Robotics and Autonomous Systems 51 (2005) 29–39

Environment–robot interaction—the basis for

mobility in planetary micro-rovers
Alex Ellery ∗
Surrey Space Centre, University of Surrey, Guildford, Surrey GU27XH, UK

Received 27 July 2004; accepted 12 August 2004

Available online 18 January 2005

Abstract

Planetary exploration through the deployment of robotic rovers on planetary surfaces such as mars imposes unique constraints
on mobile robotics. In particular, I examine the issue of mobility across a hostile planetary surface as an oft-neglected aspect
of robotic autonomy. I compare the traction performance of a wheeled concept (the rocker–bogie springless system adopted on
Sojourner), a tracked vehicle concept and a novel concept called the elastic loop mobility system (ELMS). I highlight some
limitations of the Bekker theory analysis used here in the determination of mobility characteristics of any vehicle locomotion
system.
© 2004 Elsevier B.V. All rights reserved.

Keywords: Mobile robotics; Planetary exploration; Mars rover; Bekker theory; Mobility analysis; Robot–environment interaction

1. Introduction does not have that luxury—planetary environments are

rugged, hostile and a priori unknown. Indeed, for such
Most terrestrial mobile robotics platforms are oper- applications, the environment and its characteristics
ated in relatively benign environments such as office are fundamental in determining robotic behaviour. Al-
corridors and the like despite recent emphasis on “em- though planetary surfaces are static (though the motion
bodied” or “situated” robotics paradigms which em- of the rover introduces a dynamic component) unlike
phasise the necessity for dealing with realistic (and so those encountered by terrestrial robots, they are un-
uncompromising) environments [2,10]. Evolutionary structured and rocky, requiring robust mobility over a
robotics has focussed on robotic control systems with hostile and challenging surface (Fig. 1).
some considerations of robot body morphology but The agent–environment interaction involves the dis-
still within artificial environments. Planetary robotics sipation of energy which ensures physical energy trans-
fer between the two systems. It is this approach of mod-
∗ Tel.: +44 1483 683882; fax: +44 1483 689503. elling the energy transfer of the environment and the
E-mail address: [email protected]. robot that I take here. I explore some issues relating to

0921-8890/$ – see front matter © 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.robot.2004.08.007
30 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39

ing motors substantially reduces the mass of the vehi-

cle. Hence, the analysis presented here consider pre-
cisely these parameters to determine the performance
of robotic planetary rovers.
In this paper, some of the results are presented from
a study of different mobility systems for the European
Space Agency under the Aurora programme [14].

2. Planetary rover mobility systems

Many of the proposed locomotion systems for plan-

etary rovers have paid little attention for the need to ac-
commodate significant payload capacity for scientific
Fig. 1. The Martian terrain (from NASA). instruments and to act as a stable platform for the deliv-
ery of those instruments to their selected targets—the
environmental complexity which are relevant to plan- scientific payload of a typical rover is ∼5–15% of the
etary exploration, namely the interaction of the robot’s total rover mass but it is desirable to maximise this mass
mobility system and the planetary terrain. The hostility fraction as the primary raison d’etre of the rover. Mo-
of planetary terrains make this aspect of robotic auton- bility performance is a defining characteristic for the
omy somewhat peculiar to planetary robotics, though of choice of scientific targets available to the scientific
relevance to any terrestrial robot designed to function instruments. There are five classes of locomotion sys-
in natural environments (e.g. military environments). tem in mobile robots which are applicable to planetary
The NATO Reference Mobility Model (NRMM) em- exploration rovers:
phasises performance characteristics based on [20]:
(i) wheels, e.g. automobile locomotion;
(i) maximum speed and turning radius; (ii) tracks, e.g. armoured vehicle locomotion;
(ii) traction for overcoming resistive forces to motion; (iii) legs, e.g. animal locomotion;
(iii) vehicle maneouvrability for obstacle avoidance; (iv) body articulation, e.g. snake undulation;
(iv) ride comfort. (v) non-contact locomotion, e.g. hopping.

Performance parameter (iv) is not considered fur- The chief advantage of legged robots is that they are
ther as suspension is not generally regarded as a high required to overcome only the compaction resistances
priority for robotic rovers and the traversal speeds of at the point of contact while wheels and tracks must
planetary rovers are low ∼10–20 cm/s. Performance overcome these forces continuously. Furthermore, legs
parameters (i) are difficult to clarify in any quantita- use these resistive forces to aid movement rather than
tively comparative sense but are determined by parame- hinder movement. However, legged motion involves
ters (ii) and (iii). Maximum speed will be determined by considerably more complex control algorithms than
the motor torques, slope, incidence of obstacles (which wheeled and tracked systems. Of these types of mo-
determines mean free path), and the surface traction bility system, I have considered only the first two from
on the soil. Turning radius will depend on the geom- which I have selected three candidate mobility systems
etry of the vehicle and the nature of the turning mode for study:
and strongly influences parameter (iii)—skid steering
which is adopted in tracked vehicles and small micro- (i) the wheeled rocker–bogie suspension system (as
rover vehicles offers the highest turning maneouvrabil- exemplified by the Sojourner rover);
ity at the expense of power consumption. Most vehicles (ii) a tracked system (as exemplified by the Nanokhod
with forward and aft motion capability can turn through rover suitably scaled);
skid steering and indeed, elimination of explicit steer- (iii) the elastic loop mobility system (ELMS).
 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39 31

