the first equation it is possible to compute the average longitudinal

force, from the second the lateral forces and then the sideslip angles and from
the third one the differential longitudinal force. The lateral velocity is immedi-
ately computed from the sideslip angles.
5. Examples
5.1 Mars exploration rover
5.1.1. Description of the model. Consider a six-wheeled Mars exploration
rover (MER) whose data are reported in Tab.1. The suspensions are such that
the vertical forces on the ground are equally distributed on all wheels, when
there is no load shift due to inertia forces.
Tab. 1. Data for Example 1 (Mars exploration rover)
Mass (kg) m = 180
gravitational. acceleration. (m/s2) g = 3.77
wheelbase (m) l = 1.4
position of axles (m) x = [0.6, -0.1, -0.8]
track (all axles) (m) t = 1.9

The center of mass is located slightly forward the central wheels: this is
an arbitrary assumption, taken to avoid the unlikely assumption that the rover
is centred exactly on the second axle.
Simple expressions were assumed for the lateral characteristics of the
wheels:
C
A ij
 
Fyij  f1ij A ij  sgn A ij M y max Fzij 1 e
M y max Fzij

C
A ij
 
M zij  f 2ij A ij  sgn A ij M 0ij 1 e
M y max Fzij
e
C1 A ij
.

The following values of the parameters were assumed: ymax = 0.6, C = 1100
N/rad, Moij = 1.1 Nm and C1 = 20. The forces and moments are reported as
functions of the sideslip angle in Fig. 3.
By differentiating the force and moment with respect to the sideslip angle,
the cornering and aligning stiffness are obtained
uFyij uM zij CM 0ij
 C  1100 N/rad,   17.85 Nm/rad.
uA ij uA ij M y max Fzij
 19

Conventional and slip steering for multi-wheel planetary rovers 25

Fig. 3. Cornering forces and aligning torques as functions of the sideslip angle

5.1.2. Steering response. The kinematic steering angles of the various wheels
are reported as functions of the radius of the trajectory in Fig. 4. The plot was
computed with the assumption of minimum radius, obtained by stating
l
D1  D 2  artg .
4R b a 2
2
 (34)

Fig. 4. Steering angles of the 6 wheels as functions of the trajectory curvature radius (the central
wheels do not steer)

The usual linearized value for the monotrack vehicle, obtained by assuming
that the radius of the trajectory is much larger than the wheelbase, is also
shown (dashed line). The difference between kinematic and dynamic steering in a
speed range up to 6 m/s is shown in Fig. 5. The curves were computed for speeds
giving way to a centrifugal acceleration lower than the ideal maximum value :
ymaxg.
26 Giancarlo Genta

Fig. 5. a) and b): lateral velocity v and yaw velocity r as functions of the speed for different values of
the kinematic curvature radius of the trajectory. c) and d): sideslip angle and ratio between the dynamic
and kinematic radius of the trajectory as functions of the speed

Fig. 6. Sideslip angles of the wheels on a trajectory with a radius of 5 m as functions of the speed
 Conventional and slip steering for multi-wheel planetary rovers 27

Fig. 7. a) and b): lateral velocity v and yaw velocity r as functions of the speed for different values of
the lateral slope of the ground. c) and d): sideslip angle and dynamic radius of the trajectory as
functions of the speed

Fig. 8. Sideslip angles of the wheels as functions of the speed ( = 10°)

From the radius of curvature of the trajectory it is clear that the rover is
understeering, since the dynamic radius increases with the speed. The lateral
velocity v and the yaw angular velocity r increase linearly with the speed in
case of kinematic steering, while varying in a nonlinear way for dynamic
steering.
28 Giancarlo Genta

The sideslip angle and the curvature radius of the trajectory do not de-
pend on the speed in kinematic steering, while being a function of the latter
in dynamic conditions. The sideslip angle, that in the present case is positive
in kinematic conditions, decreases with the speed, to become negative at
higher speed. When the speed tends to 0 the conditions tend to the kinematic
conditions.
The sideslip angles of the wheels related to a curvature radius of the tra-
jectory of 5 m are reported as functions of the speed in Fig. 6.
The sideslip angles are quite large even when the speed is low (a sideslip
angle of 2° is already considered as large): this is due to a combined effect
of the low trajectory radius and low gravity.

