An Adaptive Suspension System for

Planetary Rovers
G. Reina ∗ M. Foglia ∗∗
∗
Department of Engineering for Innovation, University of Salento,
73100 Lecce, Italy (e-mail: giulio.reina@unisalento.it)
∗∗
Department of Management and Mechanical Engineering,
Polytechnic of Bari, 70125 Bari, Italy

Abstract: This paper presents an innovative suspension system with variable wheel camber to
improve mobility of robots on rough-terrain. The system is optimized for planetary rovers that
employ conventional rocker-type suspensions. The main advantage of the proposed system is
that each wheel keeps an upright posture as the suspension system adapts to terrain unevenness,
maximizing tractive and climbing performance, and reducing energy consumption. The synthesis
of the variable camber mechanism is described along with details of the mechanical design,
showing the feasibility of this solution.

Keywords: Mobile robotics, rough-terrain mobility, chassis design, mechanism synthesis.

1. INTRODUCTION According to the rocker-type suspension design and with

reference to the four-wheeled robot shown in Fig. 1, the
drive wheels of either side of the vehicle are connected
Over the past two decades, there has been increasing
longitudinally by a rigid link, called rocker arm. Each one
interest in autonomous robots operating outdoor in chal-
of the two side rockers is pivoted to the main frame of
lenging environments. Numerous notable examples can be
the vehicle through the semi-axis of a differential gearbox,
found in literature. The NASA rovers, Spirit and Oppor-
whose input gear is attached to the chassis. This archi-
tunity, were launched from Earth in July 2003 and are
tecture allows the rocker arms to pivot when one wheel
still operating on the surface of Mars (Biesiadecki et al.
tries to go higher or lower than the rest. The passive
(2007)). The Demo-III eXperimental Unmanned Vehicle
pivoting action evens out the load on all four wheels,
(XUV) (Lacaze et al. (2002)), funded by the U.S. Army,
increasing mobility simply by maintaining driving and
served as a test bed for both hardware and software of
braking action on all wheels at all times. Moreover, the
outdoor robotic systems, with an emphasis on realistic
mechanical differential constraints the pitch angle of the
scenarios, such as those faced by military vehicles. The
chassis to be half of the pitch angle of either side rocker.
DARPA Grand Challenge tasked international teams of
This averaging action effectively reduces the pitching mo-
engineers with developing wheeled mobile robots capable
tion of the chassis, maintaining it at a more level pose, as
of traversing 200 kilometers of rough desert terrain in less
either side of the suspension system travels over obstacles.
than 10 hours (Buehler et al. (2005)). This competition
Overall, the rocker suspension system provides remarkable
has helped to accelerate the development of robust sensors
mobility over obstacles, allowing the rover to safely tra-
and computational algorithms for robot motion planning
and navigation. In principle, outdoor robots face all of the
same challenges as indoor robots, such as sensing, data
processing, locomotion, navigation, and interaction with
the surroundings. Outdoor robots, however, are expected
to achieve all of these tasks in much more complex and
unstructured environments such as forests, deserts, and
agricultural fields. While most researchers have focused
on sensors and algorithms to approach these issues, there
has been relatively little work in addressing changes to
the structure of the vehicle. NASA, in collaboration with
the Jet Propulsion Laboratory (JPL), proposed an artic-
ulated suspension system, referred to as rocker-bogie, for
planetary rovers in both the Mars Pathfinder and Mars Ex-
ploration Rover missions (Bickler (1998), Lindemann and
Voorhees (2005)). This solution has been demonstrated
effective in enhancing mobility and traction performance
on uneven terrain for vehicles moving at relatively low Fig. 1. The rover Dune, built at the University of Salento:
velocities (less than 20 cm/s), suggesting the development an example of 4-wheel-drive/4-wheel-steer robot em-
of alternative embodiments (Lamon and Siegwart (2007)). ploying a rocker-type suspension system
verse rocks over one and half its wheel diameter, whereas not change independently of these. There are, however,
a 4-wheeled vehicle with conventional suspension may only systems which have been designed to provide an additional
be able to climb one half its wheel diameter (Foglia and degree of freedom for camber independent of the motion
Reina (2008)). However, one critical limitation to the use of other vehicle components. In Bobbitt et al. (2007) the
of rocker-type suspensions is that they do not provide authors have created a mechanism by which the camber
any compensation of the wheel inclination angle relative angle of the wheel can be varied. The adjustment must
to the ground, i.e. the camber angle, during suspension be done manually, however, and only while the vehicle
travel. This causes the wheels to deviate from their optimal is not moving. An electronically-controlled active camber
posture, corresponding to a perfectly perpendicular con- system has more recently been seen on the Mercedes-Benz
figuration, as the robot adapts to terrain unevenness, thus F400 Carving concept car (Mackle and Schirle (2002)).
reducing the overall traction performance and promoting The purpose of the variable camber on the F400 is to