2.1. Rocker–bogie systems

The rocker–bogie wheeled concept has flown on

Sojourner and represents the US approach as it will
continue to be adopted for future perceived US rover
missions such as the current Mars Exploration Rovers
[7,26]. It is a springless suspension system with the
property of maintaining equal pressure on all six wheels
Fig. 3. Shrimp configuration (from [16]).
whilst negotiating obstacles. The rocker–bogie chas-
sis comprises two pairs of rocker–bogie arm assem- angle of the steering wheels so that during the turn, the
blies. Each pair consists of a main rocker arm and a wheels roll without skidding and follow a curved path
secondary arm whose pivot point lies at the front end with a common origin.
of the main arm. The rocker–bogie has front and cen- A variation on the rocker–bogie principle of front
tre wheels joined by bogies—each bogie pivots at the wheel obstacle negotiation is EPFL’s (Ecole Polytech-
front of a rocker link. The rocker has an aft wheel nique Federale de Lausanne) Shrimp which accom-
at its rear end. The rockers pivot at their midpoints plishes much the same approach but with six inde-
where they are attached to the hull body. The rock- pendently actuated wheels in a 4 bar rhombic linkage
ers are connected to the body using differential joints configuration [16], see Fig. 3. There are two wheels
which maintain the average pitch angle of the body be- arranged on a pivoting bogie on each side (similar to a
tween the two rockers. The rocker–bogie uses no elastic train suspension) plus single steering wheels forward
components such as springs—spring suspension on ob- and aft. The two centre-wheel bogie assemblies are
stacle climbing wheel reduces the weight distribution mounted onto a support which can rotate freely on a
over the remaining wheels which reduces the climbing central pivot. The rhombus configuration has a front
ability. The rocker–bogie maintains weight across all wheel mounted on a spring-loaded fork, a wheel in the
six wheels with increased traction. For the Sojourner rear attached directly to the rover chassis and two bo-
rocker–bogie system configuration, see Fig. 2(a). All gie wheels each side. The front fork provides maximum
six wheels are driven independently and attached to an vertical amplitude for the wheel by virtue of its parallel
articulated frame and the four corner wheels were inde- mechanism which provides a virtual centre of rotation.
pendently steered to provide double Ackermann steer- The front wheel fork has a spring suspension to main-
ing. Ackermann steering mechanically coordinates the tain ground contact of all wheels and to elevate the

Fig. 2. (a) US Sojourner robotic rover vehicle (from NASA); (b) European Nanokhod robotic rover vehicle (from ESA).
32 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39

front wheel over any obstacles. Shrimp can overcome

obstacles twice its wheel diameter and enables slope
negotiation of 40◦ while maintaining good stability.

2.2. Tracked vehicles

The tracked (or caterpillar) concept is a promising

one in that it offers a low mean maximum pressure
(MMP), a heuristic metric often used for vehicle per-
formance (especially military vehicles) [18]. Tracks are
generally preferred for general off-road applications as
they offer low vehicle sinkage, and so low resistance Fig. 4. Elastic loop mobility system (ELMS).
to motion. Tracked vehicles have similar properties to
wheeled vehicles and are generally favoured for their
bility capabilities [11–13], see Fig. 4. The elastic loop
higher tractive effort over more rugged terrain as they
comprises a single continuous band around two end
spread the load over a much wider area—a tracked ve-
wheels, one the drive wheel, the other the idler wheel
hicle will almost invariably offer higher drawbar pull
(though there is no reason why both wheels may not
than a four-, six-, or eight-wheeled vehicle of the same
be powered). The loop elasticity is due to a preformed
mass. Tracked vehicles generally comprise a tread, a
longitudinal and transverse curvature. The highly elas-
driver sprocket, an idler wheel (or alternatively it may
tic metal band curls over its width so that the sections
be powered), and optionally a number of supporting
between the end wheels are flat and taut. The elastic
bogie wheels. Tracks generally suffer from high power
loop due to its high rigidity ensures that it requires
consumption due to inefficiency from friction between
no bogie wheels. This substantially reduces the power
the track and the ground, and friction between the large
requirement of ELMS in comparison with traditional
number of bogie wheels which maintain the track pro-
tracks—even so, two motors are required in compari-
file against the ground. Tracks employ skid steering
son to the six motors for a six-wheeled vehicle. Its parts
by differentially driving the tracks for turning. This
count is thus expected to be much lower than alternative
relies on track slippage so vehicle odometry on such
mobility designs. Rotary seals may be used to protect
vehicles is lost, though this is not a major disadvan-
exposed parts from dust. The tight fit between the two
tage as wheel odometry is very inaccurate and must
end roller wheels and the track is limited to the upper
be augmented with external sensing modalities. This
third portion of the loop which reduces the possibility
is the traction system adopted on the European Space
of jamming by soil and rock particles. This wheel/track
Agency’s Nanokhod micro-rover [6], see Fig. 2(b). The
contact is maintained by the suspension arms under
Nanokhod had an obstacle climbing capability of 0.1 m
variable loads aided by the one-way damping action
and a speed of 5 m/h drawing power and data from
of the shock absorbers. The forward and aft wheels are
a tether. Two locomotion units housed drive wheels
raised above the ground so that rocks and dust can fall to
which drove tracks mounted around the locomotion
the ground and prevent jamming. The track thus acts as
units. The Nanokhod tracks being enclosed around the
a spring suspension system. Two lightweight shock ab-
locomotion units for dust sealing are rigid and provide
sorbers between the chassis and the drum support arms
the response of a rigid body over obstacles, offsetting
provide the basis for damping to reduce vertical oscil-
much of the mobility advantage of tracked vehicles.
lation and pitching to within ±45◦ . The loop elastic-
Ideally, tracks increase the ground contact area which
ity provides suspension while shock absorbers provide
is maximised if the tracks follow the contours of the
damping. The chief advantage of the track is retained in
terrain.
that ELMS distributes load over a large footprint. The
2.3. Elastic loop mobility system (ELMS) track stiffens longitudinally along straight ground due
to a preformed transverse curvature which uniformly
The elastic loop mobility system (ELMS) is a novel distribute the load over the large footprint. ELMS has
tracked concept which offers the potential for high mo- a further advantage in that its elasticity provides spring
 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39 33

