Documente Academic
Documente Profesional
Documente Cultură
Journal of Mechanisms and Robotics Copyright © 2020 by ASME JUNE 2020, Vol. 12 / 031017-1
[17]. Due to the larger contact area, tracked vehicles are able to such as acrylics, metals, and glass. It is inexpensive, durable, and
carry increased payloads [11], making them more readily scalable. capable of normal adhesion pressures up to 80 kPa. To be useful
While several tracked vehicles for climbing have been previously in a model-based design process, an adhesive model is required
studied [1,2,11,18,19], the literature shows no models that integrate that relates the maximum adhesion capability of the adhesive
properties unique to dry adhesives with models for track design that to the preloading pressure for a given climbing configuration.
evenly distribute the climbing load. This paper will present an This adhesion model for dry-adhesive materials, specifically
approach to integrate dry-adhesive models and track suspension Regabond-S, can be formulated by combining elements from a
systems into a complete model that can be used to design general suction cup model and an elastomeric model, as shown by Powel-
dry-adhesive-based, track-type climbing robot systems. son [12]. The suction cup model makes use of the fact that foam-like
materials are comprised of microscopic pockets of air which
produce suction forces when deformed. As the material is pressed
2 Model for Robot, Suspension, and Adhering onto a surface, the small voids, approximated as spheroids,
Members change volume, and assuming the adhesive behaves linearly elasti-
(a) (b)
Fig. 2 Elements of (a, left) a tracked climbing robot with (b, right) suspension [16]
Kn U = F (9)
while solving for the force vector corresponding to the known dis-
load on the system is the force from gravity acting at the center of
placement locations gives
mass of the robot chassis. Since the robot is assumed to be climbing
in a vertical orientation, this gravitational force is directed parallel to
Fk = K 11 U k + K 12 U u (12)
the climbing surface and track (Fig. 4).
The properties of each element are characterized using appropri- Equation (12) solves the forces at the stationary nodes which rep-
ate stiffness matrices formulated as traditional 1-DOF axial resent the adhering members. These forces are collected in the
members, 2-DOF beams and 3-DOF frame members [24], with
vector f s = fs,1 fs,2 · · · fs,n for the n spring elements repre-
stiffness matrices derived in local element frames as kelement,i. The
senting the adhesive material and represent the forces acting on the
adhesives and springs are represented as axial members, and bound-
adhesive. The outcome of this model predicts the forces acting
ary conditions are applied at the climbing-surface-side of the adhe-
across the suspension and represents the distribution of loading
sive, fixing the nodal translation at the climbing surface boundary.
on the adhesive at the interface of the track and the climbing
Note that this set of boundary conditions assumes that the external
surface.
forces do not exceed the available adhesive holding pressure. The
Using this model, the suspension can be designed to achieve
model requires this check to be made to ensure the boundary con-
specific pressure profiles under an expected robot loading condi-
dition assumptions are met. The elemental matrices are assembled
tion and operation in a climbing configuration. A design procedure
to develop a complete stiffness matrix for the suspension system.
consisting of a numerical search of the suspension spring and link
For this assembly, the local matrices must be transformed into the
parameters is implemented to lead to the desired suspension. For
global frame {Q}. This is done for the chassis, springs, joints,
this work, a desired suspension will provide a compressive force
and links as
on the adhesive for a small portion of the leading end of the
track to ensure the adhesive is fully bonded with the surface. It
kQ T
element,i = T i * kelement,i * T i (8) will then provide a large, distributed region of tension (pull) on
the adhesive to carry the robot payload. Finally, the trailing end
where kQ element,i and kelement,i are the stiffness matrices of the element of the suspension will provide sufficient tension on the adhesive
i in the global and local frames, respectively, and Ti is the transfor- to remove it from the surface. This desired pressure profile is
mation matrix for the element i. These local to global transforma- shown in Fig. 4. The remainder of this paper will assume a sus-
tions account for orientation of the link and suspension members pension designed to give the adhesion pressure profile closely
as shown in Fig. 3 and can also account for a variable geometry approximating the one depicted in Fig. 4, and this pressure distri-
if the suspension is expected to adapt to a non-planar climbing bution will be assumed for the design of the adhesive material
surface (e.g., climbing on a cylindrical surface). After these requirements.