undesired effects, such as wheel slip and sinkage, which allow different parts of the specially-made tire to contact
greatly affect a robot’s mobility. In addition, vehicle yaw is the ground during cornering, thus improving performance.
also induced producing unpredictable deviations from the Serra and Tetaz (2004) have taken a different approach
intended path. In principle, an upright wheel maximizes its by creating an automobile suspension system with passive
footprint, reducing ground pressure and, therefore, sinkage camber motion which is substantially independent of the
in soft terrain (Wong (2001)). At the same time, the “grip” suspension travel. The purpose of the device is to allow
of the wheel, i.e. its ability to exchange forces with the the tire to maintain proper contact with the ground even
ground, increases, thus reducing slippage. in the presence of large suspension travel, body roll, or
In this work, a variable camber mechanism is proposed intense cornering.
that can be easily integrated with conventional rocker Here, an adaptive system is proposed, optimized for
suspension systems. The idea is to obtain an adaptive rocker-type suspensions, to improve mobility of autonomous
suspension system allowing the robot wheels to adjust robots operating at moderate speeds on highly-challenging
their inclination angle in order to hold a perpendicular terrain. The system employs a wheel camber mechanism,
posture with respect to the ground. This ensures that which controls passively the wheel inclination angle as
wheels operate at peak efficiency with a regular pressure function of the suspension travels to operate each wheel
distribution and a maximum footprint area at all times, at its peak-efficiency condition at all times.
improving the overall mobility and power efficiency of the The paper is organized as follows. Section 2 describes the
robot. Although, wheel camber change plays an important behavior of a conventional rocker-type rover using a kine-
role in rough-terrain mobility, relatively little research has matic model for articulated vehicles. Section 3 introduces
been devoted to this issue in the robotics community. Wal- the proposed wheel cambering mechanism along with its
dron (1994) focused on the issue of the so-called kinematic geometrical optimization. Relevant conclusions are drawn
slip. In addition to dynamic slippage, due to terrain defor- in final Section 4.
mation or insufficient friction, a wheeled robot is affected
by kinematic slip as well, which occurs when there is no 2. KINEMATIC MODELING
instantaneous axis of rotation compatible with all of the
robot’s wheels, resulting in undesired wheel “scrubbing”. In this section, a kinematic model is introduced to study
Sreenivasan and Nanua (1999) proposed robot designs the geometrical configuration of a conventional rover dur-
to potentially reduce this issue in wheeled vehicles and ing traversal of uneven terrain. For a vehicle with m unique
suggested the use of a Variable Length Axle (VLA) with wheel-terrain contact points, at least m − 1 kinematic loop
either an unactuated or active prismatic joint to achieve closure equations can be written. When considering the
the necessary motion. As a more practical alternative to rover shown in Fig. 2, these equations can be obtained as
the VLA, Chakraborty and Ghosal (2004) introduced the P4z = P3z − 2 · l · cosφ · sin βR (1)
idea of adding an extra degree of freedom at the wheel-
axle joint, allowing the wheel to tilt laterally relative to the P4z = P2z − cos φ · (t · (cos βR − cos βL ) +
axle. This new capability, named Passive Variable Camber
(PVC), allowed the distance between the wheel-ground l · (sin βR − sin βL )) − 2 · w · sin φ (2)
contact points to change without any prismatic joints. On
a real vehicle, the PVC joints would be passively actuated P4z = P1z − cos φ · (t · (cos βR − cos βL ) +
by lateral forces at the wheel-ground interface arising from
interactions between the two surfaces. The idea of variable l · (sin βR + sin βL )) − 2 · w · sin φ (3)
camber is not new in the automotive field. Passenger auto- where Piz refers to the z component of the wheel contact
mobiles feature changing camber in response to suspension point i, φ and θ are the roll and the pitch of the robot
travel, body roll, and forces generated during cornering chassis, βR and βL the right and left rocker arm angles,
(Genta (2003)). Camber angles can also be deliberately and l, w, and t geometrical parameters as explained in Fig.
modified to affect vehicle performance. For example, neg- 2. Note that all the parameters are measured in the ground
ative camber on the two front wheels of a 4-wheeled auto- reference system. Due to the presence of the mechanical
mobile promotes straight-line stability. On the other hand, differential in this system, an additional equation can be
positive camber leads to an inherently safe under steer be- written relating the pitch θ to the angles βR and βL ,
havior. The camber angle variations in standard 4-wheeled βR + βL
vehicles are usually passive (not actuated), and depend on θ= (4)
2
other motions. For example, the camber angle is a function
of the suspension travel or of the body roll, and does The inverse kinematics problem expressed by (1)–(4) in-
volves computing vehicle’s tilt (φ, θ) and the configuration
 Table 1. Dune’s properties
Dimensions
0.74 × 0.45 × 0.32 m
(2l × 2w × t)
Wheel diameter 0.2 m
Ground clearance 0.21 m
Total mass 16 kg
Structure 10 kg
Batteries 2 kg
Motors 2 kg
Computer System 2 kg
Max speed 150 cm/s