suspension from the two loop bends over the roller nickel and titanium, respectively, whose elastic prop-
wheels. The continuous track eliminates many of the erties may be tailored to the required amount and shape
disadvantages associated with high internal friction and [21]. Shape memory alloys such as nitinol are well
mechanical complexity of traditional tracks: there are established as actuating elements—indeed, they offer
no bogie wheels (and so reduced internal friction losses higher specific power outputs than hydraulic actuators.
characteristic of most tracked vehicles) nor are there NiTi alloys offer superior properties to CuZnAl and
track links (and so no catastrophic link breakage prob- CuAlNi alloys such as high mechanical strength, high
lem) [12]. Obstacle climbing involves driving the chas- corrosion resistance and high resistivity allowing the
sis into the obstacle which pitches the nose up, deflect- use of small currents for resistive heating [23]. The
ing the rear part of the loop. The front wheel climbs martensite thermoelastic transformation starts through
the obstacle along the vertical section of the loop in- cooling at a temperature Ms and completed at a lower
dependent of the obstacle’s friction coefficient. Step temperature Mf . When this martensite crystal phase is
obstacles as high as the loop length can be negotiated. subject to deformation, it undergoes strain ∼2–10%
Trials with a prototype vehicle indicate twice the step (8% being typical) which is recoverable on heating.
height negotiation of traditional tracked systems with Shape recovery to the austenitic phase begins at a tem-
the same geometry, and three times the same for 4 × 2 perature As and is completed at a higher temperature
wheeled configurations with wheel diameters equated Af . In its most typical application, the shape memory
to the loop height [12]. It can negotiate slopes of ∼35◦ alloy relies on plastic deformation at ambient temper-
on soft soils and crevasse negotiation of 90% of its atures which become elastic on heating causing the al-
length. loy to return to its original shape—electrically driven
The ELMS is constructed from a single continu- joule heating is used to control transition of the mate-
ous track of highly elastic material. The material has rial through its transformations. However, in this case
traditionally been Ti(III) ␤ alloy (Ti with 11.5% Mo, I am interested in the super-elastic phase of nitinol.
6.5% Zr and 4.6% Sn) which required a complex fab- Its super-elastic phase lies in the hysteresis loop be-
rication process of welding flat strips, bulge-forming, tween its austenitic and martensite phases which may
age-hardening and roll-forming [12]. My modern ver- be manufactured through 0.1–0.001% alloy composi-
sion of the ELMS track is perceived to be constructed tion to lie within a wide temperature range of ∼100 K
from nitinol, a shape memory alloy of close to 50% (though more typically lies within the range ∼30 K).
The loop elasticity is due to a preformed longitudinal
and transverse curvature imposed during manufacture.
The highly elastic metal band curls over its width so
that the sections between the end wheels are flat and
taut. Nitinol (49/51) has the required structural proper-
ties for mars application—it has a density of 6.45 g/cm3
with corrosion resistance similar to stainless steel and
Ti alloy.
The ELMS track is a single-piece continuous band
around the two end wheels which eliminates all risk of
single-point failures associated with link failures typi-
cal of conventional (military) tracks. The ELMS has a
power consumption significantly lower than that of the
conventional tracked system (due to its lack of bogie
wheels and lack of pin friction) and marginally greater
than the wheeled systems. The marginally greater
power consumption of ELMS over wheeled vehicles
is compensated by the higher obstacle negotiation and
superior slope climbing capability of its competitors.
Fig. 5. Martian terrain model (Viking Lander 1 site). The only other mobility systems which can match its
34 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39

