Mechanism and Machine Theory 173 (2022) 104832

Contents lists available at ScienceDirect

Mechanism and Machine Theory

journal homepage: www.elsevier.com/locate/mechmt

Analysis of climbing in circular and rectangular pipes with a

reconfigurable sprawling robot
Ori Inbar, David Zarrouk *
Department of Mechanical Engineering, Ben Gurion University, Negev, Israel

A R T I C L E I N F O A B S T R A C T

Keywords: This paper presents an analytical description and experimental results for a reconfigurable field
Climbing robots robot that can climb inside circular and rectangular pipes. The robot is fitted with two mecha­
Field robots nisms that allow it to change its width and height and shift its center of mass (COM) to adapt itself
Mechanical design
to the size of the pipe. We start by describing the kinematic model of the robot as a function of its
Reconfigurable robot
Crawling robot
sprawl and four bar extension mechanism (FBEM). Next, we develop a force analysis based on the
Sprawl tuning robot’s geometry, its configuration, the position of its center of mass (COM), the diameter of the
pipe, and the coefficient of friction (COF). We then develop strategies for driving, climbing and
transitioning between the two modes. Although a high COF increases the robot’s grip, it reduces
its ability to reconfigure its shape, which it needs to transition between its climbing/driving
modes. Based on this analysis, we designed a control algorithm comprised of actuation sequences
to automatically drive the robot inside pipes, including the transition phases. The results show
that the robot successfully executed its climbing tasks (see video).

1. Introduction

PIPE climbing robots have multiple usages such as inspection in chemical industries or for public sewers, where it is crucial to
evaluate the condition of the pipes and take precautionary measures whenever necessary. They can also be used for search and rescue
missions since a small robot can climb through pipes and chimneys to look for survivors. To climb, these robots must apply continuous
high-pressure forces to the pipes, change their external diameter, and transition from horizontal crawling/driving to vertical climbing.
Basic climbing capabilities have been reported in worm-like robots [1–5] and wave robots [6,7] which rely on high friction [8] to
climb or use a tether to support their weight [9]. Typically, however, they are not designed to substantially modify their diameters to
adapt to different pipe diameters, execute turns along pipe curvatures and/or transition from horizontal crawling to vertical climbing.
Snake robots [10–12], which are characterized by their overly redundant kinematics, can change their external shape to adapt
themselves to different pipe diameters and crawl through pipe branches by using multiple locomotion gaits to advance [13,14].
Nevertheless, they advance by sliding, which is very energy-consuming and often necessitates tethering or large batteries that limit
their range and performance.
Multiple robots specifically designed for pipe inspection have been described in the literature [12–26], and some are even
commercially available. These robots are fitted with multiple joints and linkages (in series like snakes or with multiple parallel

This study was supported in part by the Helmsley Charitable Trust through the Agricultural, Biological and Cognitive Robotics Initiative and by
the Marcus Endowment Fund, both at Ben-Gurion University of the Negev.
* Corresponding author.
E-mail address: [email protected] (D. Zarrouk).

https://doi.org/10.1016/j.mechmachtheory.2022.104832
Received 5 November 2021; Received in revised form 25 January 2022; Accepted 7 March 2022
Available online 24 March 2022
0094-114X/© 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

mechanisms around their main body) to enlarge their external diameter to contact the walls of the pipe. To advance, they use wheels or
tacks that allow them to apply high forces with low energy consumption. Highly reconfigurable multi-linkage robots can be fitted with
wheels to provide thrust and reduce the friction [27]. Some of these robots have exhibited excellent driving and climbing performance
even in T branches [24].
Other types of propulsion techniques have also been suggested, such as spiral driving [28,29] and the two-mass inertial approach
[30]. Li et al. used a screw-drive method to change the width and contact angle of the wheels with the walls, thus making it possible to
turn and adapt to different pipe diameters [31]. Li et al [32]. described a robot that can spontaneously adjust its configuration while
rolling through pipelines. It can be used to inspect pipelines before commissioning them.
The RSTAR robot [33] presented here (Fig. 1) is a minimally actuated reconfigurable field robot that belongs to the family of STAR
robots (STAR [34], 1STAR [35], AmphiSTAR [36], and FSTAR [37]) all of which are fitted with a sprawling mechanism allowing them
to reconfigure their mechanics. The RSTAR, which only has two sets of wheels and is actuated by four motors, was originally designed
to be a medium-speed reconfigurable field robot that can drive or crawl (with a turtle gait) over different terrains and overcome
challenging obstacles. Using its sprawl and four-bar extension (FBEM) mechanisms, this robot can increase its width and height by
three-fold and shift the position of its center of mass (COM) relative to its wheels. It weighs 484 g including its battery and control
board, and has a characteristic length (rear wheel axle to front wheel axle) of 14.5 cm. The robot can be controlled by a human operator
or programmed to perform specific sets of actions using its own onboard controller.
In this paper, we analyze the factors affecting crawling inside pipes in an unmodified RSTAR. This analysis is used to develop
climbing strategies implemented in the robot controller to enable automatic maneuvers. The analysis in this work focuses on actions
that can be performed at low speeds without using dynamic maneuvers based on the inertia of the robot.
The remainder of this paper is organized as follows. We briefly present the mechanical design of the robot in Section 2. In Section 3
we present the kinematics of the robot inside a pipe and a force analysis of crawling horizontally and climbing vertically Section 4.
describes multiple experiments in which the robot was programmed or controlled to crawl inside pipes and transition from horizontal
driving to vertical climbing.