5.1.3. Response to a lateral force. Assume that the rover travels along the
direction of the X axis with the wheels locked in longitudinal direction (all
ij = 0) and that the ground slopes in the direction of Y axis (downwards in
the direction of the negative axis). The kinematic solutions yields a straight
path in the X direction, a thing that was already said to be physically impos-
sible.
The solution of the equations for dynamic steering in terms of v, r, and
R are reported in Fig. 7. The computations were performed for various
values of the lateral inclination of the ground. Fig. 7a) shows that the rover
slides sideways downwards, with a lateral velocity that is not negligible if
compared with the longitudinal speed, particularly when the latter is low.
Fig. 7b) shows that the trajectory bends too, in the direction of the –Y axis.
This could be expected, since this is the typical behavior of an undesteer-
ing vehicle, as obtained from the usual linearized model.
Angle is not negligible, particularly at low speed, and the radius of the
trajectory, that is large if is small, can be of some tens of meters if is
large.
The sideslip angles of the wheels on a slope with = 10° are reported in Fig. 8.

5.1.4. Slip steering – directional behaviour. Since the differential longitu-

dinal forces needed to perform slip steering may be quite high on sharp
corners, the interaction between longitudinal and cornering forces was
accounted for. As no detailed characteristics of the wheel were available, the
simple elliptical approximation
2
Fx
*
Fyij  Fyij 1
M x max Fz
was used to compute the actual side force F*y from the value Fy obtained
neglecting the interaction. No interaction was considered for the aligning
torque, owing to its small effect on the results.
 Conventional and slip steering for multi-wheel planetary rovers 29

Fig. 9. Slip steering: comparison between the linearized solution and the nonlinear one on a radius of
1000 m. a) and b): lateral velocity v and yaw velocity r as functions of the speed. c) and d): sideslip angle
and differential longitudinal force exerted by each wheel as functions of the speed

Fig. 10. Sideslip angles of the wheels as

functions of the speed (R = 1000 m)

As a first point the linearized response is computed and compared with

the nonlinear response on a radius of 1000 m. The results are reported in
figg. 9-10.
From Fig. 9 it is clear that the linearized and the nonlinear solutions are
quite close: for a complete accordance a larger radius, of about 10,000 m
30 Giancarlo Genta

must be used. The sideslip angle of the vehicle (Fig. 9c) and those of the
wheels (Fig., 10) are quite small, and, starting from a speed 5 m/s they have
the same sign, which mean that all side forces of the wheels act in the same
direction, i.e. against the centrifugal force.
Notice that the differential force of the wheels is quite small, and the in-
teraction between longitudinal and lateral forces (not accounted for in the
linearized solution, but accounted for in the nonlinear one) is marginal.
When travelling on a bend with smaller radius the required differential
force grows, as do the sideslip angles (Figg. 11-12). When the curve be-
comes very sharp the sideslip angles become quite large and the wheel
works in a zone of its characteristics where the sideforce is almost constant
with the sideslip angle. Due to this fact and also to the interaction between
longitudinal and cornering forces, the differential longitudinal force needed
to keep the vehicle on its trajectory does not increase much.
As shown in Fig.12, on a radius of 2 m the sideslip angles of the front
wheel remain positive, i.e. they generate a force which is opposite to those
of the other wheels.

Fig. 11. a) and b): lateral velocity v and yaw velocity r as functions of the speed for different values
of the radius of the trajectory. c) and d): sideslip angle and differential longitudinal force exerted by
each wheel as functions of the speed
 Conventional and slip steering for multi-wheel planetary rovers 31

Fig. 11. Sideslip angles of the wheels as functions of the speed on trajectories with different radii

5.1.5. Slip steering – response to a side force. Assume again that the rover
travels along the direction of the X axis and that the ground slopes in the
direction of Y axis (down in the direction of the negative axis). The equa-
tions of motion have been solved assuming that a differential force is ap-
plied so that the trajectory of the rover is straight.
The results in terms of the sideslip angles of the vehicle and the wheels
(which in this case coincide since the yaw velocity vanishes) and of the
differential longitudinal force needed to obtain a straight trajectory are
reported in Fig. 13.
In this case the yaw angle is constant, and thus the solution is a mean-
ingful one. If = 0, as assumed, the vehicle x axis is parallel to the fixed X
axis, and hence the velocity v makes an angle with the direction perpen-
dicular to the slope. This means that the trajectory is inclined downwards of
the same angle, measured on the sloping ground.
To proceed straight in a direction perpendicular to the slope, x axis must
point upwards of an angle ; however in this case 0 and the computation
should be repeated to take into account the (small) longitudinal force the
wheels must exert.
32 Giancarlo Genta

Fig. 13. Sideslip angles of the wheels and of the vehicle and differential force on the wheels as func-
tions of the lateral slope of the ground

5.2 Four-wheeled lunar buggy

5.2.1. Description of the model. Consider a light human-carrying unpres-
surized rover similar to the LRV (Lunar Roving Vehicle) of the last 3 Apollo
missions. The relevant data are reported in Tab. 2.