performance are the Marsokhods which have a com- will give a proper value of limiting rock diameter D0 :
plex chassis design including a torsional spine which H = 0.365D + 0.008 assuming that the rock height for
comprise a significant fraction of their total mass [19]. VL1 is 3/8 the rock diameter; H = 0.506D + 0.008 as-
suming that the rock height for VL2 is 1/2 the rock
diameter.
3. Rover mean free path For a vehicle to move a distance x, it must sweep out
a rectangle of length (x + D/2) by width (w + D) for
One of the most important aspects of mean free path rocks of diameter D in order to avoid a rock’s position
(MFP) is that it defines the artificial intelligence re- defined as a point particle—this is a similar process in
quirements of the robot—it may be defined as the ex- spirit to geometric path planners which shrink the ve-
pected straight-line path distance that a vehicle can tra- hicle to a point and expand obstacles to compensate,
verse over a given terrain before a heading change is except in this case the obstacles are shrunk to points
required due to the incidence of an insurmountable ob- and the vehicle dimensions expanded to compensate.
stacle. A high mean free path reduces the requirement MFP may be defined as the product of the areal cover-
for steering changes due to obstacles. The greater the age of the rover’s trajectory of (x + D/2), and the areal
number of heading changes due to obstacles, the greater density of rocks of diameter D [27,28]. To determine
the accommodation of deviance from the desired trajec- the mean distance between nearest neighbour rocks, the
tory by the autonomous navigation system. By reduc- first moment of probability density function is required.
ing this requirement for deviations from straight-line For rocks larger than D0 :
paths, the mobility approach thus distributes responsi-
∞ 
D
bility for autonomous navigation between the mobility x+ (r + D)ρ(D) dD = 1, (2)
system and the robot controller. Mean free path is a D0 2
metric used to determine the navigability of a given
where ρ(D) (=K e−qD ) is the probability density of rock
terrain for a given vehicle geometry, so is dependent
(centres/m2 ) for rocks between D and D + δD.
on the obstacle distribution. The Martian surface rock
In general, the MFP will be determined by the turn-
distribution impacts on ground clearance and straight-
ing width of the vehicle:
line trajectory segments—hence, it has a direct bear-

ing on the degree of onboard navigational intelligence    
d w 2 d w
required. Rocks of diameter D are assumed to be dis- r = l2 + + − − ,
tributed randomly according to a Poisson distribution 2 2 2 2
with an expectation value in proportion to the area, where l is the distance between axles of the steered
Fig. 5. The frequency distribution (or probability den- wheels, d the diameter of turn, and w the width of ve-
sity) of rocks for the Viking Lander sites 1 and 2 of hicle.
diameter D are given by [17] For Ackermann steering geometry, turn radius d/2
is determined by the vehicle configuration [25]:
F (D) = K e−qD , (1) For four-wheeled vehicle:
where K is the cumulative fractional area covered d 2l
= ,
by rocks of all sizes, q the exponentiation coef- 2 sin α
ficient, KVL1 = 0.069, KVL2 = 0.176, qVL1 = 4.08 and where l is the distance between axles of steered wheels
qVL2 = 2.73. and α the steering angle.
The rock distribution impacts the scale of the robotic For six-wheeled vehicle:
rover through the mean free path of the vehicle over
d 2l w
that terrain. Viking Lander site 2 is believed to repre- = + +l,
sent one of the most rocky environments on mars. Now, 2 sin α 2
MFP is determined by the rock vertical height rather where l is the distance between front/rear end and cen-
than rock diameter due to embedding in the soil. The tre wheels, w the width of vehicle and l the distance
average rock height may be determined by [17]—this between steering point to front/rear wheel.
 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39 35

For skid steering employed by tracked vehicles and For the ELMS vehicle, the curvature diameter of the
small wheeled vehicles,
√ d/2 = 0 and the largest diagonal track may be given by [12]:
of the vehicle r = l2 + w2 assuming that the vehicle 2
can turn in place—on-the-spot turning is assumed to d=√ , (4)
be the minimum turn radius provided by skid steering Fload /πbB
capability. The solution for MFP is given by where Fload is the vehicle weight (=137.3 N), b the track
width (=0.06 m), B (=Et 3 /12(1 − ν2 )) the loop bend-
∞ 
∞ 2
1 − (r/2) D0 Dρ(D) dD − (1/2) D0 D ρ(D) dD
ing rigidity, E the Young’s modulus (=83 GPa) for NiTi
x= ∞ ∞ . in austenite phase or 300 MPa in super-elastic phase,
r D0 ρ(d) dD + d0 Dρ(D) dD
ν the Poisson’s ratio (=0.3) and t the loop thickness
(3)
(=1 × 10−3 m).
I have used values that generate track curvature di-
The proper integrals may be evaluated thus ameter close to the wheel diameter—the loop curvature

∞
can be designed into the track by altering the Young’s
K −qD0
K e−qD dD =e , modulus of the track alloy and the track thickness. I
D q have utilised [12] computation for determining the ob-

