Home

Search

Collections

Journals

About

Contact us

My IOPscience

A bioinspired multi-modal flying and walking robot

This content has been downloaded from IOPscience. Please scroll down to see the full text.
2015 Bioinspir. Biomim. 10 016005
(http://iopscience.iop.org/1748-3190/10/1/016005)
View the table of contents for this issue, or go to the journal homepage for more

Download details:
IP Address: 194.27.228.12
This content was downloaded on 14/05/2015 at 20:42

Please note that terms and conditions apply.

Bioinspir. Biomim. 10 (2015) 016005

doi:10.1088/1748-3190/10/1/016005

PAPER

RECEIVED

24 June 2014

A bioinspired multi-modal ying and walking robot

ACCEPTED FOR PUBLICATION

Ludovic Daler1,3, Stefano Mintchev1,3, Cesare Stefanini2 and Dario Floreano1

12 November 2014

PUBLISHED

19 January 2015

2
3

Laboratory of Intelligent Systems (http://lis.ep.ch) at Ecole Polytechnique Fdrale de Lausanne (EPFL), CH-1015 Lausanne,
Switzerland
BioRobotics Institute, Scuola Superiore SantAnna, Polo SantAnna Valdera, Viale Rinaldo Piaggio 34, I-56025 Pontedera (Pisa), Italy
L Daler and S Mintchev contributed equally to this work.

E-mail: ludovic.daler@ep.ch
Keywords: adaptive morphology, multi-modal locomotion, ying robot
Supplementary material for this article is available online

Abstract
With the aim to extend the versatility and adaptability of robots in complex environments, a novel
multi-modal ying and walking robot is presented. The robot consists of a ying wing with adaptive
morphology that can perform both long distance ight and walking in cluttered environments for
local exploration. The robots design is inspired by the common vampire bat Desmodus rotundus,
which can perform aerial and terrestrial locomotion with limited trade-offs. Wings adaptive morphology allows the robot to modify the shape of its body in order to increase its efciency during
terrestrial locomotion. Furthermore, aerial and terrestrial capabilities are powered by a single locomotor apparatus, therefore it reduces the total complexity and weight of this multi-modal robot.

1. Introduction
Robots capable of hybrid air and ground locomotion
could be used for different search-and-rescue missions
[1, 2], exploration of hostile environments and
environmental monitoring [3]. In particular, the
combination of forward ight and ground locomotion
brings dual advantages of travelling quickly over long
distances and thoroughly exploring a specic region of
interest on the ground in order to, for example, search
for victims trapped in partially collapsed buildings [4].
Multi-modal locomotion is a feature that increases
the environmental adaptability, the locomotion versatility, and the operational exibility of robots [5].
Although multi-modal locomotion has a potentially
high impact in robotics and has recently attracted
much attention [3, 6], robots that successfully demonstrate competences in diverse environments are still at
an early stage. Current prototypes show that the
implementation of any additional locomotion mode
can potentially lead to performance losses (i.e. manoeuvrability, speed, energetic efciency) [710].
Indeed, similarly to animals [6, 1113], multi-modal
robots are subject to various trade-offs due to conicting requirements imposed by locomotion on different
substrates. Therefore, the main challenge is to identify

2015 IOP Publishing Ltd

design strategies that maximize performances over a

broad range of substrates.
Current implementations of ying and walking
robots are mainly based on an additive strategy (see
gure 1(a)), where secondary locomotion modes are
obtained by using additional actuators and mechanisms [7, 14, 15] or dedicated appendices, such as wheels
[16], cylindrical cage [17], spherical cage [18] or legs
[9] that are not directly used during ight. These additions have an impact on weight and drag during ight,
decreasing aerial efciency and manoeuvrability. Furthermore, most of these robots have a morphology
highly optimized for a primary locomotion mode, and
are therefore less effective in the secondary one.
Many animals can perform multi-modal locomotion by using an integrated strategy: this means that a
single locomotor apparatus, composed of actuators
(i.e. set of muscles for animals or a motor for robots)
and appendices (i.e. limbs for animals or mechanical
structures for robots), is used for multiple modes of
locomotion. For example, seabirds of the Alcidae
family exploit wing propulsion for both ight and
swimming [19] and some bat species use their wings
both for ying and walking [12]. In robotics, an integrated strategy is potentially advantageous because it
limits the number of mechanisms, actuators and sensors, hence the overall weight and complexity.

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

Figure 1. Examples of multi-modal land and air robots.

However, an integrated strategy is more challenging

than an additive strategy because a single locomotor
system must accommodate potentially conicting
dynamics. The integrated approach has been adopted
in a preliminary version of the ying and walking
robot [4] described here and in a jumping and gliding
robot [20].
Furthermore, animals such as ying snakes, seabirds and some salamanders adapt their morphology
to the desired locomotion mode in order to enhance
performance. Flying snakes, for example, atten their
body in order to increase lift and to travel longer distances during gliding [21, 22]. Before diving into the
sea, alcids partially fold their wings into a shape better
suited to swimming. Johansson and Aldrin [23] suggested that the partly folded wings of the alcids may act
as efcient aft-swept wingtips, reducing the induced
drag and increasing the lift-to-drag ratio.
The salamander is another example of animal that
uses adaptive morphology. Salamanders retract their
legs along their body when they transition from walking to swimming [24]. Some other species of salamanders are also capable of rolling. Their body,
originally shaped for walking, can take the shape of a
large wheel to rapidly roll downhill [25].
Among all the animals with mixed aerial and terrestrial locomotion capabilities, the common vampire
bat Desmodus rotundus is a relevant case study. In general, bats muscles and morphology are highly adapted
for apping ight [26] and, compared to other mammals, bats move awkwardly on the ground. D. rotundus is a notable exception because, most probably due
to its blood-based diet [12], it evolved remarkable terrestrial capabilities such as running [27] and jumping
[28] with its wings. According to the literature, these
terrestrial competences do not appear to negatively
2