wheel camber change. Figure 5 shows a real scenario where

Dune climbed a 160 mm-heigh obstacle. According to our
model, the robot’s body is tilted with a pitch and roll
angle in absolute value of 11.5 and 8.5 deg, respectively. It
Fig. 2. Coordinate frames and nomenclature for the rover is apparent the undesired cambering of all four wheels. The
Dune contact patches are largely reduced, affecting the overall
of the rover suspension (βR , βL ), given the shape of the ability to exchange tangential forces with the ground, and,
terrain (i.e., Piz for each wheel). For simplicity’s sake, it thus, to accelerate, brake, balance lateral loads, and climb
is assumed that the ground can be represented instanta- obstacles. Therefore, a variable camber system, aiming to
neously by four discrete horizontal planes with different keep the wheels upright throughout suspension travels,
altitudes. would be greatly beneficial for all-terrain rovers.
2.1 Step-climbing
3. VARIABLE CAMBERING MECHANISM
As a practical scenario that we will use as a running and
general example throughout the paper, let us consider the A wheel reference frame can be defined with origin at the
case of obstacle-climbing, when one wheel of the rover, center of wheel-terrain contact patch, axis Xw coincident
i.e., right front Wheel 3, rises higher than the other three with the intersection between the wheel longitudinal plane
wheels that remain in contact with the ground plane. and the ground and pointing towards the rolling direction,
The geometrical parameters used in the simulation are axis Zw perpendicular to the ground and directed upward,
collected in Tab. 1 and they refer to the rover Dune, and Yw defined accordingly such as to create a right-hand
built at the University of Salento in collaboration with reference frame. The angle between the plane Xw −Zw and
the Polytechnic of Bari (see also Fig. 1). The rock is the longitudinal wheel plane is the so-called inclination
modeled for simplicity as a step-obstacle, which, however, angle γ, defined positive when counter-clockwise (refer to
represents a worst-case condition. The rock height ho is Fig. 6). In the automotive field, γ is usually referred to
increased progressively up to 200% of the wheel radius. As as the camber angle according to the SAE terminology
a consequence, the rover is subject to tilt rotations that can (SAE J670e (1976)). Note, however, that here a different
be obtained solving the nonlinear system of equations (1)– sign convention is adopted. In general, for a four-wheeled
(4). The results are collected in Fig. 3, where the change vehicle, the inclination of the wheels changes during roll
in roll, pitch and right rocker angle is plotted as function
of the obstacle height. We refer to the final stage of the
climbing maneuver when Wheel 3 is on the top of the
step. Note also that roll, pitch and rocker arm angle are
negative, but here, for comparative reasons, their absolute
value is plotted.
It is clear the averaging effect due to the mechanical
differential that reduces the rover’s pitch to half of the
irregularities seen by the rocker arm on the ground. The
roll movement is also reduced by the rocker suspension
compared to a rigid-frame robot but by a lesser amount
than pitch. In order to get a more general sense of the
rover lateral kinematics, we can plot the roll angle φ as
function of the relative rotation between the right rocker
r
arm and the robot’s chassis βR = βR −θ, as shown in Fig. 4.
Any rotation of the rocker arm with respect to the chassis
indicates the presence of correspondent robot’s roll. In our
example, the right rocker arm rotates counter-clockwise
with respect to robot’s body to accommodate the upward
travel of Wheel 1 during the climbing stage. The left rocker
arm keeps parallel to the ground, instead. This results in Fig. 3. Change in robot’s tilt as function of the obstacle
a negative roll rotation of the robot and, in turn, in a height
 Fig. 6. Wheel camber