∞  
D0 e−qD0 e−qD0 stacle negotiation capability of the ELMS chassis:
DK e−qD dD = K + ,
D q q2 h = h0 + 0.5γ,

∞
D2 K eqD dD where h0 is the vehicle wheel height from ground,
D γ (=[(1 + l/2r)/2a2 C0 − lZ]τ) the pitch angle, τ
  (=[µ(1 + µ)/(1 + µ2 )]rW) the maximum torque, l the
D02 e−qD0 2D e−qD0 2 e−qD0
=K + + . vehicle length, r the drive wheel radius, a the half-
q q2 q3 ground contact length (=l/2), Z the forward drive
force = weight (as minimum required force), µ the co-
In order to compute the MFP of a vehicle, account efficient of friction of soil (=0.8) and C0 the soil cohe-
must be taken of its obstacle negotiation capability. sion (=1750 Pa) (average of mars soil data).
I compared the MFP for the three candidate chas- My values for this computation suggest that the
sis based on a normalised wheel diameter of 0.13 m obstacle negotiation capability of ELMS is 2.9 times
and a Sojourner sized vehicle of 0.65 m length × 0.5 m the wheel/track curvature diameter—I assume a worst-
width and experimentally-determined obstacle negoti- case of 2.5 times the wheel diameter here. These es-
ation capabilities. In fact, this falls within the estimated timates of obstacle negotiation capability require in-
maximum MFP for the Martian surface which suggests dependent normalised experimental validation which
a vehicle scaling dimension of 0.3–0.5 m [27,28]. The have not yet been conducted. There are additional fac-
MFP has been determined for three vehicles on the ba- tors to be considered such as the mass distribution of
sis of: rover vehicle and of course torque capability which
will have a major impact on obstacle crossing capa-
(i) Sojournor has a nominal obstacle negotiation bility. The following mean free path values have been
capability equal to its wheel diameter and an found based on these values in comparison to Shrimp
experimentally-determined maximum obstacle for comparative purposes, Table 1.
negotiation capability of 1.5 times its wheel di- Clearly, increasing the obstacle negotiation ca-
ameter. pability increases the mean free path of the vehicle
(ii) Nanokhod had an estimated obstacle negotiation and so will have an impact on the design of the
capability of two times its wheel diameter (R. onboard navigation system requirements. For a Mars
Bertrand, 2002, private communication). Exploration Rover range of 100 m/day a number of
(iii) ELMS has an experimentally-determined obsta- obstacles will need to be negotiated autonomously
cle negotiation capability of three times its track between communication windows sessions. Further-
curvature diameter [12]. more, vehicle scaling has a strong impact on the MFP
36 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39

Table 1 τ = C0 + σ tan φ, (5)

Mean free path comparisons for candidate mobility systems
where τ is the soil shear strength, C0 the cohesive
Obstacle MFP/m (Viking MFP/m (Viking
limit (m) Lander site 1) Landing site 2)
strength of soil (related to cementation), µ the soil co-
efficient of friction (=tan φ), φ the soil internal angle of
Nanokhod 0.25 76.3 16.7
Sojournor 0.13 66.3 14.9
friction and σ the normal stress.
(nominal) Dry sand is a frictional soil with C0 → 0 while clay
Sojournor 0.20 72.0 15.9 is a cohesive soils with φ → 0, and Martian soils are
(max) generally of the dry sandy type cemented by salts. I
Shrimp 0.22 73.7 16.2 have taken values of C0 = 0.24 kPa and φ = 35◦ with
ELMS 0.325 83.4 18.04
an average density of 1.52 g/cm3 as my representative
baseline values for the Martian surface [22]—I have
as smaller vehicles can accommodate narrower paths found that traction is a sensitive function of the soil
between obstacles. However, the superiority of the parameters [14], but the relative tractive performance
ELMS mobility system is clear given its superior of each mobility system is unchanged. The maximum
obstacle negotiation capability. tractive thrust available from the soil over a vehicle
contact area A is given by
4. Robot–terrain interaction H = AC0 + Wµ = AC0 + W tan φ, (6)
The primary goals for a planetary rover are the where A is the ground contact area and W the vehicle
capacity to navigate in an unknown, hostile terrain, weight.
recognise and negotiate obstacles, deploy scientific in- This is the Bernstein–Bekker equation [3–5]. Slip
struments and acquire samples from scientific targets. may be incorporated by [29]:
These goals must be attained with minimum mass and
H = H0 (e−sl/κ ), (7)
minimum volume and within tight mobility, power,
thermal and communications constraints. In this anal- where κ is the shear deformation
√ slip modulus, l the
ysis, I have assumed that each vehicle is normalised ground contact length (=2 (d − δ)δ), d the wheel di-
to a Sojourner sized vehicle of mass 14 kg with a ameter, δ the deflection (<0.1d) for rigid wheels (such
wheel diameter of 0.13 m and a wheel/track width of as Sojourner), s (=µW/kt l2 ) the slip in the limit of low
0.06 m. Mobility performance is quantified through velocity, kt the tangential stiffness of wheel or track
vehicle drawbar pull (DP: defined as the difference ∼4000 kN/m2 nominally, µ the coefficient of surface
between soil thrust and motion resistance). Locomo- adhesion (=1–102 N/m2 ) for mars Viking Lander site.
tion requires traction to provide forward thrust on the For grousered n-wheeled vehicle, the soil thrust is
ground in excess of the resistance which is dominated given by
by sinkage resistance. An important aspect of designing d
planetary rovers is consideration of the soil mechan- H = N(T0 + W tan φ), (8)
2
ics properties—shear strength, compressive strength, √
bearing capacity, penetration resistance, etc. Almost where N is the number of wheels, b (=2 (b − δ)δ)
all of these parameters are inter-related, e.g. bearing the wheel width, T0 = τ0 (hb/cos β sin β + h2 /cos β), β
capacity is dependent on penetration resistance. The (=45 + φ/2), τ 0 the soil shear stress and h the grouser
most important soil parameters are related to shear height.
strength which is quantified by the Mohr–Coulomb re- For a double tracked vehicle with N grousers per
track:
lation. From the effective soil shear stress (which is   
determined by the cohesion and internal angle of fric- 2h
H = 2 NblC0 1 +
tion), rover trafficability and other parameters can be b
determined. Soil shear strength is characterised by soil    
h h
cohesion C0 and internal friction angle φ quantified + W tan φ cot−1 , (9)
b b
by the Mohr–Coulomb law such that the shear stress
parallel to the plane of failure: where h the grouser height and l the ground track length.
 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39 37