affect its ight ability, although further biological data

would be necessary to validate this hypothesis [29]. D.
rotundus has evolved an integrated strategy to multimodal locomotion: a single locomotor apparatus, the
pectoral muscles and the wings, are used to locomote
in the air and on the ground. In addition, this animal
adapts the morphology of the wings during the transition from ight to terrestrial locomotion. This
remarkable combination of an integrated strategy with
an adaptive morphology explains the small trade-offs
in the multi modal capabilities of this animal and is
our source of inspiration for the design principles of
robots capable of ying and walking.
In a preliminary exploration of ying and walking
robots, we proposed a robot [4] that can use the tips of
its wings as whegs [30] (see gure 1(b)). The advantage of this approach lies in the weight mitigation
achieved by using the same structure for two locomotion modes. However, this explorative prototype uses
additional actuators dedicated only to walking and has
a xed morphology.
The present paper describes the implementation
of a novel ying wing with additional walking capabilities, which fully exploits the aforementioned biological design principles. According to an integrated
strategy, the robot adopts the same actuators and
appendices, called wingerons, for both ight control
and walking on the ground. In addition, this robot has
foldable wings, which provide morphological adaptation for switching from a wing shaped for ight to a
more compact morphology adapted to ground locomotion: deployed wings maximize lift during ight,
while folded wings enhance the efciency of the robot
on the ground by increasing the grip of the wingerons.
With the proposed design, terrestrial competences are
successfully endowed on a ying wing while losses of

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

Figure 2. Novel multi-modal ying and walking robot. The robot is equipped with wingerons that seamlessly integrate ight control
and terrestrial walking. Foldable wings adapt the morphology of the robot either for aerial or ground locomotion.

aerial performance are minimized in terms of ight

manoeuvrability and cost of transport (COT) during
ight.

2. Platform design
We aim at a robot that could be used in search-andrescue missions where it will have to cover long
distances in order to reach remote areas, therefore its
primary mode of locomotion will be forward ight.
Then, the robot should be capable to land and explore
the environment in order to locate victims, therefore
its secondary mode of locomotion will be terrestrial
locomotion. Among all the possible congurations of
ying vehicles, we have chosen a ying wing because
the absence of fuselage and tail makes it simple to
construct and robust in landings. The next step is to
investigate how the basic components of a ying wing
can be used when moving on the ground. According to
the integrated design strategy, the same set of actuators
should be used for both ight control and ground
locomotion. In the proposed design (see gure 2), the
two extremities of the wings, called wingerons, are
used to control the pitch and the roll axes of the robot
during ight and are also used as whegs to power the
ground locomotion in unstructured environments.
These wingerons are designed in such a way that the
two modes of locomotion have compatible dynamics.
Moreover, the robot has foldable wings in order to
improve its performance on the ground.
2.1. Dual use wingerons
The effective use of the wingerons for multiple
locomotion modes in a ying wing (see gure 3(A))
has been achieved using a two-step design process:
The rst step was to identify the best shape of the
wingerons in order to accommodate the different
constraints imposed by ight control and ground
locomotion.
The second step involved the selection of suitable
actuators and the sizing of wingerons for their
3

Figure 3. (A) The wingeron is the portion of the wing used to

control the ight. (B) Zoom on one of the wingerons, which
shows the different parameters which have to be dimensioned
and the constraints; (1) the centre of aerodynamic pressure
(black dot) should be on the rotation axis, (2) the axis of
rotation should be in the centre of the wingeron (w1 = w2 )
and (3) the trailing edge should be horizontal. (C) The
solution that fulls all these requirements.

double use with the smallest possible ight efciency loss.

Regarding the rst step, three important aspects
constrain the shape of the wingerons (see gure 3(B)):
(1)The axis of rotation of the wingeron should be close
to its aerodynamic centre. The aerodynamic centre
of any airfoil is the point where the pitching
moment coefcient does not change with the angle
of attack; this is also the point where the lift is
applied [31]. Thus, if the axis of rotation of the
wingeron goes through the aerodynamic centre,
the torque that must be applied by the motor to
turn the wingeron is minimized and it also makes
the calculations easier since this torque will not
change with the angle of attack. The aerodynamic
centre is located at approximately 25% of the Mean
Aerodynamic Chord (MAC) [31].

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

(2)For the ground locomotion the axis of rotation of

the wingeron has to be in its centre in order to
minimize the peak torque required by the motor.
(3)The trailing edge of the wingeron should be parallel
with its axis of rotation, in order to maximize the
grip on the ground.
In order to satisfy these three requirements, we are
going to demonstrate below that the wingerons should
have a triangular shape, as in gure 3(C). The MAC
length is given by the following equation:
2(A B) A + B
2

MAC = A
,
3(A + B)

(1)

where A is the root chord of the wingeron and B is its

tip chord (see gure 3(B)). The swept distance at
MAC, c, is given by:
c=S

A + 2B
,
3(A + B)

(2)

where S is the swept distance at the tip of the wingeron

(see gure 3(B)). The following equations must be
satised in order to, respectively, minimize the motor
torque, have an horizontal trailing edge, and have the
centre of rotation at 25% of the MAC distance:
w1 = w2,

(3)

A = S + B,

(4)

w1 = c +

MAC
.
4

(5)

Thus, by replacing c and MAC by their expression

and with w1 + w2 = S + B we nd:
S 2 2AS + A2 = 0.

(6)

This equation has only one solution: S = A and

therefore B = 0. It follows that the wingerons should
have a triangular shape as shown in gure 3(C). This
triangular shape has a large swept angle which creates a
vortex at the tip of the wingeron. This vortex prevents
stalling of the wingerons at large angles of attack and
thus increases the performance during aerobatics
manoeuvres that require high deection of the
wingerons.
Regarding the second step, namely the selection of
actuators and the scaling of the wingerons, it is important to consider that ight manoeuvrability requires
rapid wingeron movements and low torque, while
ground locomotion demands high torque and lower
rotational speed. These differences in torque and rotational speed requirements make the selection of a single actuator difcult.
To appreciate this issue, lets consider the operating range of a conventional dc motor (see gure 4). It
is constrained by two main factors: the torque that can
be continuously delivered by the motor is limited by
heat dissipation, and the maximum speed is primarily
limited by the wear effect in the commutations
4

Figure 4. Operating range of a dc motor suited for terrestrial

locomotion (rotational speed n versus torque M). The
dynamics of ight control and terrestrial locomotion are
shown at point A and B, respectively. The diagonal line
illustrates the dynamics of a dc motor operating at
V = Vnominal , where V is the voltage applied to the motor and
Vnominal is the maximum voltage recommended by the
manufacturer. The continuous torque is the maximum
torque that can be applied continuously by the motor without
overheating and nmax is the maximum speed of the motor.
The optimization of the wingerons size moves the dynamics
of ight control within the continuous workspace of the
actuator dimensioned for terrestrial locomotion.