Fig. 5 Intersection of adhesion model with robot model for design and how the solution points shift when (a, left) the suspension
variables or (b, right) robot parameters change
3 Design Guidelines
Equations (15)–(19) define the robot climbing state that satisfies Fig. 6 Suspension design chart—adhesive preload pressure
both the adhesion and suspension model requirements. In particular, versus r2 ratio
Eqs. (17) and (18) show the coupling of pressure regions in the sus-
pension with a linear relationship between preload and force
required to remove adhering members, while Eq. (19) shows the four regions can be manipulated to adjust the preload pressure
tight, nonlinear coupling that exists between the preload and response w1. This is shown to illustrate the influence of combina-
1
maximum adhesion due to the adhesive behavior. For the purpose tions of r1 and r2 in Fig. 6, where r4 is chosen to be 16 and r3 is
of design, those two equations are held separate to create a graphical found with Eq. (21). The horizontal lines indicate the preferred
representation of the climbing condition. Valid climbing conditions range of preload pressures and suggest bounds on the suspension
exist when Eq. (18) intersects Eq. (19), as illustrated in Fig. 5, where region when considering the specific conditions of Fig. 6.
the maximum pulloff pressure of the adhesive w4 adh corresponds to The load-carrying capability of a climbing robot with suspension
the necessary pressure to achieve kinetostatic equilibrium w4 sus for and dry adhesive can be characterized by rewriting Eq. (18) to
a given preload w1. From a design standpoint, increasing the sus- expose the load as
pension stiffness of Region 2 leads to an increase in preload pres-
sure, thus decreasing the slope for a given adhesion force (Eqs. md
wlcf =
(16), (18), and Fig. 5(a)). Suspension stiffness w2 can be increased bL2
by decreasing the factor of safety (FoS) in Eq. (16), resulting in a 2
η r2 r3 r2
decreased slope (Eq. (18)), yielding an equilibrium between the sus- = w4 + − r3 r4 − r42 − w1 (r12 + 2r1 r2 + r1 r3 )
2g FoS FoS
pension and adhesion models with higher preload and adhesion
force (Fig. 5(a)). Meanwhile, decreasing the payload decreases (22)
the y-intercept in Eq. (18) yielding an equilibrium between the sus- where wlcf is the “load-carrying factor” and is a function of the
pension and adhesion model, again with higher preload and adhe- pulloff adhesive pressure w4 and suspension parameters, r1
sion force (Eq. (18) and Fig. 5(b)). Similarly, increasing L and b through r4, which determine the distribution of track pressures.
results in a decrease in the y-intercept (Eq. (18)). The load-carrying factor is written in terms of load-carrying mass
Next, the effect of the suspension design parameters on overall and moment created by that mass (m, d ) and the available width
system performance is considered. In this case, the suspension and length squared of the track contact area (b, L 2). Assuming a
design parameters are assumed to be the relative lengths of the pres- specified selection of the suspension parameters (i.e., specified
sure regions l1 through l4 (Fig. 4), and these determine the distribu- wlcf), Eq. (22) shows that the available payload mass varies linearly
tion of track pressures, w1 through w4. For effective climbing, it is with robot track width b and increases with the square of the track
assumed that a dry adhesive will have a preferred preload, w1. For length. This result is generalized to a variety of dry-adhesive mate-
example, in Regabond-S, higher preload regions increase maximum rials and shown in Fig. 7. The figure demonstrates the load-carrying
pulloff pressure, but it should not be loaded so heavily that the adhe- factor for different dry adhesives commonly used in robotics, as
sive is compressed to the point of permanent damage. This region is demonstrated in the literature, based on η = 90%, FoS = 1, and
approximately between 20 and 70 kPa (Fig. 1) but could be much [r1:r4] = [2, 11, 2, 1]/16. Each adhesive is shown to encompass a
lower for adhesives that primarily rely on shearing forces and range of values representing a minimum and maximum preload as
require normal preloading only to insure sufficient contact. This defined in Fig. 6. These load-carrying factors can be used to esti-
preload is dictated by selecting the pressure region lengths. Consid- mate a theoretical payload from a robot with a given length and
ering the pressure profile shown in Fig. 4, the lengths of the pressure adhesive contact area, demonstrating the significant payloads that
regions are related to the overall length L by the proportionality con- can be achieved with a tracked vehicle and various dry adhesives
stants r1 through r4. These constants are defined such that using a suspension designed in the manner shown in this paper.