relative camber to contrast the effect of body roll and

to maintain an optimal posture even on highly-irregular
terrain. This is achieved using a simple driving linkage,
Fig. 4. Change in roll angle as function of the rocker arm- which controls the cambering motion of the wheels of
chassis relative rotation the same side of the robot during relative rotations with
respect to the chassis. The proposed adaptive rocker
motion. This mechanism is shown in Fig. 7. The inclination suspension system is shown in the CAD drawing of Fig. 8.
angle γi of wheel i, i=Right, Left, can be expressed as In contrast with the rigid connection of the original design
γi = γc,i + φ (5) (compare with Fig. 1), each rocker arm is now attached to
the semi-axis of the differential via a revolute joint, which
where φ is the roll angle of the vehicle, and γc,i the allows it to pivot around its longitudinal axis, modifying
camber angle of the given wheel in its relative motion with the inclination angle of the attached wheels. The variable
respect to the body, which depends on the geometry of cambering linkage is shown in more detail in the close up
the suspension. From an analysis of the rocker suspension of the lower part of Fig. 8. Link 1 and Link 2 form the
kinematics, the derivative of the inclination angle with roll driving linkage that controls the wheel inclination change
angle can be determined in the vicinity of the equilibrium in the right rocker-chassis relative motion. An equivalent
configuration equal to mechanism operates on the left side. Link 1 is attached
∂γi to Link 2 and to the chassis through two spherical joints,
=1 (6) whereas Link 2 is rigidly connected with the rocker arm.
∂φ
Link 1 and Link 2 serve as crank and coupler of the driving
This derivative value outlines one disadvantage of the mechanism. This can be better seen when considering the
rocker suspension system, that is any roll angle translates projection of the linkage in the transverse plane of the
into an equivalent change in wheel camber. For this reason, robot, as shown in Fig. 9, which allows one to analytically
the more wheels deflect from their static position, the more study the system referring to a simple model without
difficult it is to maintain an optimal camber angle (see Fig. losing much of accuracy. Input to the linkage is the vertical
5). In theory, a null derivative value, i.e. roll and relative displacement of the slider B that is produced by the
wheel camber change equal and opposite, would ensure the rotation of the right rocker link with respect to the robot’s
best performance. This requires an appropriate suspension body. The linkage reacts by changing the angle of Link
design with a wheel camber change increasingly positive 2, i.e. the wheel inclination angle, which we can consider
as the suspension deflects upward. as the output of the mechanism. In order to maximize
In this paper, an innovative design of rocker suspension the footprint area so that optimal tire tractive effort and
system is proposed, enabling the wheels to modify their even pressure pattern can be obtained, the wheel needs to
remain upright as the suspension complies with terrain
unevenness. The proposed mechanism fulfils the design
requirement that the wheel gains a positive camber during
its upward travel and a negative camber during downward
travel. Wheel camber is assumed equal to zero when the