systems and in the bearings. Moreover, the maximum

continuous torque decreases with the speed, thus there
is a trade-off between torque and speed that can be
delivered by the motor. Usually, if two locomotion
modes have very different dynamics (i.e. one needs
very high speed and the other very high torque), an
actuator dimensioned for one locomotion mode will
not be suited for the other and vice-versa. For example, an actuator dimensioned for terrestrial locomotion (slow with high torque, point B), which operating
range is shown in gure 4, is not able to continuously
provide the high speed required by ight control
because its working point A is outside of the continuous workspace of this actuator. Alternatively, an
actuator dimensioned for ight control, which would
have a different operating range compared to the one
shown in gure 4 (i.e. higher nmax and lower continuous torque), would overheat during ground locomotion because of the too high torque.
A similar problem is encountered in the skeletal
muscles required for animal locomotion. Skeletal
muscles generate their maximum power and efciency
in a small range of bre strain, contraction speeds and
load [32]. When birds need less power to y due to
favourable environmental conditions, instead of reducing the contraction speed of the muscles, they alternate between gliding and apping phases in order to
maintain an optimal contraction speed of the muscle
during the apping phase [11, 33]. Using this strategy,
avian muscles work in their optimal range, maximizing efciency and output power. Similarly, when transitioning between substrates with different physical
properties [34], animals, like robots have to address
the limitations of their biological actuators. In

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

robotics, three solutions exist to address the limited

operating range of dc motors:
(1)The use of a single oversized dc motor that is
simultaneously compatible with both locomotion
modes. A single dc motor that matches both
dynamics has to be simultaneously fast and strong
resulting in a heavier solution than the use of two
different motors each independently optimized for
a single working point. Hence, an integrated
approach with a single locomotor apparatus can be
heavier than using two locomotion systems as
suggested by the additive strategy. A numerical
example below supports this conclusion.
(2)The use of a transmission with controllable gear
ratio which would allow a single actuator to match
the dynamics imposed by ight control and
terrestrial locomotion by changing this ratio. This
approach involves additional components (i.e.
clutch, gear shift actuator and multiple gear trains)
that increase both the complexity and the weight of
the robot [14]. Therefore, the advantage with
respect to an additive strategy is questionable.
(3)The rotational speed of the wingerons during ight
control could be reduced until it is compatible with
the operating range of the actuator suited for
ground locomotion (as shown by the dotted arrow
in gure 4). This can be achieved without loss of
ight manoeuvrability if the size of the wingerons is
increased proportionally to the reduction of speed.
In the third solution, the two locomotion modes
become dynamically compatible within the limited
operating range of a single actuator; hence they can be
seamlessly integrated in a single locomotor system
optimized for one of the mode of locomotion. In this
condition, the additional terrestrial competences have
a minimal impact on the aerial performances of the
robot. Furthermore, because the wingerons are
already part of the ying wing, additional weight and
drag are minimized, hence reducing the impact of
walking on the aerial COT.
2.1.1. Analysis of solution 1
When two locomotion modes have different
dynamics, the use of a single apparatus is not
convenient. To this aim, the dynamic data reported in
table 1(a) are considered, and two actuation strategies
have been compared:
A single actuator dimensioned to match the
dynamics of both working points (integrated design
strategy).
Two different actuators for ight control and for
ground walking (additive design strategy).

Weight is a good metrics to compare the different strategies since it is a key parameter in the
design of ying robots. Weight comparison is based
on the fact that biological and articial effectors
(the system that converts an input energy into an
output mechanical work) have constant power densities [35, 36]: their weight increases with rated
power capabilities. Considering a complete actuator
(i.e. electromagnetic effector + reduction stage), it is
still possible to assume that its weight increases with
its rated power. According to this assumption, the
weights of two strategies can be compared by evaluating the maximum power associated with the
actuators that are involved.
Dc motors have been selected for actuating the
wingerons since they can be easily implemented and
because their control techniques are well established.
Nevertheless, the overall methodology presented here
can be generalized to other actuators (e.g. SMAs,
EAPs). The dynamic behaviour of a dc motor is described by the following equation, which expresses the
rotational speed, n (in radians per second), as a function of the motor output torque, M:
n (M ) = n 0 kM ,

(7)

where n0 is the no-load speed and k is the speed/

torque gradient. The output power of the motor, P, is
given by:
P (M ) = n (M ) M = (n 0 kM ) M .

(8)

The maximum power, Pmax, that can be delivered

by the actuator is then evaluated as follows:
dP ( M )
= n 0 2kM ,
dM

n 0 1 n 02
=
Pmax = P M =
.

2k 4 k

(9)

(10)

The two strategies of actuation are illustrated in

the diagram in gure 5, which shows rotational speed,
n, versus torque, M. The points A and B represent the
working points of the wingerons during ight control
and walking, respectively. According to equation (7),
each dc motor is associated with a line connecting the
no-load speed to the stall torque. An actuator is suited
for a specic working point if it lies on the actuator
line. The additive strategy requires a single actuator for
each locomotion mode: actuator a for ight control
and actuator b for terrestrial operations. In the integrated design approach a single actuator c can be used
for both.
Considering at rst the strategy of multiple actuators, the motor parameters (n0 and k) can be optimized in order to minimize the peak power of each
actuator, and therefore their weight. Indeed, considering a generic working point (nx, Mx), the actuator
dynamics are described by:
n x = n 0, x k x Mx

(11)

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

Pmax, c

Figure 5. Working point A of wingerons during ight,

working point B of wingerons during ground locomotion and
dynamic capabilities of motors a, b and c. Data experimentally
measured from the rst DALER prototype.

and, according to equation (10):

Pmax, x

1 n x + k x Mx
=
4
kx

).

Pmax, x has a minimum value for k x =

ing to:
Pmax, x

1 nx + nx
=
4 n x Mx

nx
Mx

(12)

correspond-

= n x Mx .