2. Summary description of the robot design

2.1. Robot description

The RSTAR, which we recently reported [33] and is presented in Figs. 1 and 2, is an upgraded version of the original STAR [34]. The
RSTAR is fitted with a four-bar extension mechanism (FBEM) on its sprawl mechanism which allows it to increase the distance between
its body and legs and move its COM in the fore-aft and vertical directions. The RSTAR is primarily designed as a medium speed search
and rescue robot that can overcome obstacles using a combination of its sprawling and extension mechanisms. The RSTAR can also
crawl over different challenging surfaces such as gravel and grass, as well as granular and slippery surfaces using its turtle gait.
The robot can increase its height and width 3-fold, which allows it to penetrate small cracks and/or pipes and crawl beneath
obstacles. It can also expand its width to be able to apply pressure to the surrounding walls to increase its friction or climb vertically.
Further details on the design of RSTAR and a movie featuring its capabilities can be found in our previous work [33].

2.2. Robot design

The RSTAR has minimalistic design in which each set of wheels (left and right) and the sprawl and FBEM mechanisms are actuated
by a single motor (4 motors in total). It weighs 484 g including its battery and control board, and has a characteristic length (rear wheel
axle to front wheel axle) of 14.5 cm. The rigid body core houses its controllers, onboard batteries, and receivers. The two sides of the
robot are phased together and move symmetrically relative to its center.

Fig. 1. The reconfigurable RSTAR climbing vertically in a transparent pipe.

2
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Fig. 2. (A) An isometric view of the RSTAR. (B) the robot is depicted in negative sprawl and (C) in positive sprawl. The COM can be shifted forward
(D) or backward (E). See video.

2.2.1. The sprawl and FBEM mechanisms

The sprawl mechanism varies the relative angle ρ between the legs and the main body (see Fig. 2). The sprawl angle is defined as ρ
= 0 when the legs are coplanar with the ground while the positive sense is downward. The sprawl angles at both sides are actuated
symmetrically using a single motor to insure an identical sprawl on both sides. The sprawl can be actuated from -90 to +90 , allowing
◦

the robot to drive even when flipped over.

The four-bar extension mechanism (FBEM) is attached to the sprawl mechanism and rotates together with it. The two parallel bars
of the FBEM that connect the sprawl mechanism to its set of legs are identical in length Lbar. The rotation angle of the four-bar
mechanism is denoted by γ and can be varied from (-75 to +75 ). The rotation angle γ is zero when the two bars are perpendic­
◦ ◦

ular to the body and wheel housing.

2.3. Actuation and sensors

The RSTAR is actuated by a total of four motors. For driving, it implements two 12 mm diameter, 6–9 V, high power, 1:300 gear
ratio Pololu motors which can be fitted with encoders. The driving motors provide a torque of 50 Ncm. Similar motors but with a
1:1000 gear ratio are used to actuate the sprawl and FBEM mechanisms. The robot is powered with two 3.7 V 800 mAh LiPo batteries
connected in series. The angles of the sprawl and FBEM mechanisms are measured using potentiometers attached to them. The robot is
controlled by a Teensy 3.5 controller which allows for automatic sequencing or remote control by an operator using a 2.4 GHz RC
digital controller. In the automatic sequencing mode, the controller can change the sprawl angle and FBEM or drive the wheels forward
based on a pre-planned sequence. The controller can also actuate different motors in parallel; for example, to keep the width constant
while moving the COM.

2.4. Manufacturing

The RSTAR body is manufactured primarily by 3D printing. We used both SLA and FDM technologies whose accuracies are
respectively 0.05 mm and 0.2 mm. Some of the gears that need to sustain heavier loads were 3D printed in brass. Considerable efforts
were made to simplify parts replacement such as the motors, the controller, the bars of the FBEM, and the wheels. Easy part
replacement emerged as essential for experimentation since the robot often fell during the initial experiments while climbing vertically
in the pipe. This caused damage to the mechanical parts and some motors burned up while the robot was applying side forces at full
duty cycle.

2.5. Width of the robot and location of the center of mass

The RSTAR can vary its width using either its sprawl or its FBEM mechanism, or both. Alternatively, by actuating its sprawl and
FBEM mechanisms, it can shift its center of mass (COM) without modifying its width.
Denoting the term L1 + L2 + R by LT, the position of the COM in the fore-aft direction relative to the flat configuration where both ρ
and γ are zero is:
2Lbar sin(γ)mbody
xCOM = (1)
mbody + 2mleg

In the vertical direction, the position of the COM is:

(LT + Lbar cos(γ))sin(ρ)mbody
zCOM = . (2)
mbody + 2mleg

3
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

The width W (ρ,γ) of the contact point of the wheels with the pipe is (as defined in Fig. 3) is:
W(ρ, γ) = 2L3 sin(ρ) + Lw + (2LT + 2Lbar cos(γ))cos(ρ). (3)
Alternatively, for a given width W, the FBEM angle γ can be calculated as a function of the sprawl:
( )
W − 2L3 sin(ρ) − Lw − 2LT cos(ρ)
γ = acos . (4)
2Lbar cos(ρ)
By coordinating its sprawl and FBEM mechanisms, the robot can shift its center of mass without changing its width. The range of
COM positions for W = 20 cm and W = 25 cm is presented in Fig. 4, and shows that the COM can be shifted by 6 cm and 12 cm
respectively in the fore-aft and vertical directions without modifying its width. Moving the COM serves to increase the robot grip with
the wall. For example, if the robot starts climbing when it is slightly tilted, the COM can be shifted closer to the pipe wall to increase the
hold of the front or back wheels.