Tab 2. Data for Example 2 (lunar buggy)

mass (fully loaded) (kg) m = 660
gravitational acceleration (m/s2) g = 1.62
wheelbase (m) l = 2.3
position of axles (m) x = [1.2, -1.1]
track (all axles) (m) t = 1.8
max. lateral force coefficient -- ymax = 0.6
cornering stiffness N/rad C = 5000
parameter for aligning moment Nm/rad Moij = 2
parameter for aligning moment -- C1 = 20

The center of mass is located slightly behind the mid-wheelbase point.

The same simple expressions used in the previous example were assumed
for the lateral characteristics of the wheels. The forces and moments are
reported as functions of the sideslip angle in Fig. 14.
The cornering and aligning stiffness are obtained by differentiating the
force and moment with respect to the sideslip angle
 Conventional and slip steering for multi-wheel planetary rovers 33

uFyij uM zij CM 0ij

 C  5000 N/rad,   62.5 Nm/rad.
uA ij uA ij M y max Fzij

Fig. 14. Cornering forces and aligning torques as functions of the sideslip angle

5.1.2. Steering response. The kinematic steering angles of the various

wheels are reported as functions of the radius of the trajectory in Fig. 15.

Fig. 15. Steering angles of the 4 wheels as functions of the trajectory curvature radius
34 Giancarlo Genta

The plot was computed with the assumption of minimum radius, obtained
by stating
(34)
l
D1  D 2  artg .
4R2 b a 2 
The usual linearized value for the monotrack vehicle, obtained by assum-
ing that the radius of the trajectory is much larger than the wheelbase, is also
shown (dashed line).
The difference between kinematic and dynamic steering in a speed range
up to 10 m/s is shown in Fig. 16. The curves were computed for speeds
giving way to a centrifugal acceleration lower than the ideal maximum value
ymaxg. The maximum speed at which the vehicle can travel on a curved
trajectory is low, owing to low gravity.

Fig. 16. a) and b): lateral velocity v and yaw velocity r as functions of the speed for different values
of the kinematic curvature radius of the trajectory. c) and d): sideslip angle and ratio between the
dynamic and kinematic radius of the trajectory as functions of the speed.

The rover is oversteering, since the dynamic radius decreases with the
speed, as could be expected from the position of the center of mass.
The sideslip angle and the curvature radius of the trajectory do not de-
pend on the speed in kinematic steering, while being a function of the latter
in dynamic conditions. The sideslip angle is negative also in kinematic
conditions and further decreases with the speed. When the speed tends to 0
the conditions tend to the kinematic conditions.
 Conventional and slip steering for multi-wheel planetary rovers 35

The sideslip angles of the wheels due

to a curvature radius of the trajectory
of 5 m are reported as functions of
the speed in Fig. 17. As expected
from an oversteering vehicle the
sideslip angles of the rear wheels are,
as an average, larger than those of the
front wheels, but owing to the short
radius of the trajectory the difference
between the sideslip angles of the
right and left wheels is large.
Fig. 17. Sideslip angles of the wheels on a
trajectory with a radius of 5 m as functions of the
speed

5.1.3. Slip steering – directional behaviour. Since the differential longitu-

dinal forces needed to perform slip steering may be quite high on sharp
corners, the interaction between longitudinal and cornering forces was
accounted for using the same elliptical approximation seen in the previous
example. No interaction was again considered for the aligning torque. The
linearized response was first computed and compared with the nonlinear
response on a radius of 10000 m. The results are reported in figg. 18-19.