(13)

In conclusion, power expressions associated with

optimized actuators that work in the dynamic conditions A and B, Pmax, a , Pmax, b and Pmax, a + b , are:
Pmax, a = n a Ma ,

(14)

Pmax, b = nb Mb ,

(15)

Pmax, a + b = n a Ma + nb Mb ,

(16)

where ni and Mi are respectively the speed and the

torque outputs of a motor i. This example shows that,
within the additive design approach, in order to
minimize weight, an optimal actuator delivers its
maximum power in the working point of interest.
By adopting an integrated design approach, a single actuator c is used. With this strategy, the two motor
parameters (n0, c and kc) are constrained by the following equations:
n a = n 0, c kc Ma ,

nb = n 0, c kc Mb ,

(17)

n 0, c =

(18)

n a Mb nb Ma
.
Mb Ma

(19)

Finally, the maximum power associated with the

actuator c, Pmax
, can be evaluated according to
c
equation (10) as:
6

)(

2.1.2. Analysis of solution 3

In the authors opinions, the best solution for a
successful integrated design is illustrated in gure 4. It
consists of translating the working point A to point A
in order to match the properties of actuator b. In fact,
the working points A and B are dynamically well
matched to actuator b, which can be effectively selected
to power both locomotion modes. By doing this the
velocity of the wingerons decreases, thus the manoeuvrability of the ying wing may be compromised.
The question is then if it is possible to optimize the size
of the wingerons in order to make the two operational
points dynamically compatible, with minimal impact
on ight manoeuvrability. To answer this question, let
us consider the lift generated by a wingeron, L:
L=

n a nb
,
Mb Ma

Considering the values of torque and speed associated with the two working points A and B [4], and
according to equations (14)(16) and equation (20),
the overall power consumption of the two actuation
strategies are summarized in table 1(a).
It can be observed that the power required by the
single actuator strategy is 37% higher than the overall
power required by multiple actuators. Since power is
assumed as an index of weight, the conclusion is that
an integrated design approach with a single actuator is
not optimal in terms of weight. This is due to the fact
that the two operational points have very different
dynamics, and therefore a single dc motor that matches both requirements has to be both fast and strong,
thus resulting in a heavier solution than two single
motors each optimized for a single working point. For
example, considering motors from Maxon (Sachseln,
Switzerland), good candidates are the following: RE8
with a 64:1 reduction, weighs 7.8 g for ight control;
DCX 10 L with a 400:1 reduction, weighs 18.7 g (considering the weight of the available 1024:1 reduction
stage) for ground locomotion; RE-max 13 with a 100:1
reduction, weighs 31 g with plastic gears and 41 g with
metal gears (considering the available 67:1 reduction
stage) for both locomotion modes. In summary, the
weight associated with a single actuator strategy is
1957% heavier than the solution with two different
motors, which is in good agreement with the calculation presented above.

which leads to:

kc =

2
n a Mb nb Ma
1 n 0,c
1
=
=
. (20)
4 kc
4 Mb Ma n a nb

1 2
v A wn C L ,
2

(21)

where represents the air density, v is the air speed, Awn

is the area of the wingeron and CL is the lift coefcient.
The lift coefcient of an airfoil in a steady airow can be
expressed as a function of its angle of attack, , as:
C L = Kc + C L0,

(22)

where Kc is a parameter that allows to evaluate lift

variation depending on the angle of attack and CL0 is

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

Table 1. Dynamic requirements.

the lift coefcient for = 0 . Since the aim is to

preserve ight manoeuvrability, the wingerons must
generate the same lift variation (that is ultimately
responsible for pitch and roll control) in the same time
interval, and therefore:
dL 1 2
= v A wn Kc wn
dt
2
1
= v 2A wn Kc n wn = const,
2

(23)

where nwn ( wn ) is the rotational speed of the

wingeron. This equation demonstrates that the ight
manoeuvrability is not compromised when the wingeron is slowed down if its area is increased according
to the following relationship:
A wn n wn = const.

(24)

Concerning torque requirements, in rst approximation, wingerons producing the same manoeuvrability need the same actuation torque. For example,
when the area of a wingeron is doubled its lift doubles
as well. Nevertheless, the stroke (angle ) is reduced
and consequently the velocity is divided by two. These
two effects compensate each other, resulting in a constant torque requirement for each wingeron that
ensures the same level of manoeuvrability. Thus, as
shown by the dashed arrow in gure 4, the working
point A can be translated vertically until it crosses in A
by simply increasing its width L (see gure 3). Torque
and velocity of the working points A and B, Ma , na ,
Mb and nb are reported in table 1(b) as well as the associated power requirements.
2.2. Foldable wings
In the previous section 2.1, a single locomotor
apparatus suited for both ight control and walking
has been described. However, the challenge of adapting a morphology mainly optimized for ight to one
more suited for terrestrial locomotion is discussed
here. For example, ight and walking require different
positions of the centre of mass (CM) of the robot. For
ground locomotion the CM of the robot must be close
to the centre of rotation of the wingeron to avoid the
wingerons slipping on the ground [4] (see
gure 7(B)). In ight the CM must be instead far from
the centre of rotation of the wingeron in order to
create torques required to control the ight (see
7

gure 7(A)). Furthermore, for ight stability reasons,

the CM must be in front of the aerodynamic pressure
centre of the wing. In this prototype, the use of foldable
wings to adapt the morphology of the robot to either
ight or ground locomotion is proposed. With
reference to gure 7, when the wings are deployed, the
robot has a ight-adapted morphology:
The lift is augmented due to the large wingspan.
The distance between the axis of rotation of the
wingerons and the CM is increased, enhancing
ight manoeuvrability.
When the wings are folded, the morphology of the
robot becomes suitable for ground locomotion:
A short wingspan improves the robots agility in
cluttered terrestrial environments.
The axis of rotation of the wingerons is closer to the
CM, thus enhancing the grip between the wingerons and the ground.
The use of these deployable wings allows adaptation of the morphology of the robot and therefore it
satises the requirements on the relative positioning of
the CM and of the wingeron axis of rotation for the
two modes of locomotion.

2.3. Mechanical design

This subsection presents the mechanical design of the
second DALER prototype, which is based on the
theoretical considerations presented above. As illustrated in gure 6 the DALER comprises ve main
body sections: a central frame housing the propeller,
electronics and battery, two foldable sections and two
wingerons for ight control and ground locomotion.
The robot has a wingspan of 72 cm and a weight of
393 g. The frame of the DALER is designed in order to
minimize weight, while providing enough stiffness for
efcient ight. To this aim, the frame has a central
body and multiple ribs connected together by carbon
ber spars. This frame is covered with Icarex, a lightweight polyester fabric, inextensible and resistant
to wear.