li = L * ri (20)
where i = 1:4. This relationship naturally leads to the physical 4 Design Procedure
system constraint
This section will provide a design procedure for track-type robots
4 using a suspension with dry adhesives. The procedure assumes that
ri = 1 (21) m, d, FoS and w1 are chosen and will guide the selection of track
i=1
and suspension variables L, b and r1 − r4. Choosing m, d and FoS
Furthermore, the pulloff region associated with r4 is a function of are based on an expected payload and climbing stability margin,
the track element size and elasticity of the adhesive and is, therefore, while the choice of w1 is based on a preferred rate of preload for
also considered a constant. With these two constraints, two of the the dry adhesive as discussed in Sec. 3. Finally, the effective track
5 Experimental Verification
5.1 Prototype Design. The design process laid out in Sec. 4 is
now demonstrated experimentally with the creation of a tracked
robot using the dry-adhesive Regabond-S to climb acrylic. Using
a payload of 20 kg as the driving factor behind the design, the
creates an area of compression (Region 1). After the adhesive has limits of the prototype were assessed. After each successful climbing
been preloaded, the suspension system uses constant force springs test, additional payload was added to the robot. Eventually, the
to apply tension to the subsequent 20 track elements, “pulling” robot’s total mass equaled 40 kg. At this mass, the robot was still
the robot into the climbing surface (Region 2). After this, the adhe- able to climb the entire height of the wall and remain there until it
sive is re-compressed as it passes by the caster wheels due to the was purposefully removed. Figure 9 shows three different views
nature of the force balance (Region 3). As the very last track of the 40-kg prototype robot scaling the climbing wall. The robot
element is pulled from the surface at the end of the adhered outperforming the design expectations can most likely be contrib-
length of track, it also creates a tensile force (Region 4). uted to the conservative estimates chosen for FoS and η. It should
The robot was designed to match the pressures predicted by the be noted that there is a relationship between the tension in the sus-
design procedure: w1 ≈ 40 kPa, w2 ≈ 12 kPa, w3 ≈ 175 kPa, and pension springs and the overall payload. As mass is added to the
w4 ≈ 60 kPa (dashed curve in Fig. 8). Figure 8 shows an acceptable robot, the suspension springs are tightened to maintain the desired
agreement between the actual and desired suspension pressure pro- preload. The relationship between the mass, tension in suspension
files for Region 1’s preload pressure and Region 2’s load- springs, and mass are demonstrated in Fig. 5.
distributing pressure. However, the magnitude of the pressures in
Regions 3 and 4 produced by the robot is smaller than predicted.