Fig. 18. Slip steering: comparison between the linearized solution and the nonlinear one on a radius
of 10000 m. a) and b): lateral velocity v and yaw velocity r as functions of the speed. c) and d): sideslip
angle and differential longitudinal force exerted by each wheel as functions of the speed
36 Giancarlo Genta

Fig. 19. Sideslip angles of the wheels as functions of the speed (R = 10000 m)

The linearized and the nonlinear solutions are close but not identical: to
obtain a complete accordance an even larger radius, up to 100 km, must be
used. The conclusions regarding the sideslip angle of the vehicle and those
of the wheels are the same as for the previous example, while the differen-
tial force of the wheels decreases with the speed instead of increasing: this is
due to the fact that this vehicle is oversteering while that of the previous
example was understeering. The behaviour on a bend with smaller radius is
shown in figg. 20-21. The required differential force grows, but only up to a
limit. The large interaction between longitudinal and lateral forces, due
mainly to the low value of gravity, allows the wheels to slip sideways with-
out requiring large yaw torques, thus reducing the requirements for the
differential force.

Fig. 20. a) and b): lateral velocity v and yaw velocity r as functions of the speed for different values
of the radius of the trajectory. c) and d): sideslip angle and differential longitudinal force exerted by
each wheel as functions of the speed
 Conventional and slip steering for multi-wheel planetary rovers 37

The sideslip angles become very large on the rear wheels, as could be
expected in an oversteering vehicle.

Fig. 21. Sideslip angles of the wheels as functions of the speed on trajectories with different radii

6. Conclusions
Three simple mathematical models for the study of the lateral (direc-
tional) behaviour of wheeled rovers for planetary explorations have been
developed. The models are based on
– kinematic steering,
– dynamic steering obtained by steering some (or all) wheels,
– dynamic steering obtained by slip steering.
The first approach is quite common in studies on the control of wheeled
robots. The sideslip (and also the longitudinal slip, which is however of
little importance in the present study) is neglected, and the wheels are mod-
elled as nonholonomic constraints. As it is well known in motor vehicle
technology, this approach leads to large inaccuracies in predicting the trajec-
tory at high speed, since it neglects completely the over- or understeering
behaviour of the vehicle. However, this approach is inaccurate also at low
speed (as a limit, even when the speed tends to zero) when the ground has a
lateral slope. The models based on assuming that the wheels are non-
holonomic constraints can thus be used only in structured environments,
where the robot travels on perfectly smooth and flat terrain, while in un-
structured environment, like that found in off-road motion on Earth and in
planetary exploration, they are not applicable.
38 Giancarlo Genta

The second model takes into account the sideslip of the wheels, and then
requires a good knowledge of their characteristics, particularly for what the
cornering forces are concerned. The usual assumptions of large trajectory
radius (large if compared with the wheelbase of the vehicle) and small
sideslip angles are dropped, yielding a fully nonlinear model. This is due to
the fact that, while in automotive technology the lateral behaviour is impor-
tant at high speed, which can be reached only on large radii, in planetary
exploration speeds are at best moderate (usually very low) and narrow radii
are needed to manoeuvre in unstructured environments littered with obsta-
cles.
In the examples independent steering with a steering controller able to
realize a perfect Ackerman steering of all wheels is assumed: the wheels
thus steer in a way that makes it possible to obtain kinematic steering (when
the speed tends to 0 and no lateral force is present). The model is however
able to include any steering control law and thus also the case of coordi-
nated steering (which is however seldom considered in the case of planetary
rovers).
In the third model slip steering is considered. In this case the sideslip an-
gles of the wheels can be quite large, up to almost 90° when turning on the
spot. The presence of (differential) longitudinal forces, required for steering,
that may be larger than the lateral forces, makes it compulsory to take into
account the interaction between longitudinal and lateral forces and thus
requires a knowledge of the behaviour of the wheel that is deeper than that
needed for the previous models.
Slip steering is particularly appealing for planetary rovers since it allows
a large simplification in the rover architecture (both of the electromechnical
layout and of the control system, since no steering wheels are required) and
provides a good handling, up to turning the vehicle on the spot. The simula-
tions did show that, owing to low gravity and to interaction between the
longitudinal and lateral forces, the differential longitudinal forces needed to
control the trajectory remain fairly low even on sharp bends (even when
turning on the spot). The main objections against slip steering - requiring
large driving power, producing large wheel wear and stresses in the suspen-
sions and the vehicle body – are shown to be oversted, particularly in low
gravity environments.
The models are largely nonlinear, which prevent from obtaining closed
form solutions or defining gains, as it is common when linearized vehicle
dynamics models are used.
The general models for non-steady state operation can be easily inte-
grated numerically in time, even if the presence of algebraic loops suggests
to take the usual precautions.
 Conventional and slip steering for multi-wheel planetary rovers 39