3. Pipe climbing analysis

Climbing in a pipe is generally composed of three stages: transitioning from a horizontal to a vertical configuration, climbing
vertically, and transitioning back from the vertical to the horizontal configuration. In order to successfully climb a pipe, the robot must
meet multiple requirements. Its sprawl and FBEM mechanisms must be able generate large side forces Fside on the pipe to produce
sufficient friction to match its weight. The driving wheels must generate enough thrust to advance and overcome this weight. Aside
from these two basic stipulations, the robot must be able to change its width in the pipe to increase its contact with the walls (Fig. 5 A
and B). In the following analysis we use the given values of the actual robot as listed in Table 1.

3.1. Horizontal to vertical quasi-static transitioning

The first step in climbing consists of transitioning from horizontal driving to vertical climbing. Success is a function of the COF of
the wheels with respect to the ground and the wall, respectively µG and µW and the relative position of the COM is αL (α = 0 and α = 1
imply respectively that the COM is above the back and front wheels, whereas α = 0.5 implies that the COM is in the center of the robot).
Before starting to climb, when the robot is still horizontal, the front and back wheels provide forward thrust. As the robot sprawls
slightly upwards, the front wheels disengage from the ground. At this stage, the rear wheels provide a horizontal friction force FR which
applies force to the walls while the front wheels provide the vertical friction force FF that lifts the body upwards (Fig. 6). Assuming
quasi-static climbing, the sum of the torque around the contact point of the rear wheels with the ground can be expressed as:
FF (L + R) + NF ⋅ R = αL ⋅ mg, (5)

where m is the mass and g is the gravity. Given that the maximum friction force that the wheels can apply vertically is FF=μNF, and
rearranging Eq. (5), we obtain:
αL ⋅ mg
NF = . (6)
μW (L + R) + R
The normal front force with the wall is equal to the rear friction FR. In sliding conditions, the rear friction force is equal to the
normal rear force times the ground COF (FR=μGNR). The sum of the forces in the vertical direction yields:
FR = μG (mg − μW NF ) = NF . (7)

Fig. 3. The dimensions of the RSTAR and the position of its COM as a function of its sprawl angle ρ and its FBEM angle γ.

4
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Fig. 4. The possible positions of the COM for two different values of the width W = 20 cm and W = 25 cm.

Fig. 5. After initially entering the pipe in the horizontal mode at high sprawl (and small width), the RSTAR must be able to increase its width (A to
B) and transition from horizontal driving to vertical climbing in the pitch (C) and yaw (D) axes.

Table 1
The main specifications of the RSTAR.
Robot characteristics Value

Characteristic length LR = 14.5 cm

Wheel/whegs radius R = 3.25 cm
Lwidth; Lbar; L1; L2; L3 5.6; 6; 1.4; 1.8; 1.5 [cm]
Sprawl range -90→90◦
FBEM range -75→75◦
Maximum width Wmax=30.5 cm
cm
Minimum width Wmin = 8.6 cm
Robot mass m M = 0.484 kg
Main body mass 0.252 kg
Driving mechanism mass 2 × 0.116 kg
Battery 800 mAh, 7.4 V
Robot speed VR = 0.4 m/s (0.25 m/s vertical)
Max sprawl rate 35◦ /s; 0.61 rad/s
Max FBEM rate 26 eg/s; 0.45 rad/s

Substituting the value of NF into Eq. (6), we obtain:

(R + L − αL)μG μW + RμG − αL = 0. (8)
Alternatively:
RμG + (R + L)μG μR
α= . (9)
μG μR L + L
If both COF are identical and the wheel radius is small relative to the length L, Eq. (8) becomes:

5
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Fig. 6. (Top) The forces acting on the RSTAR as it starts climbing the walls of a pipe. (Bottom) The relative position of the COM α as a function of µG
and µW.

( )1
αL 2
μ= . (10)
L − αL
The location of the COM which permits transitioning as a function of the ground and wall COF is presented in Fig. 6. These
conditions are not symmetrical relative to the two COF since a higher wall COF is required than for a ground COF. For example, if α =
0.5 and the ground COF is 0.6, the wall COF must be nearly 0.8 in order to climb. In our case, given that the measured COF is 0.6
between the plexiglass and the robot’s wheels (see Section 0), the required α must be below 0.4 which is easily achievable with our
robot (0.3≤α≤0.7).

3.2. Changing width during horizontal driving in a pipe

The robot contacts the pipe at two points forming an arc of 2θ (2θ ≤ 180∘ ). In the horizontal driving state, along the experiments, it

Fig. 7. The forces acting on RSTAR as it attempts to increase its width during the horizontal driving.

6
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

was physically impossible to increase the width of the robot (even with infinite torque and force) beyond a maximal arc 2θHmax. This is
due to the side friction of the wheels with the pipe (see Fig. 7). Forcefully attempting to increase the width beyond 2θHmax damaged the
motors as well as the sprawling and FBEM mechanisms.
When in the horizontal state, the forces acting on the robot are its weight mg, the normal forces N of the pipe, and the friction of the
wheels against the pipe. As the robot attempts to increase its width, it applies a side force Fside which increases the normal force N and
generates a friction resisting force µN. The static force equilibrium in the horizontal direction is:
Nsin(θ) + μNcos(θ) − Fside = 0. (11)
In the vertical direction, the equilibrium is:
Ncos(θ) − μNsin(θ) − mg = 0. (12)
The side force Fside as in Eqs. (11) and (12) as a function of θ and the COF is presented in Fig. 8. The arc 2θ can be calculated using:
( )
W
2θ = 2asin . (13)
2Rpipe

Given that the robot can apply a sprawl torque of 65 Ncm (in the safe mode at 60% duty cycle), the maximum force that the robot
can generate sideways is a function of its sprawl (presented as the dashed line in Fig. 8). In the experiments while in the safe mode, the
maximum arc that the robot could achieve was 2θHmax = 75 when the COF was 0.6, but increased to 140 and 145 when the COF was
◦ ◦

0.1 or 0.05.