These Bekker theory based formulations provide Wheel sinkage z for an N-wheeled vehicle is given by
the basis for computing the tractive thrust available to  2/(2n+1)
the vehicle from the soil. The tracked vehicles gener- 1 3W
z= √ . (12)
ate an order of magnitude increase in soil thrust over N (3 − n)k d
wheeled vehicles due to their much greater ground con-
In fact, this over-estimates wheel sinkage as
tact area—Sojourner yields a soil thrust of 36.7 N com-
front wheels experience greater sinkage resistance—
pared to the tracked vehicles which yield 950.3 N soil
subsequent wheels follow the rut produced by the front
thrust for Nanokhod and 2085.4 N soil thrust for ELMS
wheels causing a drop of compaction resistance of 15%
(including slippage). The higher soil thrust from ELMS
or so in comparison to the front wheels. However, for
is primarily due to the deflection of the track increasing
design purposes, this provides a small margin by under-
its ground contact area which also increases its grouser
estimating the vehicle performance.
engagement to the ground.
Bulldozing resistance becomes a problem when
To obtain drawbar pull, resistive forces must be sub-
wheel sinkage exceeds around 0.06 times the wheel
tracted from the soil thrust. There are a number of dif-
diameter—it is typically much lower than compaction
ferent sources of resistance to forward motion which
resistance. Narrow wheels or tracks suffer less from
are dominated by soil compaction with other resistance
bulldozing as significant portions of the soil are pushed
sources such as soil bulldozing and rolling resistance
to the sides of the wheel or track. Bulldozing resistance
being negligible in comparison [14]. Sinkage depends
for N-wheeled or N-tracked vehicle is given by [3,5]
on the properties of the soil and the properties of the
vehicle [3–5] such that: b sin(α + φ)
Rb = N (2zC0 kc + γz2 kγ )
Soil compaction resistance: 2 sin α cos φ

  πγl2 (90 − φ) πC0 l2
kb + +
Rc = N zn+1 , (10) 540 180
n+1  
φ
+C0 l2 tan 45 + , (13)
where N is the number of wheels/tracks, b the 2 w
wheel/track width, k (=kc + bkφ ) the modulus of
soil deformation due to sinkage (soil consistency) where kc = (Nc − tan φ) cos2 φ, kγ = (2Nγ /tan φ +
(N/mn+2 ), kc the modulus of cohesion of soil defor- 1) cos2 φ, α = cos−1 (1 − (2z/d)) is the approach an-
mation, kφ the modulus of friction of soil deformation, gle, γ the soil density (=1.52 g/cm3 ) for average Mar-
n the soil deformation exponent, where 0 < n < 1.2, z tian soil, Nc,γ the coefficient of passive earth pres-
[=(W/A(k/b))1/n = (p/(k/b))1/n ] the sinkage, W the sure (=0.43) for homogeneous coarse, sandy soil and l
wheel load (pA), p the ground pressure load and A the (=z tan2 (45 − φ/2)) the distance of rupture.
wheel/track contact area. Bulldozing for tracks involves only the first term of
Given the absence of actual Martian data on the Eq. (13) while bulldozing for wheels involves all terms.
Martian soil deformation constants, I have adopted as For my three candidate vehicles, I obtained the fol-
my baseline actual values from lunar soil as [9]: κ lowing results for sinkage and soil compaction resis-
the shear deformation slip modulus (=0.018 m) for the tance (Table 2).
moon, kc the modulus of cohesion of soil deformation Bekker theory yields a high sinkage for Sojourner
(=0.14 N/cm2 ) for the moon, kφ the modulus of fric- reflecting the findings of Richter and Hamacher [24]
tion of soil deformation (=0.82 N/cm3 ) for the moon
Table 2
and n the soil deformation exponent (=1.0) (Gerstner Soil resistance for candidate mobility systems
soil such as sand).
Vehicle Soil sinkage (m) Soil compaction
Track sinkage for dual tracks is given by resistance (N)

