You are on page 1of 9

A Model-Based Two-Arm Robot various empirical details, the model is sufficient to predict the

robot behavior. This conclusion further suggests that the behavior


With Dynamic Vertical and Lateral development of the robot can indeed be explored and evaluated by
using the simple climbing model in the simulation environment in
Climbing Behaviors place of extensive trial-and-error on the physical robot.
[DOI: 10.1115/1.4032777]
Wei-Hung Ko1
Department of Mechanical Engineering, 1 Introduction
National Taiwan University, The climbing robot has various potential applications for
Taipei 106, Taiwan replacing humans to perform tasks in dangerous work environ-
e-mail: b99502083@ntu.edu.tw ments. Examples include unknown-environment exploration,
building wall maintenance, military missions, and disaster relief.
Wei-Hsuan Chiang1 Various climbing robots have been developed in the last 20 yrs. In
general, the robots can be categorized into several different cate-
Department of Mechanical Engineering, gories according to their locomotive mechanisms, such as legged
National Taiwan University, type [1–14], wheeled type [15–17], tracked type [18–21], and
Taipei 106, Taiwan snake type [22]. On the other hand, the robots can be categorized
e-mail: b99502085@ntu.edu.tw into several different categories according to their attachment
mechanism as well, such as the suction type [1,11,16,19,22], mag-
Ya-Han Hsu netic type [3,17,21], mechanical-lock type [5,8–10], biomimetic
Department of Mechanical Engineering, type [6,7,15,18], electroadhesion [20], wet adhesion [14], friction
[2,4,13], and hot melt adhesion [12].
National Taiwan University, While animals such as geckos [23] and cockroaches [24] per-
Taipei 106, Taiwan form robust, agile, and dynamic climbing locomotion with a
e-mail: b99502115@ntu.edu.tw strong resistance to external disturbances, most of the climbing
robots that have been developed perform slow quasi-static loco-
Ming-Yuan Yu motion because of the limited attachment force between the robot
Department of Mechanical Engineering, and the wall. Though the availability of adequate attachment
National Taiwan University, mechanisms has already been a critical constraint for the empiri-
cal climbing robot to climb even quasi-statically on general surfa-
Taipei 106, Taiwan ces, the animals’ climbing behaviors reveal that the dynamic
e-mail: b98502039@ntu.edu.tw climbing strategy is also an important factor and should be inves-
tigated simultaneously. Previously, Goldman et al. demonstrated
Hung-Sheng Lin that a reduced-order two-arm model can represent the dynamic
Department of Mechanical Engineering, climbing locomotion of multilegged animals through extensive
National Taiwan University, experimental work on the climbing cockroaches and geckos as
Taipei 106, Taiwan well as simulations via the commercial software working model
2D [24]. Following this research, the robotic team at UPenn had
e-mail: r02522813@ntu.edu.tw scaled up and revised the model into a form that can be physically
built while maintaining the dynamic characteristics, and then, the
Pei-Chun Lin2 team built a bio-inspired climbing robot with this specific mor-
Mem. ASME phology [9]. By utilizing the designed passive claws, which pro-
Department of Mechanical Engineering, vided an adequate attachment and detachment mechanism
National Taiwan University, between the robot and the vertical carpet wall, the robot could
Taipei 106, Taiwan dynamically climb the vertical wall. In addition, the team also
found that a sprawled posture is required for stability and sug-
e-mail: peichunlin@ntu.edu.tw gested that a sprawled posture may yield a significant energetic
advantage [25]. The team studied the relation between the sprawl
posture and the wall inclination as well [10].
We report on the model-based development of a climbing robot Here, following our initial presentations at Refs. [26] and [27],
that is capable of performing dynamic vertical and lateral climb- we report on the sequential development of the dynamic two-arm
ing motions. The robot was designed based on the two-arm climbing robot. The contribution of this work lies in two aspects.
vertical-climbing model inspired by the dynamic climbing motion First, instead of using commercial simulation software, the
of cockroaches and geckos, with the extension of introducing the reduced-order two-arm dynamic model (i.e., not computer-aided
arm sprawl motion to initiate the lateral climbing motion. The design (CAD) model or kinematic model) was further simplified
quantitative formulation of the model was derived based on and redeveloped based on Lagrangian mechanics to explore the
Lagrangian mechanics, and the numerical analysis of the model relation between the individual model parameters and the result-
was conducted. The robot was then built and controlled based on ant dynamic behavior of the model. In addition, the model behav-
the analysis results of the model. The robot can perform the ior also served as the reference for robot development and
behaviors predicted by the model in which the climbing speed behavior planning. Second, in addition to the original design in
decreases when the swing magnitude increases, and the lateral which the robot had one active degree-of-freedom (DOF) per arm,
climbing motion can be initiated when the arm sprawl motion is the robot is revised to include one extra active DOF for control-
introduced. The experimental validation of the robot confirms that ling the arm sprawl angle for better maneuverability. By coordi-
though the reduced-order two-arm model is abstract and ignored nating this DOF with the other two-arm DOFs based on the model
analysis, the robot can be maneuvered to perform not only the
1
vertical-up climbing locomotion but also the tilted or horizontal
W.-H. Ko and W.-H. Chiang contributed equally to this work.
2
Corresponding author.
climbing locomotion. Because lateral locomotion has a different
Manuscript received April 15, 2015; final manuscript received February 5, 2016; attach/detach pattern between the claw and the wall, the claw was
published online March 10, 2016. Assoc. Editor: Xilun Ding. also revised to include a controllable attach/detach mechanism.