3.3. Vertical climbing in the center of the pipe

To climb vertically, the RSTAR must apply a sufficient side force to the walls of the pipe to produce enough friction to overcome its
own weight (Fig. 9).
Assuming that the friction and side force are equally divided between the top and bottom wheels, the average vertical friction force
per wheel is:
mg
fV = . (14)
4
The required side force is:
2fV mg
Fside > = . (15)
μ 2μ

where the equality holds for the minimum required side force. A safety margin must be added to ensure that the robot does not slip
since the COF can vary slightly in different parts of the pipe. Note that if one set of wheels (either front or back) disconnects due to
variations in pipe diameter for example, the entire side force will be applied to the other set of wheels. However, the climbing re­
quirements will remain the same.

3.3.1. General climbing model

Stricter conditions are required to satisfy equilibrium if the robot is not climbing exactly in the center of the pipe. This is because the
contact force that the pipe applies to the wheels has a component in the z (vertical) direction which causes the robot to slide.
The force equilibrium at the contact point of the wheels with the pipe in the z direction is (see Fig. 1):
(NT + NB )sin(θ) + (fTH + fBH )cos(θ) − Fside = 0. (16)

where NT and NB denote respectively the normal forces applied to the pipe walls by the top and bottom wheels. The friction forces fTH
and FBH are respectively the horizontal top and bottom friction forces acting on the wheels. The vectorial sum of the horizontal and

Fig. 8. The required side force Fside and the maximum angle θHmax as a function of the COF. The dashed line representing Fside is the maximum force
that the RSTAR can generate.

7
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Fig. 9. The normal and friction forces acting on the robot as it climbs vertically in the center of a tube.

vertical friction forces is equal to the total friction force:

( 2 2
)12
fTH + fTV = fT ≤ μNT
( )12 . (17)
2 2
fBH + fBV = fB ≤ μNB

Assuming that the weight is equally supported by all 4 wheels, the fore-aft component (z direction) of the friction force becomes:
mg
fTV = fBV = . (18)
4
In the y direction, the force equilibrium is:
(NT + NB )cos(θ) − (fTH + fBH )sin(θ) = 0. (19)
Inserting Eqs. (17) and (18) into Eqs. (16) and (19) and solving them yields the minimum arc θVmin in which the robot can still
adhere to the pipe as a function of the side force Fside. The results are presented in Fig. 11 and compared to the maximal arc 2θHmax.
Interestingly, for any given COF, the two angles θVmin and θHmax never intersect. The physical implication of this result is that by using
the sprawl and FBEM mechanisms, it is impossible to sufficiently increase the width of the robot while in horizontal mode to transition
from horizontal driving to vertical climbing.
If the COM is shifted by δCOM (positive or negative) relative to its wheels, Eq. (19) is replaced by the torque equilibrium around the
bottom wheels:
( )
mgδCOM + 2LR fT H sin(θ) − NT cos(θ) = 0. (20)

where LR, the characteristic length, is the distance between the axes of the front to the back wheels (see Table I. ).
If δ is positive, more load is applied to the top wheels, which may then lose contact with the pipe. In this scenario, the robot may fall
on its back or slide down the pipe. If δ is negative, the bottom wheels may lose contact with the walls but the robot will remain clinging
to the pipe. The minimum θVmin required for climbing as a function of the side force is presented in Fig. 12.

3.4. Reducing the COF by inducing sliding in the fore-aft direction

Given that the actual COF of the wheels with the pipe is 0.63 (based on our experiments), the robot cannot extend its width beyond
θHmax = 37 in the safe mode or 46 at 100% of the duty cycle (see Section 5). A special maneuver is required to increase the width to the
◦ ◦

climbing requirements of θVmin = 65 (see Fig. 11).

◦

One way to increase the width is by pitching the robot upwards in the elbow and actuating the sprawl and FBEM mechanisms.
Another strategy is to decrease the friction resistance sideways by producing sliding in the fore-aft direction (while sliding sideways). If
the speed of the driving wheels is denoted by VR and Vside is the speed of the width extension (see Fig. 13), the sliding speed is:
Vsliding = Vside + VR
( )1 . (21)
Vsliding = Vside 2 + VR 2 2

According to Coulomb’s law, the direction of the friction force is antiparallel to the sliding speed. As a result, the friction force that
opposes the expansion of the width is the side component alone fside:
Vside
fside = f ( )12 . (22)
VR 2 + Vside 2

8
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Alternatively, the effective side COF µES is:

Vside
μES = μ ( )12 . (23)
VR 2 + Vside 2

For example, in our robot, if the wheels slide at VR = 40 cm/s in the fore-aft direction and sideways at Vside = 4 cm/s, the effective
side COF µES is nearly one-tenth of the actual COF (in our case it becomes 0.063).
Note that sliding in the fore-aft direction can be done in two ways. The first is by driving forward during the transition from
horizontal to vertical. The second is by driving only one side of the robot forward/backward while the second side remains static (the
advancing side will slide).