Fig. 5. Wheel camber change due to a 160 mm-high rock Fig. 7. Wheel camber mechanism
 l1 · sin θ1 + l2 · sin θ2 = d (8)
l1 · cos θ1 + l2 · cos θ2 = h + ∆yB (9)

By squaring both sides of (8)–(9), adding and rearranging

the result, it is possible to get an explicit equation of θ2
as a function of the system input
k1 · cos θ2 + k2 · sin θ2 + k3 = 0 (10)

where
k1 = 2 · l2 · (h + ∆yB ) (11)
k2 = 2 · d · l2 (12)
k3 = l12 2
− (d + l22 + (h + ∆yB ) )2
(13)

Equation (10) can be solved using the standard trigono-

metric identities for half angles, one gets
(k3 − k1 ) · t2 + 2 · k2 · t + (k1 + k3 ) = 0 (14)
θ2
t = tan (15)
2
Fig. 8. The adaptive rocker suspension system using a
variable cambering mechanism Finally, solving for t gives
p
rocker arm is parallel with the body, i.e. perfectly planar −k2 + σ · k22 − k32 + k12
t= (16)
ground. The proposed system features cost-effectiveness, k3 − k1
simplicity and requires small and affordable modifications
to the conventional rocker suspension design. where σ = −1 is a sign variable identifying the assembly
mode. There are two solutions for θ2 corresponding to the
3.1 Mechanism synthesis two values of σ. These correspond to the two assembly
mode for the linkage. Note that the variable t may be
In this section, the synthesis of the variable cambering complex when k12 +k22 < k32 . If this happens, the mechanism
mechanism is discussed. The coupler Link 2 is the driver can not be assembled. After θ2 is known, equations can be
in the linkage: any relative rotation of the rocker arm with solved for θ1 . Dividing (8) by (9) and solving for θ1 gives
 
respect to robot’s body produces a correspondent vertical d − l2 · sin θ2
displacement of the slider B θ1 = arctan (17)
h + ∆yB − l2 · cos θ2
r
∆yB = b · sin βR (7)
Equations (16)–(17) give a complete and consistent solu-
where b is the distance of Link 2 attachment along the tion to the position problem as function of the geometrical
r
right rocker arm from the differential semi-axis. In order to parameters l1 , l2 and b and the input variable βR . Note
determine the relationship between ∆yB and the variables that the value of θ2 at rest is set to θ20 =90 deg, and d is
θ1 and θ2 , as expressed in Fig. 9, the position analysis of a known geometrical parameter (d=0.065 m for the rover
the mechanism needs to be analytically solved using the Dune). Finally, given θ2 , the wheel camber of Wheel 3 and
closure condition (B-O)=(A-O)+(B-A). Rewriting in its Wheel 4 that are attached to the right rocker arm can be
component equations, one gets obtained as
π
γR = − θ 2 (18)
2
The driving linkage is to be synthesized in order that
the coupler will move through an angle θ2 satisfying a
given function. This type of synthesis approach is usually
referred to as function generation problem. The functional
r
relationship between the two variables βR and θ2 can be
generally expressed as
r
f (βR , θ1 , θ2 , l1 , l2 , b) = 0 (19)

where l1 , l2 and b are design variable defining the system

geometry. Our goal is to design the system to minimize
the cost function
r r
g(βR , θ2 ) = φ(βR ) − γR (θ2 ) (20)
r
where the function φ = φ(βR ) is shown in Fig. 4 and
Fig. 9. Planar schematization of the variable camber obtained numerically as explained in Section 2.1. This
mechanism optimization problem can be solved using a least-square
 (a)