Journal of Mechanisms and Robotics Copyright V


C 2016 by ASME AUGUST 2016, Vol. 8 / 044503-1

Downloaded From: http://mechanismsrobotics.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/journals/jmroa6/935090/ on 02/11/2017 Terms of Use: http://www.asme.o


The remainder of the paper is organized as follows: Section 2 where a and a are the geometrical parameters defining the relative
introduces the two-hand model, and Sec. 3 discusses the effect of position of the left shoulder joint (SL ) and the COM. Equation (1)
the model parameters on the model’s dynamic behavior. Section 4 reveals that rL is a function of uL and hm , and r_ L is a function of u,
_
describes the mechanic design of the climbing robot. Section 5 uL , h_ m , and hm . Unlike rL , the orientation configuration of the
reports on the performance evaluation of the robot, and Sec. 6 COM with respect to the origin (hL ) is passively determined by
presents the conclusion of this paper. the model dynamics.
The kinetic energy (T) and potential energy (V) of the model
2 The Two-Hand Climbing Model can be expressed as
The reduced-order two-arm model was first proposed by Gold-   2  1
m 2
man et al. [24]. It had a rigid body and two arms with translational T¼ ðr_ L Þ2 þ rL2 h_ L þ mR2g h_ b (2)
2 2
movement, and the arm was composed of two serially connected
components: a linear actuator and a passive spring–damper sys- V ¼ mgrL sin hL (3)
tem. The actuator provided the climbing force to pull the body up,
and the passive system acted as the interface to smooth the dra- where hb ¼ hL  wL  hm and g are the body orientation and
matic force change after the engagement of the hand to the wall. gravity constants, respectively. The angle included in the arm (uL )
In this study, the two-arm model was further simplified, as and the rL vector, wL , can be expressed as
shown in Fig. 1. The new model also has two massless arms and a
rigid body with mass (m) and moment of inertia (I ¼ mR2g , where wL ¼ cos1 ððu2L þ rL2  a2 Þ=2uL rL Þ (4)
Rg is the radius of gyration), but the serially connected passive
spring–damper system on the arm was removed so the arm length The angle hL is set as the generalized coordinate, so the
(uL ) could be actively controlled by the linear actuator without Lagrange’s equation can be expressed as
involving any passive dynamics. In addition, the arm sprawl angle 
d @L @L
(hm ) was treated as an active and controllable DOF, and the two  ¼ QL (5)
arms were assumed to have the same hm to configure symmetri- dt @ h_ L @hL
cally respective to the body, as defined in Fig. 1. Thus, the model
has three active DOFs in total ðuL ; uR ; hm Þ. Several assumptions where the Lagrangian function is L ¼ T  V, and QL is a noncon-
have been made to simplify the analysis: (i) There was one arm servative force. The QL of the model includes the resultant
attached to the wall at all time. (ii) When the arm was in motion, Coulomb friction (F) and resultant viscous damping with the
its hand attached to the wall was regarded as a revolute joint fixed damping coefficient C acting on the model COM, and it can be