3.5. Vertical to horizontal transitioning

In order to transition from a vertical to a horizontal configuration, the robot must shift its COM to one side by a distance d and then
decrease its normal forces. In
Fig. 15, the COM is shifted right by d, which decreases the robot’s total side force (the sum of the normal forces) to φmg/μ where φ is
less than 1. The friction with the walls is not sufficient to keep the robot static and it will rotate (clockwise) and fall. The dynamic
equations of motion are:
FF + FR − mĝz = maCOM , (24)

where FF and FR are respectively the friction forces on the front and rear wheels. The vector aCOM is the vector acceleration of the COM.
Note that the friction forces oppose the motion which are exerted in the opposite direction to the sliding direction:
μNF μN F
FF = V F ; FR = VR . (25)
|VF | |VR |

where VR and VF are the vector speed of the rear and front wheels. The sum of the torque acting on the robot relative to its COM is:

(XF − XCOM ) × FF + (XR − XCOM ) × FR = mθ̈̂

y. (26)
By numerically solving the two ordinary differential equations (ODE) in Eqs. (24) and (26), we obtain the location and orientation
of the robot. The location and orientation as well as the linear and angular speeds of the robot for α= 0.3, 0.5 and 0.7 are presented
respectively at the top and bottom of
Fig. 15 and in Table 2. Although the location of the COM α (α = 0 and α = 1 imply respectively that the COM is above the back and
front wheels while α = 0.5 implies that the COM is in the center of the robot), this does not substantially affect the linear speed of the
COM during its fall, since a smaller α increases the rotation speed so that the robot falls a shorter distance before landing on the ground.
On the other hand, a larger α results in more shifting of the COM towards the target (rightward in this example as in Fig. 14).

4. Experiments

This section presents the results of experiments conducted on the robot. We measured the forces that the RSTAR is able to apply to
the walls, experimentally evaluated the mechanical model, and tested the climbing algorithm in a 25 cm circular plexiglass pipe and in
a rectangular pipe. We measured the speed of the robot and found that it can drive horizontally at 40 cm/s and climb vertically in the
pipes while applying a side force at 25 cm/s.

4.1. Measuring the side forces that RSTAR can apply

We developed an experimental system to measure the forces that RSTAR can apply to a pipe wall. The system, as presented in
Fig. 16, is composed of two plexiglass walls (material similar to a pipe) whose distance could be adjusted. The left wall was attached to
a 6 DOF Nano 25 force sensor that measured the normal and thrust forces the robot applied to the left wall. Given that the RSTAR can
change its width using either its sprawl or its FBEM mechanisms, we measured the forces that each mechanism could apply separately
(the robot can apply the minimum of the two). In these experiments, the duty cycles of the sprawl and FBEM motors were limited to
60% in order to avoid damaging the motors or the mechanisms. Throughout the experiments, we set the width of the walls to be 25 cm
(similar to the diameter of the experimental pipe Rpipe).
In the first part of this experiment, the FBEM was set to zero degrees and the sprawl was actuated while the force was being

Table 2
The location of the COM and rotation speed while transitioning from a vertical to a horizontal position.
α Δt [s] ΔYCOM [cm] ΔXCOM [cm] VCOM [cm/s] Ang. Vel. [deg/s]

0.3 0.108 3.1 -0.2 28.7 -18.0

0.5 0.122 4.0 0.1 19.3 -18.9
0.7 0.146 5.8 0.5 81.7 -16.9

9
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

measured using the force sensor. The normal side force Fside was 11.5 N. In the second part of the experiment, the FBEM was set to zero
degrees and then the sprawl mechanism was actuated (the sprawl was 36 ). The robot applied a side force of 9.3 N to the walls which
◦

translated into a sprawl torque of 65 Ncm. The results of the experiments are presented in Fig. 17.

4.2. Measuring the coefficient of friction (COF)

In the second experiment (Fig. 18), we measured the side and friction forces (i.e., the thrust forces) that the wheels could generate
on the wall while they were sliding. The robot, which was prevented from advancing, was programmed to apply an initial side force
and then drive forward while the normal and friction forces were being measured. The COF, which was calculated by averaging the
ratios of the thrust to the normal forces from 5 runs, was 0.63.

4.3. The maximum width the robot can achieve when in the horizontal configuration

In the first part of this experiment (Fig. 19), we measured the maximum θHmax that could be achieved by simply expanding the
width of the robot using the sprawl and FBEM mechanisms (without sliding forward). Starting in (A), the robot was repeatedly unable
to expand its width beyond θHmax = 45 (B) even at a 100% duty cycle. This result was consistent with our theoretical expectation of θ
◦

= 46 as shown in Figure 8. Further attempts to increase the width resulted in damage to the sprawl mechanism.
◦

In the second part of the experiment, the robot was attached to a wire (C), which prevented it from advancing, and was driven
forward. The wheels slid in the fore-aft direction while the sprawl was increased. In this case, the robot was able to expand its width to
an arc of 2θ = 134 (D). This experimental result was 4% less than our theoretical expectation of 140 (see Section 3 A and Fig. 8).
◦ ◦

4.4. Climbing vertically in a pipe

The robot was tested while it was climbing vertically (see Fig. 20 B1 and B2). When δ was negative, the robot appeared to be
pressing against the pipe while its bottom wheels were slightly away from the wall. However, when δ was positive, the robot quickly
lost hold and, in some cases, fell all the way down the pipe. The climbing speed of the robot was 25 cm/s. The climbing angle of θ = 69
◦

was slightly larger than the calculated θVmin = 65 in our theoretical estimates. Based on this result and given that the maximal width of
◦

the robot is 30.5 cm, the robot can climb in pipes ranging from 12 cm (the smallest diameter which allows it to fin in) up to 33 cm
(31.5/sin(69o)). The maximum diameter can be further increased if the bars of the FBEM are increased. For example, if the bars are
increased to 10 cm, the minimum diameter increases to 16 cm but the maximum diameter becomes 42 cm.