Fig. 10. Wheel camber change using the adaptive suspen-

sion system. Note that wheel camber contrasts closely
the robot’s roll
approach, yielding to the following results: l1 =0.1 m,
l2 =0.054 m, b=0.071 m. When considering a driving mech-
anism with this geometry, a variation of wheel camber
as function of vehicle’s roll is obtained as shown in Fig.
10. The discrepancy with the desired function can be
expressed as absolute average of the residuals and results
in RES=0.81 deg. If the analysis is limited to the usual
operating range of the rover, i.e. roll angle between -10–10 (b)
r
deg (βR =-7–7 deg and obstacle height ho =-150–150 mm),
the residuals decreases to RES=0.53 deg, showing a good Fig. 11. Comparison of the conventional rocker suspension
agreement and attesting to the feasibility of this approach. (a), with the adaptive alternative (b)
Finally, Fig. 11(b) shows the configuration of the adaptive
rover for the step climbing condition previously considered Chakraborty, N. and Ghosal, A. (2004). Kinematics of
in Fig. 5. Using the wheel cambering mechanism, each wheeled mobile robots on uneven terrain. Mech. and
wheel keeps unaltered its footprints even in presence of Machine Theory, 39, 1273–1287.
substantial body roll. This can not be achieved by conven- Foglia, M. and Reina, G. (2008). Locomotion performance
tional rovers, as shown in Fig. 11(a). evaluation of an all-terrain rover. Int. J. of Mechanics
and Control, 9(2), 13–25.
4. CONCLUSION Genta, G. (2003). Motor Vehicle Dynamics. World
Scientific Publishing, Singapore.
An adaptive suspension system with wheel cambering Lacaze, A., Murphy, K., and DelGiorno, M. (2002). Au-
compensation was presented that can be integrated with tonomous mobility for the demo III experimental un-
all-terrain rovers employing rocker-type suspensions. The manned vehicles. In Proc. of the AUVSI. Orlando, Fl.
geometrical optimization of the variable cambering mech- Lamon, P. and Siegwart, R. (2007). 3D position tracking
anism was discussed, aimed to keep wheels always perpen- in challenging terrain. The Int. J. of Robotics Research,
dicular to the ground as the robot tilts adapting to terrain 26(2), 167–186.
unevenness. The proposed system would allow a robot to Lindemann, R. and Voorhees, C. (2005). Mars exploration
improve its mobility and autonomy by maximizing tractive rover mobility assembly design, test and performance. In
and climbing performance and reducing energy consump- Proc. of the IEEE Conf. Syst. Man, Cybern., 450–455.
tion. New Orleans, LA.
Mackle, G. and Schirle, T. (2002). Active tire tilt control
das neue fahrwerkkonzept des F400 carving. In 11th
REFERENCES
Aachen colloquium on Vehicle and Engine Technology.
Bickler, D. (1998). Roving over mars. Mechanical Engi- SAE J670e (1976). Vehicle and Dynamics and Terminol-
neering Megazine. ogy. Society of Automotive Engineers, Warrendale, PA.
Biesiadecki, J., Leger, P., and Maimone, M. (2007). Trade- Serra, L. and Tetaz, C. (2004). Vehicle suspension with
offs between directed and autonomous driving on the camber control. U.S. Patent 6,668,620.
mars exploration rovers. Int. J. of Robotics Research, Sreenivasan, S. and Nanua, P. (1999). Kinematic geometry
26, 91–104. of wheeled vehicle systems. Trans. of the ASME J. of
Bobbitt, J., Schult, D., and Butler, J. (2007). Adjustable Mech. Des., 121, 50–56.
control arm for vehicle suspension mechanisms. U.S. Waldron, K. (1994). Terrain adaptive vehicles. ASME
Patent 7,278,648. Journal of Mech Design, 117, 107–112.
Buehler, M., Iagnemma, K., and Singh, S. (eds.) (2005). Wong, J. (2001). Theory of Ground Vehicles. Wiley-
The 2005 DARPA Grand Challenge. Springer. Interscience, New York.