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

Figure 6. 3D model of the DALER. (A) The left wing is deployed, while the right one is completely folded. (B) Zoom on the central
frame which shows the motor for wing deployment. (C) Zoom on the wingerons drive mechanism.

Flight control and ground locomotion rely on a

single locomotor system composed of two independent wingerons and their actuators. The dc motors are
housed in the external ribs (rib B) and are coupled
with the wingerons using a synchronous belt. For control purposes, the angular position of the wingerons is
measured by Hall effect sensors, which detect the
orientation of small magnets mounted on the wingerons axis of rotation.
The two foldable sections of the robot are equipped with an articulated frame controlled by a single dc
motor. The foldable frame is composed of two carbon
spars that are serially connected together and also to
ribs A and B through revolute joints. Each foldable
section is controlled by two separate cables, one to
open and the other to close the wings. The two cables
are connected to a double arm that is directly controlled by a dc motor. The arm is used instead of a
pulley because it acts as a self-locking mechanism that
prevents unwanted rotation of the motor when the
wings are fully deployed or collapsed. The cable
responsible for wing folding is connected to a spring
and when the wing is deployed, the spring is pulled
and the covering fabric is tensioned. This is of paramount importance in order to pre-load the fabric to
avoid uttering, thus maximizing ight efciency.
Furthermore, the spring can absorb energy, limiting
damage to the wings in case of a frontal collision. If a
wing collapses due to a collision, the spring is stretched. Each spring can absorb a maximum energy of
12.5 J which corresponds to a collision at 8.4 m s1.
Wing morphing can be performed in 12 s. The length
of each foldable section can be reduced from 17 to
6 cm (65%); this corresponds to a 30% reduction in
the overall wingspan.

3. Results
This section presents the analysis of the performance
of the robot on the ground as well as in the air, the
analysis of the level of mass integration of the two
8

modes of locomotion and the analysis of the versatility

and complexity of the robot.
3.1. Ground locomotion analysis
This subsection presents the measurements of the
COT of the robot on the ground. In gure 8(a) the
COT of the robot on a wooden oor as a function of
the speed can be seen for different openings of the
wings. For each measurement two complete revolutions of the wingerons were performed and the
following parameters were measured:
The time (s).
The travelled distance (m).
The current in the dc motors (A).
The voltage applied to the dc motors (V).
A PID controller running at 100 Hz controls the
rotational speed of the wingerons. At each update of
the controller (every 10 ms) the current in both
motors is measured along with their voltage. The electrical power is computed at each step and then lowpass ltered. At the end of the run, the mean power is
computed and multiplied by the time of the run in
order to nd the total energy used for the run (in joules
(J)). Finally, the COT (in joules per kilogramme and
per metre (J Kg1 m1)) is computed by dividing the
energy by the mass of the robot and by the travelled
distance (speed is computed as the ratio between the
travelled distance and the time of the run).
Concerning the experiments presented in
gure 8(a), the wingerons were set to seven different
revolution speeds and each experiment was repeated
ve times. The smaller dots represent all the measurements and the larger dots represent the mean COT at
the mean speed of the robot for the seven imposed
rotational speeds of the wingerons. The three different
markers shapes (diamonds, triangles and squares)
represent the congurations with the wings open,
half-closed and fully closed respectively. With

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

reference to gures 6 and 7 the width of the foldable

section for the three different openings was set to 17,
11 and 6 cm respectively and creating a distance
between the CM and the axis of rotation of the wingerons (d axis d CM) of 13, 11 and 9 cm.
It can be seen that the three lines show the same
trend; a high COT at low speeds, then a minimal COT
at around 75% of the maximum rotational speed of
the wingerons, and again a high COT at high rotational speeds of the wingerons. At high rotational
speeds there is a sharp increase of COT due to the fact
that the wingerons slip more on the oor. The COT of
the robot when the wings are closed (i.e. folded) is
much lower than the COT for open wings, especially at
high speeds. Furthermore, the speed of the robot is
much higher for the same rotational speed of the wingerons (35% increases). These two arguments clearly
demonstrate the need and the advantage of having an
adaptive morphology (i.e. foldable wings).
Another interesting observation is that at speeds
lower than 0.04 m s1 the conguration with halfclosed wings has a lower COT than the one with fully
closed wings. This indicates that the morphology of
the robot (i.e. wings opening) should be continuously
adapted as a function of the speed to minimize the
COT. This can be explained by the fact that, on one
hand, when the wings are folded the CM of the platform has to rise higher at each step than when the
wings are open, while on the other hand, at low speeds
there is static friction between the ground and the wingerons while at very high speeds the friction is mainly
dynamic (i.e. lower), therefore the wings start to slip
more. Thus, by adapting the morphology of the robot
the optimal trade-off between CM lifting and wingerons slippage that minimizes the COT can be found.
Figure 8(b) shows the same experiment but with a
small wheel at the tip of the robot that is used to reduce
the friction between the central frame of the robot and
the ground. It can be seen that if this friction is reduced
to almost zero the adaptive morphology is required
less than previously and also that the robot travels
much faster for the same rotational speed of the wingerons. In outdoor, eld applications the friction of
the central body cannot be so drastically reduced, thus
this situation will not happen, however it demonstrates the importance of maximizing the friction on
the wingerons and minimizing the friction on the central frame. From these experiments it can be seen that
COT values between 15 and 20 J Kg1 m1 at speeds of
more than 0.1 m s1 can be reached, similar to small
running animals [37].
Table 2(a) summarizes the performance of the
DALER prototype presented in this paper. Different
experiments have been performed in order to evaluate
the capabilities of the robot on the ground. The maximum gap that the robot can overcome repeatedly
(100% success over ve trials) is 9 cm, which corresponds to 0.25 body-length (BL); above this distance
the robot gets stuck or falls in the gap. The maximum
9

Figure 7. (A) Robot seen from above with deployed wings.