4.5. Automatic climbing and transitioning from horizontal driving to vertical climbing and vice-versa

Based on our analysis and preliminary experiments, we programmed the robot to automatically climb in a pipe and perform
transitions. We first demonstrate the transition from the horizontal to vertical climbing program that consisted of five sequences (see
Fig. 21 and video).
The robot enters the pipe (A) and advances (B) until it reaches the pipe’s elbow (C). At this point, the robot continues advancing
while it increases its width (D) until reaching a sufficient width (E) and then continues climbing vertically in (F).
To reliably transition from vertical to horizontal driving, the robot must shift its COM towards the elbow of the pipe. Starting in (A)
as the robot is climbing vertically, the robot reaches the elbow of the pipe (B) and shifts its COM (C) forward without modifying its
width to ensure that it is still applying sufficient side force to grip the pipe. At this point, the robot continues driving up the pipe (D).
Note that vertical to horizontal transitioning is the riskiest maneuver since if the robot slides, it can fall and sustain damage. After
finding values of the sprawl and FBEM actuation as well as the driving distance of the wheels from (A) to (D) and the motion of the
COM, the movement was uploaded to the controller which could thus perform the sequence automatically. The automatic sequencing
had a high success rate in transitioning along the pitch axis, unlike the human controller who was generally unsuccessful in performing
the same maneuvers (see Fig. 22).
In the third experiment, the robot drove forward using its left side alone (A) and actuated its sprawl to increase its width. As a result,
the right side remained static while the left side reached the top of the pipe (B). At this point, the robot continued advancing towards
the elbow (C) and rotated along its yaw axis. It continued climbing until the next elbow (D) and rotated clockwise (E) and continued
driving through the pipe (F). Note that the transition along the yaw axis was relatively simple and this experiment was performed
successfully by a human operator (Fig. 23).
In the last experiment (Fig. 24), the robot climbed in a rectangular canal. Starting from a horizontal configuration on the floor (A),
the robot first pitched upward by contacting the wall (B), then climbed along the canal (C-D) until it reached the top (E) and tran­
sitioned back to a horizontal configuration (F)

5. Conclusions

This paper presented the mechanics, conditions and strategies enabling the RSTAR reconfigurable robot to climb inside circular and
rectangular pipes. The RSTAR robot was originally developed for search and rescue applications and has multiple unique capabilities
that enable it to overcome a variety of obstacles. It has a low weight and relatively high energy efficiency. The RSTAR can drive at
speeds of up to 40 cm/s horizontally and up to 25 cm/s when climbing inside pipes. We developed driving/climbing strategies and

10
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Fig. 10. The forces acting on the robot as it climbs vertically in a pipe.

Fig. 11. The required side force Fside for climbing vertically as a function of θ for 3 different COF. The dashed line presents the maximum θ which
can be reached by increasing the side force when in the horizontal configuration.

Fig. 12. The required side force Fside for climbing vertically, as a function of the minimum θVmin required for climbing, for δ = 1 cm and δ = 2 cm for
3 different COF (0.6, 0.8, 1).

determined the conditions of locomotion by calculating the required side forces and friction forces as a function of the robot’s width
relative to the pipe, and the position of its COM.
Our analysis shows that the robot can climb vertically even if it is not in the center of the pipe. With a COF of 0.63, it can climb at an
arc (2θ) of only 130 (and theoretically at 100 if the COF is 1). As a result, the robot can climb inside pipes whose diameters range from
◦ ◦

12 to 33 cm. The diameter size can be further increased if the length of the bars in the FBEM is extended. For example, if the length of
the bars is increased from 6 to 10 cm, the maximum diameter of the pipe that the robot can climb in rises to 42 cm.
The robot can safely (without endangering its mechanisms) apply side forces up to 10–12 N (on each side) and a traction force of
10–12 N which is double its weight. The RSTAR can also climb vertically at a COF as low as 0.3. Below that level, the robot will not be
able to climb vertically unless its sprawl and FBEM mechanisms are strengthened.

11
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Fig. 13. The direction of the friction force and the effective COF as the robot expands its width while the wheels are sliding in the fore-aft direction.

Fig. 14. (A) The robot before falling. (B) The robot starts falling. (C) The robot is horizontal.

Fig. 15. (Top) The height of the COM and orientation of the robot. (Bottom) The linear speed of the COM and the angular speed of the robot.

Although a higher COF would be more advantageous and would simplify climbing, it would prevent the robot from varying its
width during the transition phases (which are needed to increase or reduce its grip). After entering the pipe in the driving mode (at a
small width), the robot was unable to increase its width beyond an arc of 90 by simply actuating its sprawl and FBEM mechanisms
◦

because of friction resistance. Therefore, there is a difference between the maximum width that the robot can achieve while driving
horizontally and the minimal arc required for climbing. The analysis showed that this difference (maximum width to minimum
requirement for climbing) always exists regardless of the side force that the robot applies to the pipe or the COF between the wheels
and the pipe. Overcoming this difference is possible at the pipe’s elbow when the robot is pitching upwards or by intentionally causing

12
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Fig. 16. The experimental setup used to measure the side forces that the RSTAR can apply to the walls.

Fig. 17. The measured normal forces applied to the walls when actuating the sprawl and FBEM Mechanisms.