to the wall. Thus, when the robot climbed, its motion was similar expressed as
to a pendulum with a fixed pivot point and a variable length. (iii)
rL2 h_ L
The model was assumed to have Coulomb friction and viscous QL ¼ F qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  C rL2 h_ L (6)
damping acting on the center of mass (COM) of the model. This 2
r_ þ r 2 h_
2
L L L
energy nonconservative assumption makes the model more practi-
cal to be mapped to the physical system. (iv) When the model
Note that the energy loss along with the direction rL can be
switched the hands, the linear momentum of the model in the
actively compensated by the arm actuation, so only the energy
unconstrained direction (i.e., perpendicular to the vector formed
loss in the perpendicular direction to the rL is considered. Thus,
by the hand and the COM) remained unchanged.
the fraction multiplied by F was utilized to compute the vector
The quantitative formulation of the model was developed based
component of the friction force along the direction perpendicular
on Lagrangian mechanics. Figure 1 shows the scenario when the
to rL , and the term multiplied by C is the COM velocity along the
left arm is attached to the wall as the demonstration. The origin of
direction perpendicular to rL . Next, by substituting Eqs. (2), (3),
the Cartesian coordinate system (OL ) is set on the left hand (i.e.,
and (6) into Eq. (5), the equation of motion (EOM) of the model
the revolute joint). When the model climbs, the arm length (uL )
when the left arm of the model attaches to the wall can be
and sprawl angle (hm ) change according to the control commands,
expressed as
so the resultant distance between the COM and the origin (rL ) is a
function of these two variables and can be actively determined as  
m rL2 þ R2g €h L þ 2mrL r_ L h_ L þ mgrL cos hL
  
p 1=2  
rL ¼ a2 þ u2L  2auL cos hm þ a þ (1) rL2 h_ L
€ þ €h m  F qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ¼ mR2g w L  CrL2 h_ L (7)
_r 2L þ rL2 h_ L 2

Note that rL and wL are functions of controlled inputs and model


parameters, so the only unknown variable to be determined is hL .
Similarly, the model EOM when the right arm attaches to the wall
can be expressed as
 
m rR2 þ R2g €h R þ 2mrR r_ R h_ R þ mgrR cos hR
  rR2 h_ R
¼ mR2g w€ þ €h m  Ff qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  C rR2 h_ R (8)
R
2
2 2
r_ R þ rR h R _

As shown in Eqs. (7) and (8), the two-arm model is a two-


model system, and its continuous climbing motion requires the
model’s switching mechanisms to be clearly defined. As shown in
Fig. 2(a), when the left arm of the model is attached to the wall
(hereafter referred to as the “left-arm model”), the arm length (uL )
Fig. 1 The sketch and notations of the two-arm model (a) and decreased to pull the body up. In the meantime, the right-arm
the physical robot (b) length (uR ) increased to reconfigure its state for the next wall

044503-2 / Vol. 8, AUGUST 2016 Transactions of the ASME

Downloaded From: http://mechanismsrobotics.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/journals/jmroa6/935090/ on 02/11/2017 Terms of Use: http://www.asme.o