(B) Robot seen from above with folded wings. (C) Robot seen
from the side walking on a at terrain.

step that the robot can climb is 6 cm, which corresponds to 1 body-height (BH). The maximum upward
slope, on a wooden oor, that the robot can walk on is
9. The maximum forward speed measured on a at
wooden oor is 7 cm s1, which is 0.2 BL s1 and the
maximum rotational speed of the robot (on spot) is
24 s1 (15 s for one complete revolution). The autonomy of the robot is very much dependent on the type
of terrain; as demonstrated above, the COT changes
drastically with the variation of the friction between
the robot and the ground. On a at wooden surface the
maximum autonomy has been measured at close to
60 min and in rough terrains the robot can walk for
about 30 min with a full battery (3 cells LiPo, 1 Ah).
3.2. Flight analysis
The drag force during ight is the same as on a wing
capable of only ying since no additional appendices
have been added for ground locomotion. When the
fabric that covers the wings is properly stretched by the
deployable mechanism, the lift produced by the wings
is similar to the one produced by a rigid wing. The only
difference to a regular ying wing is the weight added
for the deployable wings, which represents less than
10.7% of the weight of the platform, 42 g over 393 g
(see table 2(b)). Thus, for retaining the same wing
loading (i.e. the same total weight) as on a regular
wing; the weight of the battery must be reduced by the
weight of the deployable wings (42 g). The DALER has
a 1 Ah (3 s) battery which weighs 90 g, which means
that a 1.5 Ah (approx. 130 g) battery can be used in a
robot without deployable wings, resulting in a difference in battery capacity equal to 33%.
The minimum ight speed of the robot, before
stalling, has been measured at 6 m s1 and therefore
the robot can easily be launched by hand. The maximum ight speed of the robot has been measured
above 20 m s1. The autonomy of the robot at cruise
speed (approx. 12 m s1) has been measured at
between 15 and 20 min with a 1 Ah battery. The

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

Figure 8. Cost of transport versus speed of the robot on the ground.

Table 2. Performance and weight analyses.

maximum constant pitch and roll rates, measured by

performing an inside loop manoeuvre and a full roll
manoeuvre, are 120 s1 and 180 s1 respectively.
The robot shows excellent ight performances that
can be seen in the accompanying video.
3.3. Integrated design analysis
The working area of the wingerons motors has been
measured during ight. For a standard ight, the speed
of the wingerons varies between 0 and 25 rpm and the
torque varies between 100 and 150 Nmm. This torque
is mainly caused by friction in the belt transmission
and does not change with the speed of the robot in the
air. The speed during ground locomotion varies
between 0 and 15 rpm (which is the speed that
minimizes the COT) and the torque varies between
650 and 850 Nmm depending on wing opening, speed
of the robot and friction with the ground. Figure 9
shows the two designed working points, A and B, for
ight and ground locomotion respectively and the two
measured working areas of the motor during ight
and ground locomotion are also shown. The two
10

measured areas are below the line of the actuator and

this demonstrates that the same motor can be used for
the two modes of locomotion and that these modes of
locomotion are dynamically compatible.
Table 2(b) provides the weight distribution of the
DALER prototype. All the components are sorted into
three categories; the components shared for both
modes of locomotion, the components used only for
ground locomotion and those used only for ight.
Shared components are: the frame which is composed
of all the mechanical parts which are needed to build a
xed wing without deployable wings (central body,
ribs, carbon rods, wingerons and fabric), the wingerons control which includes the actuation and the
transmission used to control the wingerons, the electronics which includes the autopilot board and the dc
motor board, and nally the battery. These shared
components weigh a total of 301 g. The components
which are added only for ground locomotion are part
of the foldable wings, which includes all the joints,
bearings and springs which are added to enable the
folding of the wings. They together weigh 42 g, which

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

Figure 9. Designed working points A of wingerons during

ight and B during ground locomotion and measured working areas of the motor during the two modes of locomotion.

brings the total for the ground locomotion to 343 g.

The parts used only for the ight are the propulsion system (brushless motor, propeller and speed controller)
and the motor holding. These components weigh 50 g,
which brings the total mass for the ight mode of locomotion to 351 g. The total mass of the robot
weigh 393 g.
According to the mass integration metrics recently
introduced in [20], which is a measure of the percentage of the total integrated robot mass required to
combine the modes without any integration between
them, the mass integration of the DALER prototype is
1.77. This results shows that 77% of the mass is used by
both modes of locomotion; the robot is 43.5% lighter
than it would be with an additive design approach. The
other multi-modal robot evaluated with this metric is
the MultiMo-Bat [20], which scores 1.69. The result
achieved by the DALER is even more remarkable considering that it combines active aerial and terrestrial
locomotion, while in the MultiMo-Bat ight is passive
(gliding), and therefore it does not require actuators.
In the DALER, such high level of integration can been
achieved since the same actuators are used for both
ight control and walking on the ground.
3.4. Multi-modal locomotion analysis
The versatility and complexity metrics dened in [5]
have been used to evaluate the DALER. The metrics
used to evaluate the versatility of a mobile robot is
dened as an extension of mobility that includes
operation in and transition among multiple domains,
and the complexity metrics is dened as the number of
actuators multiplied by the number of degrees of
freedom of the robot. The objectiveness of the
versatility metrics is questionable since arbitrary
grades are given by the person doing the evaluation of
a given robot. However, to the best of the authors
knowledge, there is no objective metrics to evaluate
the versatility of a mobile robot.
Based on the rules dened in [5], the versatility of
the DALER prototype has been evaluated and compared to the ones of the rst DALER prototype [4], of
BOLT [9] and of MMALV [15]. These robots have
been chosen since they are the only robots with wings
that can also move on the ground and thus they can be
11

compared to our prototype. Figure 10(a) shows the

results of these evaluations (refer to [5]). The robots
are graded for their capabilities in the aerial and terrestrial domains and for their capability of transition
between these domains; they obtain a grade between 0
(cannot do it) and 2 (does it well) for each mobility.
The mobilities in the terrestrial domain are further
categorized between the manoeuvrability and the
capacity to overcome different types of obstacles.
Some of these grades are based on measured values
given in table 2(a) and some are evaluated by the
authors. The versatility of the DALER is 0.183 which is
higher than the other prototypes; 0.141 for the rst
DALER, 0.128 for MMALV and 0.118 for BOLT. The
rst DALER prototype obtains a lower versatility
because it has lower ight performances compared to
the new DALER prototype. Furthermore, according to
gure 4 of [5], which shows the versatility of many
mobile robots, none of them has a versatility higher
than 0.16. This result shows that our prototype has a
very high versatility compared to state of the art
mobile robots.
The complexity of these robots has also been evaluated and the results are given in gure 10(b). The
complexity of the DALER is 16; it has four actuators
(the two dc motors for the wingerons, the brushless
motor for the propeller and the dc motor for the
adaptive morphology) and has as many degrees of
freedom. The previous DALER prototype had a complexity of 49 (seven actuators and seven DOFs), BOLT
has a complexity of only 12 (three actuators and four
DOFs; its wings and legs are power with a single actuator) and MMALV has a complexity of 36 (six actuators and six DOFs). Figure 10(c) gives the versatility
versus the complexity of this four robots. It can be
seen that the new DALER prototype has a higher versatility and a lower complexity than the rst DALER
prototype and than MMALV. BOLT has a lower complexity but also has a lower versatility compared to the
others.