Fig. 18. The normal versus sliding forces as measured by the ATI Nano 25 force sensor. The average COF was μ = 0.63.

the wheels to slide in the fore-aft direction while increasing their width. This strategy can decrease the effective COF in the side di­
rection by nearly 10- fold.
The robot can move its COM in the fore-aft and vertical directions without changing its width. We used this feature to improve the
reliability of the vertical to horizontal transition along the pitch axis by shifting the robot’s COM towards the pipe’s exit point.
Our analysis and climbing conditions were verified on an experimental robot which we used to perform multiple contact exper­
iments to evaluate the COF and determine the limits of climbing. Based on the analysis and preliminary experiments we programmed
the robot’s controller to perform a sequence of actions, such as applying pressure to the walls, shifting the COM without modifying the
width, and transitioning from the horizontal to vertical modes and holding onto the pipe. The robot climbed and transitioned between
the horizontal to vertical modes and vice-versa, both in the yaw and pitch axes, thus outperforming our human operator (see attached
video) who had to control multiple functions simultaneously.
In future work, we plan to increase the autonomy of the RSTAR so that it can perform some of these maneuvers on its own. We are
also applying a deep learning algorithm to find optimal techniques to overcome obstacles, which can be uploaded to the controller of
the robot. Another possible future direction would involve using dynamic maneuvers (although they are riskier) in which the robot
would move at a higher rate (speed and acceleration) and use its inertia to perform climbing and transitioning sequences.

13
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Fig. 19. The RSTAR (prevented from advancing by a wire) expands its width by sliding in the fore-aft direction while actuating its sprawl and FBEM
mechanisms (see attached video).

Fig. 20. The RSTAR climbing in the pipe when not in the center. (A1) and (A2) are side views, whereas (B1) and (B2) are top views (see
attached video).

Fig. 21. The robot starts by driving horizontally into the pipe, expands its width and climbs vertically in the pipe (see attached video).

14
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

Fig. 22. The RSTAR transitioning from vertical climbing to horizontal driving. (See attached video).

Fig. 23. The RSTAR transitioning along the yaw axis (see attached video).

Fig. 24. The RSTAR climbing in a rectangular canal. (see attached video).

Declaration of Competing Interest

There are no conflicts of interests.

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.mechmachtheory.2022.
104832.