  Sojourner 0.0503 23.073

1 W 1/n Nanokhod 0.0023 0.031
z= . (11)
2 kl ELMS 0.0023 0.031
38 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39

Table 3 ory as an analytical tool to investigate the mobility per-

Mobility performance of candidate mobility systems formance of three different mobility systems. I have
Vehicle Soil thrust (N) Compaction Drawbar been limited in that Bekker theory is the only theo-
including slip resistance (N) pull (N) retical tool available for analysing the interaction (and
Sojourner 36.75 23.07 6.88 so tractive performance) between a vehicle and natural
Nanokhod 950.3 0.03 943.5 terrain. Unlike traditional terrestrial mobile robotics,
ELMS 2085.4 0.03 2078.6
planetary robotics must address the issue of natural ter-
rains and their effect on mobility as a matter of priority.
Bekker theory, despite its limitations, however, does
that small, wheeled vehicles have excessively degraded
have value in relative ranking of tractive performance
performance according to Bekker theory which makes
of different mobility systems. It is clear, however, that
large wheel size assumptions in its formulation. I have
experimental studies must be undertaken to provide ab-
included the additional sources of resistance in my
solute data on vehicle tractive performance. It is my
formulation—namely bulldozing resistance and rolling
intention to develop the Bekker theory models, cali-
resistance—to provide minor compensation as recom-
brated with experimental data, to provide the basis for
mended by Bekker [3].
low-level behaviour-control schema within the poten-
Powered wheels produce traction and this trac-
tial field medium for an integrated autonomous naviga-
tion must exceed the resistance forces for forward
tion architecture utilising models of the form [8,1,30]:
movement—this is drawbar pull (DP), which is the dif-
ference between soil thrust H and motion resistance R: z n
τ = C0 + σ tan φ, where σ = (kc + bkφ ) . (15)
DP = H − R, (14) b
Such schema may provide the basis for traction-based
where R = Rc + Rb + Rr , Rc is the compaction resistance,
reactivity to accommodate changes in soil characteris-
Rb the bulldozing resistance and Rr the rolling resis-
tics and a degree of traction control through variable
tance.
wheel torquing. The primary role for such a capability
The total resistance to motion of tracks is always less
as part of an onboard autonomous navigation system
than that for comparable wheels. However, the most
is to provide for robust autonomous obstacle negotia-
important factor is in generating soil thrust: tracked
tion capability. This kind of robust locomotion as part
vehicles by virtue of their ground contact area provide
of an autonomous navigation system is a pre-requisite
much greater soil thrust—this illustrates the emphasis
for effective planetary rover missions given the limited
placed on ground pressure metrics such as maximum
scope for communication with such a remote robot.
mean pressure as an approximate relative measure of
mobility performance (Table 3).
Clearly, tracked systems are superior in performance
to wheeled vehicles by virtue of distribution of weight Acknowledgements
over a large ground contact area. ELMS is superior to
traditional tracked vehicles by virtue of its suspension This work was supported by European Space
and deflection properties which enhance its soil thrust. Agency (ESA) Aurora study contract Elastic loop
mobility/traction system study for Mars micro-rovers
16221/02/NL/MV (March 2003).
5. Conclusions

The elastic loop mobility system is one of the candi- References

date chassis designs for the Vanguard mission proposal
micro-rover [15]. I computed the mean free path met- [1] R. Arkin, Motor schema based mobile robot navigation, Int. J.
Robot. Res. 8 (4) (1987) 92–112.
ric for each mobility system outlined based on their [2] R. Beer, A dynamical systems perspective on
obstacle negotiation capabilities, but this will require agent–environment interaction, Artif. Intell. 72 (1995)
experimental verification. I have also used Bekker the- 173–215.
 A. Ellery / Robotics and Autonomous Systems 51 (2005) 29–39 39