4. Generalized design method

This section proposes a design method extracted from
the example presented in this paper. For the design of a
robot capable of multi-modal locomotion two different strategies can be adopted: an integrated strategy or
an additive strategy. Nature provides many examples
of animals using one of these two strategies [6].
However, when it comes to robotics an integrated
strategy will allow a higher reduction of total weight
and complexity, and will consequently lead to better
performances in the different modes of locomotion.
The main idea behind using an integrated strategy (i.e.
using a multi-purpose locomotor apparatus) is to
maximize the versatility of the robot by minimizing
the complexity of the mechanical structure and the
number of actuators required by each additional

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

Figure 10. Versatility versus complexity of multi-modal robots computed with the metrics dened in [5]. Only robots capable of
winged-ight and ground locomotion have been compared in this study.

locomotion mode. However, effective implementation of an integrated approach is challenging; while in

an additive strategy each apparatus can be separately
optimized for a specic locomotion mode, a dual use
apparatus must simultaneously full the different
dynamics imposed by each locomotion mode. The
steps to follow for the design of an effective multimodal robot based on the proposed methodology are
given below:
(1)Dene the mission requirements of the robot.
(2)Dene what is the principal mode of locomotion
and choose an appropriate basic design.
(3)Dene how the existing structure and actuators of
this basic design could be reused for the second
mode of locomotion. If necessary modify the
placement of the locomotor apparatus to t the
needs of the two modes of locomotion.
(4)Optimize the shape and dimensions of the locomotor apparatus; the two modes of locomotion must
have compatible dynamics.
(5)Use adaptive morphology if the two modes of
locomotion have different constraints on the placement of the CM.
(6)Take advantage, if possible, of the adaptive morphology to continuously adapt the shape of the
robot in order to optimize the performance of the
robot for different locomotion tasks.
12

The rst three steps were presented in greater

detail in [4] while the steps 46 are covered by this
paper. The six steps were followed for the design of the
prototype presented in this paper and the authors
believe that they can be applied to the design of other
multi-modal locomotion robots.

5. Discussion
Nature has evolved multiple strategies to implement
multi-modal locomotion. These strategies can be
successfully applied to the development of robots with
locomotion capabilities in multiple environments
with minimal compromises.
A comparison can be made between animals that
exploit an additive strategy with multiple single-use
locomotor apparatus, or an integrated strategy with a
single apparatus with competences in multiple substrates. For robots, it is shown that the latter strategy is
convenient if the two locomotion modes impose
dynamics that are compatible with the operating range
of the actuator used in the single locomotor apparatus.
In this condition, secondary locomotion modes can be
added with minor impact on the primary locomotion
mode. In addition, the overall structural mass and
complexity of the robot are minimized as well. For
example, in the proposed prototype, ground locomotion can be performed with wingerons, introducing
minimal losses in ight manoeuvrability and minimal
increase in robot weight.

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

Furthermore, many animals exploit adaptive

morphologies in order to accommodate the requirements imposed by different modes of locomotion. We
have shown that also this strategy is a good solution to
minimize trade-offs. Indeed, a robot optimized for
ight can improve its terrestrial capabilities with foldable wings.
In addition, multi-modal locomotion in robotics
could provide new perspectives for understanding
multi-modal locomotion in animals. For example, the
concept of dynamically compatible locomotion modes
could explain why the Desmodus rotundus does not
apparently show compromises caused by terrestrial
competences in a body optimized for ight. It is possible to speculate that this bat evolved a unique running
gait compatible with a locomotor apparatus evolved
for ight [29]. The concept of adaptive morphology is
supported by biological observations of animals that
actively control their body to improve locomotion
performance in multiple environments [23].
Our aim in the near future is to deepen the understanding of the biological principle of multi-modal
locomotion, also integrating ad hoc control techniques and multi-purpose perception capabilities into
the robot. Concerning the locomotion exibility of the
DALER, we would like to add hovering capabilities,
which are needed for vertical take-off and ight in
cluttered environments. Furthermore, we are currently developing a mass-and-power model of this
future version of the DALER prototype in order to be
able to optimize its geometrical parameters (e.g. wingspan, chord, etc) to different mission scenarios.

Acknowledgments
This work was supported by the Swiss National
Science Foundation through the National Centre of
Competence in Research Robotics.

References
[1] Tadokoro S 2009 Rescue Robotics: DDT Project on Robots and
Systems for Urban Search and Rescue (Berlin: Springer)
[2] Nagatani K et al 2013 Emergency response to the nuclear
accident at the Fukushima Daiichi nuclear power plants using
mobile rescue robots J. Field Robot. 30 4463
[3] Siddall R and Kova M 2014 Launching the aquamav:
bioinspired design for aerialaquatic robotic platforms
Bioinsp. Biomim. 9 031001
[4] Daler L, Lecoeur J, Hhlen P B and Floreano D 2013 A ying
robot with adaptive morphology for multi-modal locomotion
IEEE/RSJ Inter. Conf. Intelligent Robots and Systems (IROS)
13616
[5] Nie C, Pacheco-Corcho X and Spenko M 2013 Robots on the
move: versatility and complexity in mobile robot locomotion
IEEE Robot. Autom. Mag. 20 7282
[6] Lock R J, Burgess S C and Vaidyanathan R 2014 Multi-modal
locomotion: from animal to application Bioinsp. Biomim. 9
011001
[7] Kossett A, DSa R, Purvey J and Papanikolopoulos N 2010
Design of an improved land/air miniature robot IEEE Int.
Conf. on Robotics and Automation (ICRA) (Piscataway, NJ:
IEEE) pp 6327