15
O. Inbar and D. Zarrouk Mechanism and Machine Theory 173 (2022) 104832

References

[1] D. Zarrouk, M. Shoham, Analysis and design of one degree of freedom worm robots for locomotion on rigid and compliant terrain, ASME J. Mech. Robot. 134 (2)
(2012).
[2] J. Lim, H. Park, S. Moon, B. Kim, Pneumatic robot based on inchworm motion for small diameter pipe inspection, in: Proceedings of the IEEE International
Conference on Robotics and Biomimetics, 2007, pp. 330–335.
[3] J. Qiao, J. Shang, A. Goldenberg, Development of inchworm in-pipe robot based on self-locking mechanism, IEEE ASME Trans. Mechatron. 18 (2) (2013)
799–806.
[4] J. Lim, H. Park, J. An, Y.S. Hong, B. Kim, B.J. Yi, One pneumatic line based inchworm-like micro robot for half-inch pipe inspection, in: Proceedings of the 8th
International IFAC Symposium on Robot Control 18, 2008, pp. 315–322.
[5] Q. Xie, S. Liu, X. Ma, Design of a novel inchworm in-pipe robot based on cam-linkage mechanism, Adv. Mech. Eng. 13 (2021) 1–11, https://doi.org/10.1177/
16878140211045193.
[6] D. Zarrouk, M. Mann, N. Dgani, A. Hess, Single actuator wave-like robot (SAW): design, modeling, and experiments, Bioinspiration Biomim. 11 (4) (2016),
https://doi.org/10.1088/1748-3190/11/4/046004.
[7] A.S. Boxerbaum, A.D. Horchler, K.M. Shaw, H.J. Chiel, R.D. Quinn, Worms, waves and robots, in: Proceedings of the IEEE International Conference on Robotics
and Automation, Saint Paul, MN, 2012, pp. 3537–3538, https://doi.org/10.1109/ICRA.2012.6224805.
[8] Y. Liu, B. Lim, J.W. Lee, J. Park, T. Kim, T. Seo, Steerable dry adhesive linkage-type wall-climbing robot, Mech. Mach. Theory 153 (2020), 103987.
[9] K. Seo, S. Cho, T. Kim, H.S. Kim, J. Kim, Design and stability analysis of a novel wall-climbing robotic platform (ROPE RIDE), Mech. Mach. Theory 70 (2013)
189–208.
[10] T. Maneewarn, B. Maneechai, Design of pipe crawling gaits for a snake robot, in: Proceedings of the IEEE International Conference on Robotics and Biomimetics
(ROBIO), 2009, pp. 1–6.
[11] H. Schempf, E. Mutschler, V. Goltsberg, G. Skoptsov, A. Gavaert, G. Vradis, Explorer: untethered real-time gas main assessment robot system, in: Proceedings of
the 1st International Workshop on Advances in Service Robotics (ASER’03), 2003.
[12] A. Kakogawa, Y. Oka, S. Ma, Multi-link articulated wheeled in-pipe robot with underactuated twisting joints, in: Proceedings of the IEEE International
Conference on Mechatronics and Automation (ICMA), Changchun, 2018, pp. 942–947.
[13] I. Virgala, M. Kelemen, P. Bozek, Z. Bobovsky, M. Hagara, E. Prada, P. Oscada, M. Varga, Investigation of snake robot locomotion possibilities in a pipe,
Symmetry 12 (6, 12, 939) (2020) 1–25, https://doi.org/10.3390/sym12060939.
[14] I. Nasibullayev, O. Darintsev, D. Bogdanov, In-pipe modular robot: configuration, displacement principles, standard patterns and modeling, in: Proceedings of
the 16th International Conference on Electromechanics and Robotics, 2021, pp. 85–96.
[15] L. Hun-ok, O. Taku, Development of pipe inspection robot, in: Proceedings of the ICCAS-SICE, Fukuoka, 2009, pp. 5717–5721.
[16] S. Hirose, H. Ohno, T. Mitsui, K. Suyama, Design of in-pipe inspection vehicles for 25, 50, 150 pipes, in: Proceedings of the IEEE International Conference on
Robotics and Automation 3, 1999, pp. 2309–2314.
[17] T. Li, S. Ma, B. Li, M. Wang, Z. Li, Y. Wang, development of an in-pipe robot with differential screw angles for curved pipes and vertical straight pipes, J. Mech.
Robot. 9 (2017), 051014.
[18] Y.S. Kwon, B. Lee, I.C. Whang, W.K. Kim, B.J. Yi, A flat pipeline inspection robot with two wheel chains, in: Proceedings of the IEEE International Conference on
Robotics and Automation, 2011, pp. 5141–5146.
[19] S.A. Fjerdingen, P. Liljeback, A.A. Transeth, A snake like robot for internal inspection of complex pipe structures (PIKo, in: Proceedings of the IEEE IROS, 2009,
pp. 5665–5671.
[20] T. Nishimura, A. Kakogawa, S. Ma, Screw-drive in-pipe robot with pathway selection mechanism, in: Proceedings of the IEEE CASE, 2012, https://doi.org/
10.1109/CoASE.2012.6386388.
[21] S.G. Roh, H.R. Choi, differential-drive in-pipe robot for moving inside urban gas pipelines, IEEE Trans. Robot. 21 (1) (2005) 1–17.
[22] M. Horodinca, I. Doroftei, E. Mignon, A. Preumont, A simple architecture for in-pipe inspection robots, in: Proceedings of the International Colloquium on
Mobile and Autonomous Systems, 2002, pp. 1–4.
[23] A. Kakogawa, S. Ma, Mobility of an in-pipe robot with screw drive mechanism inside curved pipes, in: Proceedings of the IEEE International Conference on
Robotics and Biomimetics, 2010, pp. 1530–1535.
[24] A. Kakogawa, S. Ma, An in-pipe inspection module with an omnidirectional bent-pipe self-adaptation mechanism using a joint torque control, in: Proceedings of
the IEEE IROS, 2019, https://doi.org/10.1109/IROS40897.2019.8968221.
[25] D. Guo, Z. Yuan, S. Bao, J. Yuan, S. Ma, L. Du, Visualized small-size pipeline, model building using multilink-articulated wheeled in-pipe inspection robot, in:
Proceedings of the IEEE International Conference on Real-time Computing and Robotics, 2021, pp. 492–497.
[26] S. Roh, D.W. Kim, J.S. Lee, H. Moon, H.R. Choi, In-pipe robot based on selective drive mechanism, Int. J. Control Autom. Syst. 7 (1) (2009) 105–112.
[27] A. Kakogawa, Y. Oka, S. Ma, Multi-link articulated wheeled in-pipe robot with underactuated twisting joints, in: Proceedings of the IEEE International
Conference on Mechatronics and Automation, 2018, pp. 942–947.
[28] Z. Wang, Z. Deng, P. Lei, Research on technology of pipeline detection robot based on spiral propulsion, in: Proceedings of the ICIRA 2021, LNAI 13016, 2021,
pp. 67–78, https://doi.org/10.1007/978-3-030-89092-6_7.
[29] A. Kakogawa, S. Ma, Mobility of an in-pipe robot with screw drive mechanism inside curved pipes, in: Proceedings of the IEEE International Conference on
Robotics and Biomimetics, 2010, pp. 1530–1535.
[30] D. liu, J. Lu, Dynamic characteristics of two-mass inertial pipeline robot driven by noncircular gears, in: Proceedings of the ICIRA, LNAI 13016, 2021, pp. 79–90,
https://doi.org/10.1007/978-3-030-89092-6_8.
[31] P. Li, M. Tang, C. Lyu, M. Gang, X. Duan, Y. Liu, Design and analysis of a novel active screw-drive pipe robot, Adv. Mech. Eng. 10 (2018) 1–18, https://doi.org/
10.1177/1687814018801384.
[32] H. Li, R. Li, Y. Wang, Design and control of in-pipe inspection robot for pre-commissioning, in: Proceedings of the ICIRA 2021, LNAI 13016, 2021, pp. 67–78,
https://doi.org/10.1007/978-3-030-89092-6_6.
[33] D. Zarrouk, Liran Yeheskel, Rising STAR, a highly reconfigurable sprawl tuned autonomous robot, IEEE Robot. Autom. Lett. 3 (3) (2018) 1888–1895.
[34] D. Zarrouk, A. Pullin, N.J. Kohut, R.S. Fearing, STAR - sprawl tuned autonomous robot, in: Proceedings of the IEEE International Conference on Robotics and
Automation, 2013, pp. 20–25.
[35] D. Zarrouk, R.S. Fearing, Controlled in-plane locomotion of a hexapod using a single actuator, IEEE Trans. Robot. 31 (1) (2015) 157–167.
[36] A. Cohen, D. Zarrouk, The ASTAR high speed amphibious sprawl tuned robot: design and experiments, in: Proceedings of the IEEE IROS, 2020, https://doi.org/
10.1109/IROS45743.2020.9340986.
[37] N. Meiri, D. Zarrouk, Flying STAR, a hybrid crawling and flying sprawl tuned robot, in: IEEE International Conference on Robotics and Automation, 2019,
pp. 5302–5308, https://doi.org/10.1109/ICRA.2019.8794260.

16