[3] M. Bekker, Introduction to Terrain Vehicle Systems. Part 1. The [18] J. Hetherington, Applicability of the MMP concept in spec-
Terrain and Part 2. The Vehicle, University of Michigan Press, ifying off-road mobility for wheeled and tracked vehicles, J.
Ann Arbor, USA, 1969. Terramech 38 (2001) 63–70.
[4] M. Bekker, Theory of Land Locomotion: The Mechanics of [19] A. Kemurdijan, et al., Small Marsokhod configuration, in: Pro-
Vehicle Mobility, University of Michigan Press, Ann Arbor, ceedings of the IEEE International Conference on Robotics and
USA, 1959. Automation, Nice, France, May, 1992, pp. 165–168.
[5] M. Bekker, Off The Road Locomotion, University of Michigan [20] A. Lessem, et al., Stochastic vehicle mobility forecast using the
Press, Ann Arbor, USA, 1960. NATO Reference Mobility Model, J. Terramech. 33 (6) (1996)
[6] R. Bertrand, et al., European tracked micro-rover for planetary 273–280.
surface exploration, in: Proceedings of the 5th ESA Workshop [21] I. Mihalcz, Fundamental characteristics and design method for
on Advanced Space Technologies for Robotics and Automation NiTi shape memory alloy, Periodica Polytechnic Ser. Mech.
(ASTRA’98), ESTEC, Noordwijk, Holland, 1998, Paper No. Eng. 45 (1) (2001) 75–86.
3.4-1. [22] H. Moore, et al., Surface materials of the Viking landing sites,
[7] D. Bickler, New family of JPL planetary surface vehicles, J. Geophys. Res. 82 (28) (1997) 4497–4523.
in: Proceedings of the International Symposium on Mis- [23] D. Reynaerts, H. Van Brussel, Design aspects of shape memory
sions, Technologies and Design of Planetary Mobile Vehicles, actuators, Mechatronics 8 (1998) 635–656.
CNES/Cipaduhs-Editions, Toulouse, France, September 1992, [24] L. Richter, H. Hamacher, Investigating the locomotion per-
1993, pp. 301–306. formance of planetary microrovers with small wheel diame-
[8] R. Brooks, A robust layered control system for a mobile robot, ters and small wheel loads, in: Proceedings of the 13th In-
IEEE J. Robot. and Autom. 2 (1986) 14–23. ternational Conference ISTVS, Munich, Germany, September
[9] W. Carrier, Physical properties of the lunar surface, in: G. 14–17, 1999, pp. 719–726.
Heiken, et al. (Eds.), Lunar Sourcebook, Cambridge Univer- [25] B. Wallace, N. Rao, Engineering elements for transportation on
sity Press, Cambridge, UK, 1991, pp. 522–530. the lunar surface, Appl. Mech. Rev. 46 (6) (1993) 301–312.
[10] A. Clark, An embodied cognitive science? Trends Cogn. Sci. 3 [26] C. Weisbin, JPL space robotics: present accomplishments and
(9) (1999) 345–351. future thrusts, in: Proceedings of the ANS 6th Topical Meeting
[11] N. Costes, A mobility concept for Martian exploration, in: Pro- on Robotics and Remote Systems, Monterey, California, USA,
ceedings of the ASME Space Conference, Albuquerque, USA, 1995.
1998. [27] B. Wilcox, et al., A nanorover for mars, Space Technol. 17 (3–4)
[12] N. Costes, W. Trautwein, Elastic loop mobility system—a new (1998) 163–172.
concept for planetary exploration, J. Terramech. 10 (1) (1973) [28] B. Wilcox, A. Nasof, R. Welch, Implications of statistical rock
89–104. distributions on rover scaling, in: International Conference on
[13] N. Costes, et al., Terrain–vehicle dynamic interaction stud- Mobile Planetary Robots and Rovers Roundup, Santa Monica,
ies of a mobility concept (ELMS) for planetary surface ex- CA, USA, January 23–February 1, 1997.
ploration, in: Proceedings of the AIAA/ASME/SAE 14th [29] G. Wong, Theory of Ground Vehicles, 2nd ed., Wiley, New
Conference on Structures, Structural Dynamics and Material, York, USA, 2001.
Williamsburg, Virginia, March 20–22, USA, 1973, pp. 73– [30] K. Yoshida, et al., Slip-based traction control of a planetary
407. rover, Preprint, 2002.
[14] A. Ellery, Elastic loop mobility system study for mars
micro-rovers, ESA-ESTEC Final Contract Report No. Alex Ellery is a lecturer in space engineer-
16221/02/NL/MV, Noordwijk, Holland, 2003. ing and robotics at the Surrey Space Centre,
[15] A. Ellery, et al., Design options for a new European University of Surrey, UK. He has a B.Sc.
astrobiology-focussed mars mission—Vanguard, in: Proceed- (Hons) degree in physics from the Univer-
ings of the World Space Congress 2002, Houston, USA, Octo- sity of Ulster, his M.Sc. degree in astron-
ber 2002, Paper No. IAC-02-Q.3.2.04. omy from the University of Sussex, and a
[16] T. Estier, et al., An innovative space rover with extended climb- Ph.D. degree in astronautics and space en-
ing abilities, in: Proceedings of the 4th International Con- gineering from Cranfield Institute of Tech-
ference on Robotics for Challenging Situations and Environ- nology. His main area of specialisation is
ments, Albuquerque, NM, USA, 27 February–2 March, 2000, in space robotics in which he has written a
pp. 333–339. textbook An Introduction to Space Robotics
[17] M. Golombek, D. Rapp, Size-frequency distributions of rocks published by Praxis-Springer. His major areas of interest are: robotic
on mars and earth analogue sites: implications for future on-orbit servicing, planetary rover design, robotic deployment of
landed missions, J. Geophys. Res. 102 (E2) (1997) 4117– scientific instruments, adaptive behaviours, force control, robotic in-
4129. telligence, and biomimetic robotics.