13

[8] Kovac M, Hraiz W, Fauria O, Zufferey J-C and Floreano D

2011 The ep jumpglider: a hybrid jumping and gliding
robot with rigid or folding wings IEEE Int. Conf. on Robotics
and Biomimetics (ROBIO) (Piscataway, NJ: IEEE)
pp 15038
[9] Peterson K and Fearing R S 2011 Experimental dynamics of
wing assisted running for a bipedal ornithopter IEEE/RSJ Int.
Conf. on Intelligent Robots and Systems (IROS) (Piscataway, NJ:
IEEE) pp 50806
[10] Li F, Liu W, Fu X, Bonsignori G, Scarfogliero U,
Stefanini C and Dario P 2012 Jumping like an insect: design
and dynamic optimization of a jumping mini robot based on
bio-mimetic inspiration Mechatronics 22 16776
[11] Lovvorn J R, Croll D A and Liggins G A 1999 Mechanical versus
physiological determinants of swimming speeds in diving
Brunnichs guillemots J. Exp. Biol. 202 174152
[12] Riskin D K, Parsons S, Schutt W A, Carter G G and
Hermanson J W 2006 Terrestrial locomotion of the New
Zealand short-tailed bat Mystacina tuberculata and the
common vampire bat Desmodus rotundus J. Exp. Biol. 209
172536
[13] Elliott K H, Ricklefs R E, Gaston A J, Hatch S A,
Speakman J R and Davoren G K 2013 High ight costs, but
low dive costs, in auks support the biomechanical hypothesis for ightlessness in penguins Proc. Natl Acad. Sci. 110
93804
[14] Kossett A and Papanikolopoulos N 2011 A robust miniature
robot design for land/air hybrid locomotion IEEE Int. Conf. on
Robotics and Automation (ICRA) (Piscataway, NJ: IEEE)
pp 4595600
[15] Bachmann R J, Boria F J, Vaidyanathan R, Ifju P G and
Quinn R D 2009 A biologically inspired micro-vehicle capable
of aerial and terrestrial locomotion Mech. Mach. Theory 44
51326
[16] Itasse M, Moschetta J-M, Ameho Y and Carr R 2011
Equilibrium transition study for a hybrid mav Int. J. Micro Air
Veh. 3 22946
[17] Kalantari A and Spenko M 2013 Design and experimental
validation of hytaq, a hybrid terrestrial and aerial quadrotor
IEEE Int. Conf. on Robotics and Automation (ICRA) (Piscataway, NJ: IEEE) pp 444550
[18] Briod A, Kornatowski P M, Zufferey J-C and Floreano D 2014
A collision resilient ying robot J. Field Robot. 31 469509
[19] Gaston A J and Jones I L 1998 The Auks: Alcidae (Oxford:
Oxford University Press)
[20] Woodward M A and Sitti M 2014 Multimo-bat: a biologically
inspired integrated jumping-gliding robot Int. J. Robot. Res. 33
151129
[21] Socha J J 2002 Kinematics: gliding ight in the paradise tree
snake Nature 418 6034
[22] Socha J J, ODempsey T and LaBarbera M 2005 A 3d kinematic
analysis of gliding in a ying snake, chrysopelea paradisi J. Exp.
Biol. 208 181733
[23] Johansson L C and Aldrin B S W 2002 Kinematics of diving
atlantic pufns (Fratercula arctica l.): evidence for an active
upstroke J. Exp. Biol. 205 3718
[24] Ijspeert A J, Crespi A, Ryczko D and Cabelguen J-M 2007 From
swimming to walking with a salamander robot driven by a
spinal cord model Science 315 141620
[25] Garca-Pars M and Deban S M 1995 A novel antipredator
mechanism in salamanders: rolling escape in Hydromantes
platycephalus J. Herpetology 29 14951
[26] Norberg U M and Rayner J M 1987 Ecological morphology
and ight in bats (mammalia; chiroptera): wing adaptations,
ight performance, foraging strategy and echolocation Phil.
Trans. R. Soc. B 335427
[27] Riskin D K and Hermanson J W 2005 Biomechanics:
independent evolution of running in vampire bats Nature 434
292292
[28] Schutt W, Altenbach J S, Chang Y, Cullinane D M,
Hermanson J W, Muradali F and Bertram J 1997 The dynamics
of ight-initiating jumps in the common vampire bat Desmodus rotundus J. Exp. Biol. 200 300312

Bioinspir. Biomim. 10 (2015) 016005

L Daler et al

[29] Riskin D K, Bertram J E and Hermanson J W 2005 Testing the

hindlimb-strength hypothesis: non-aerial locomotion by chiroptera is not constrained by the dimensions of the femur or
tibia J. Exp. Biol. 208 130919
[30] Quinn R D, Nelson G M, Bachmann R J, Kingsley D A, Of J T,
Allen T J and Ritzmann R E 2003 Parallel complementary
strategies for implementing biological principles into mobile
robots Int. J. Robot. Res. 22 169
[31] Etkin B and Reid L D 1996 Dynamics of Flight: Stability and
Control (New York: Wiley)
[32] Rome L, Funke R, Alexander R, Lutz G, Aldridge H,
Scott F and Freadman M 1988 Why animals have different
muscle bre types Nature 335 8247
[33] Pennycuick C 1991 Adapting skeletal muscle to
be efcient Efciency and Economy in Animal

14

[34]

[35]

[36]

[37]

Physiology (Cambridge: Cambridge University Press)

pp 3342
Lock R J, Vaidyanathan R and Burgess S C 2014 Impact of
marine locomotion constraints on a bio-inspired aerialaquatic wing: experimental performance verication J. Mech.
Robot. 6 011001
Hunter I W and Lafontaine S 1992 A comparison of muscle
with articial actuators Solid-State Sensor and Actuator Workshop 1992. 5th Technical Digest., IEEE (Piscataway, NJ: IEEE)
pp 17885
Huber J, Fleck N and Ashby M 1997 The selection of
mechanical actuators based on performance indices Proc. R.
Soc. A 453 2185205
Tucker V A 1975 The energetic cost of moving about Am. Sci.
63 4139