The models for steady state operations yield sets of nonlinear algebraic
equations. The Newton-Raphson method, with a numerically (or partially
numerically and partially analytically) computed Jacobian matrix, proved to
converge to the required solution in a small number of iterations, at least in
the first 2 cases. When dealing with slip steering some convergency prob-
lems were found, probably linked with the larger sideslip angles that cause
the wheels to work in a more nonlinear part of their characteristics. These
problems were fully solved by slowing down the convergency by multiply-
ing the inverse of the Jacobian matrix by a factor smaller than one.
References
[1] G. GENTA, Motor Vehicle Dynamics, World Scientific, Singapore, 2004.
[2] G. GENTA, L. MORELLO, The Automotive Chassis, Springer, New York,
2009.
[3] D.S. APOSTOLOPOULOS, Analytical Configuration of Wheeled Robotic Loco-
motion, CMU-RI-TR-01-08, The Robotics Inst., Carnegie Mellon University,
Pittsburgh, April 2001.
[4] G. GENTA, Introduction to the Mechanics of Space Robots, Springer, 2012.
[5] A. ELLERY, An Introduction to Space Robotics, Springer Praxis, Chichester,
2000.
[6] Lunar Roving Vehicle Operations Handbook, http://www.hq.nasa.gov/
office/pao/History/alsj/lrvhand.html.
[7] G. ISHIGAMI, A. MIWA, K. YOSHIDA, Steering Trajectory Analysis of Plane-
tary Exploration Rovers Based on All-Wheel Dynamic Model, Proc. Of the 8th
Int. Symp. on Artificial Intelligence, Robotics and Automation in Space, Mu-
nich, Sept. 2005.
[8] G. ISHIGAMI, K. YOSHIDA, Steering Characteristics of an Exploration Rover
on Loose Soil Based on All-Wheel Dynamic Model, 2005 IEEE/RSJ Int.
Conf. on Intelligent Robots and Systems, Edmonton, Aug. 2005.
[9] M.R. FLANNERY, The Enigma of Nonholonomic Consraints, Am. Journ. Of
Physics, Vol. 73, n. 3, 2005, p. 265-272.
[10] O.M. MORESCHI, G. CASTELLANO, Geometric Approach to Non-holonomic
Problems Satisfying Hamilton’s Principle, Revista de la Union Matematica
Argentina, Vol. 47, n. 2, 2006, pp. 125-135.
[11] E. BAKKER, L. LIDNER, H.B. PACEJKA, Tire Modelling for Use in Vehicle
Dynamics Studies, SAE Paper 870421
[12] E. BAKKER, H.B. PACEJKA, L. LIDNER, A New Tire Model with an Applica-
tion in Vehicle Dynamics Studies, SAE Paper 890087.
[13] H.B. PACEJKA, Tire and vehicle dynamics, Elsevier, New York, 2006.
 40 Giancarlo Genta

[14] K. YOSHIDA, G. ISHIGAMI, Steering Characteristics of a Rigid Wheel for

Exploration on Loose Soil, 2004 IEEE/RSJ Int. Conf. on Intelligent Robots
and Systems, Sendai, Oct. 2004
[15] G. ISHIGAMI, A. MIWA, K. NAGTANI, K. YOSHIDA, Terramechanics Based
Model for Steering Maneuver of Planetary Exploration Rovers on Loose Soil,
Journ. Of Field Robotics, vol. 24, n. 3, 2007, pp. 233-250.
[16] T.B. RHYNE, S.M. CRON, Development of a Non-Pneumatic Wheel, Tire
Science and Technology, Vol. 34, n. 3, pp. 150-169 (September 2006).

Symbols
a distance between the center of mass and L Lagrangian function
the front axle
Qi ith generalized force
b distance between the center of mass and
R radius of the trajectory
the front axle
S Jacobian matrix
g gravitational acceleration
T kinetic energy
l wheelbase
U potential energy
m mass
V velocity of the vehicle
n number of axles
X, Y axes of the inertial frame
p vector function
X configuration vector
r yaw angular velocity
lateral slope angle of the ground
rz radius of gyration
ij sideslip angle of the ijth wheel
qi ith generalized coordinate
sideslip angle of the vehicle
ti track of the ith axle
steering angle
u component of velocity V along x axis
qi ith virtual displacement
v component of velocity V along y axis
W virtual work
x, y axes of the body-fixed frame
Lagrange multiplier
xi coordinate of the ith axle
force coefficient
C cornering stiffness
yaw angle
F force
F differential force
Jz yaw moment of inertia

View publication stats