In particular, it is believed that the magnitude of Region 4’s pressure 6 Discussion
is reduced because of a combination of pulling and twisting that is
applied to the track as it leaves the end of the suspension and travels The integration of load-distributing suspension and dry adhesives
around the rear sprocket. Some deviations between the measured are shown to offer high theoretical performance for climbing vehi-
and desired model are also noted. The measured pressure profile cles. The desired adhesive loading profile chosen for this work was
departs from the ideal model during regions of transition from com- reduced to a model consisting of four regions of constant pressure
pression to tension on the adhesive members. This is in part due to applied to the adhesive. This modeling, fabrication, and testing of
clearances in the prismatic or sliding suspension elements. The full the prototype robot, with results shown in Fig. 8, demonstrates
potential pressure is only fully realized once the track element is the ability in a practical design to reasonably achieve this desired
completely engaged in the suspension. Additionally, a prototype loading profile. The simplified loading profile allows the designer
suspension is sensitive to tolerances between track elements, as to develop intuition about the complex relationships between the
offsets between subsequent elements can cause concentrated load- design parameters that would not be possible if a more complex
ings. This is seen most prominently in Region 2 where the pressure model was used throughout. As a prime example, the load-carrying
profile oscillates about the nominal pressure. factor (Eq. (22)) shows that the available payload is proportional to
a robot length dimension cubed—the square of the robot’s track
length multiplied by the track width (bL 2)—which is dimensionally
5.3 Climbing Performance. A climbing wall was constructed equivalent to volume. This implies that climbing robot load capacity
from 19-mm thick clear acrylic to test the robot climbing perfor- is not necessarily limited by scale as reported in Refs. [14,28].
mance. Initially, the prototype robot was tested independently of Rather, such a climbing machine can be designed to perform at
additional payload to ensure the suspension and adhesive were func- higher payloads. This significant departure from the current litera-
tioning according to the model. Next, an additional mass was added ture in the design of climbing robots is further highlighted by com-
to the robot’s chassis until the total robot’s mass (called payload) paring the reported load-carrying capacities of 25 other climbing
equaled the 20-kg design payload, and additional climbing tests robots. For comparison, Fig. 10 shows payload plotted against the
were performed. Again, the robot was able to successfully climb adhered area for the proposed prototype as well as other
the entire height of the wall, where it remained until it was purpose- dry-adhesive climbing robots found in the literature. Especially
fully removed. Successful climbing was further ensured by observ- notable is the performance of other track-based systems. It should
ing that, as new adhering members enter the suspension, they also be noted that, although the proposed prototype demonstrates
become fully adhered to the climbing wall. Finally, the upper a payload that is an order of magnitude higher than other dry-
adhesive climbing robots in the literature, not all of the robots in the inspection—for example, manufacturing, maintenance, and rescue-
literature were trying to maximize the payload. type jobs. Further, this work has detailed a design procedure for
One advantage of track-based systems is that they tend to offer such systems. Due to the tight coupling of the adhesive pressures,
more surface area than legged systems. However, previous robots wherein dry adhesives require a preloading in order to achieve adhe-
in the literature do not realize the payload increase that should sion, the process of designing these suspensions requires simulta-
result from using an increased adhesive area with corresponding neous modeling of the adhesive.
load-distribution suspension. By using a force distributing suspen- Some additional guidelines for the design for maximum payload
sion to utilize the entire adhered area, this work demonstrates pay- capability are provided. For maximum payload, the r2 region should
loads an order of magnitude higher than previous designs. This be as large as possible with r1 tuned such that the adhesive is pre-
allows robots of this type to be used for tasks other than loaded as strongly as possible without damaging the adhesive.
Fig. 10 Payload plotted against adhesion area for several dry-adhesive climbing
robots
Appendix
1/2
{[C1 ln(w1 )2 + C2 ln(w1 ) + C3 w1 + C4 ] + C5 ln(w1 ) + C6 }
w4 = ± (A1)
C7
2
• C1 = G2 = 1.6216 * 103 kPa
• C2 = 2EG ln(10) + 2GK ln(10) + 2πDGR2 atm ln(10) = 1.3290 * 105 kPa2
• C3 = 4πDR2 atm ln (10)2 = 1.5652 * 103 kPa
• C4 = E 2 ln (10)2 + K 2 ln(10)2 + 2EK ln(10)2 + D2 R4 atm2 π 2 ln (10)2 − 2πDER2 atm ln(10)2 + 2πDKR2 atm ln(10)2 = 1.6928 * 106 kPa2
• C5 = G = 40.269 kPa
• C6 = K ln(10) − E ln(10) + πDR2 atm ln(10) = −1.5493 * 103 kPa
• C7 = 2 ln(10) = 4.6052