R
Fig. 2 The motion sequence of the model in vertical climbing (a) and lateral climbing (b) simulated using MATLABV

engagement. More specifically, the arms were assumed to have lateral climbing toward the right side as the demonstration. If the
the same motion profile but with a 180 deg phase difference. The arm sprawl angle moves to its maximum when the right arm
switching time is set to activate when the arms reach their maxi- extends to its extreme, the right hand can grab the position, which
mum/minimum length states. In a complete climbing stride with is further toward the right side in comparison with the condition
an actuation period tp , the robot is pulled up by the left arm and without the arm sprawl motion. Next, the arm sprawl angle moves
right arm in a sequence. When the model switches from the left- to its minimum when the left arm retracts to its extreme, and the
arm to the right-arm at t ¼ tp =2 (i.e., the final state of the left-arm left hand can grab the position, which is also further toward the
model and the initial state of the right-arm model), the coordinate right side. Therefore, by repeating the motion, the robot can per-
system also switches, and the configuration relations of the two form a lateral climbing motion. The described planning trajectory
coordinate systems (i.e., the relation of the two hands) can be requires the periodic motion of the arm length and the sprawl
defined by a vector h as shown in Fig. 1 motion to be synchronized and in phase where both arms reach
the extreme simultaneously.
h ¼ rL ðcos hL f i^ þ sin hL f jÞ
^  rR ðcos hR i i^ þ sin hR i jÞ
^ (9)
3 Dynamic Behavior of the Model
where the subscripts f and i represent the final and initial states,
respectively. At this point, the orientation configuration of the In addition to the model formulation described in Sec. 2, the
COM with respect to the new origin (hR i ) is a function of that of active DOFs ðuL ; uR ; hm Þ should also be clearly defined before
the old origin (hL f ) and the robot configuration, and it can be the performance analysis of the model. Because these 3DOFs in
expressed as the physical robot are controlled by the actuators, empirical con-
siderations should be incorporated to make the defined motion
hR i ¼ hL f  ð2hm þ wL þ wR Þ (10) feasible. The arms of the model ðuL ; uR Þ are assumed to stretch
and retract alternately and periodically. Though a linear motor
In addition, the COM velocity perpendicular to the rL direction may fit the motion requirement from a mechanism aspect, it is not
when the model switches remains unchanged, so the initial angu- empirically suitable for the following reasons. First, it is designed
lar speed h_ R i can be expressed as to have a high-speed and low-force motion profile, which did not
fulfill the low-speed and high-force demand for this application.
1     Second, in this application, the linear motor had to accelerate and
h_ R i ¼ r_ L sin hL f  hR i þ rL h_ L f cos hL f  hR i (11)
rR decelerate to achieve an arm reciprocating extend/retract motion.
This would cause the motor to operate in the energy inefficient
Similarly, when the model switches from the right-arm to the left- region where nontrivial joule heat generates, and it may even
arm, the initial angular speed h_ L i can be expressed as damage the motor. Thus, as in the previous design [9], a rotational
actuator with a crank-and-slider mechanism is utilized to generate
1    
h_ L i ¼ r_ R sin hR f  hL i þ rR h_ R f cos hR f  hL i (12) the reciprocating motion, as shown in Fig. 1(b). In this case, the
rL actuator continues to rotate in the same direction, so the motor
can be operated nearly constant speed. In contrast, though the arm
Finally, by importing the initial conditions (hL i , h_ L i ) or (hR i , sprawl motion of the robot is also reciprocal, its motion requires
h_ R i ) shown in Eqs. (10)–(12) into the left-arm model shown in less torque because of a lower gravity effect. Thus, a rotational
Eq. (7) or the right-arm model shown in Eq. (8), the dynamic actuator can be utilized to move reciprocally.
response of the model can be numerically simulated. In addition, The formulation of the active DOFs is described as follows.
by using Eq. (9), the dynamic responses of the model in multiple The arm sprawl angle (hm ) is parameterized as
strides can be sequentially composed. As a result, by giving
the initial conditions of the model in its first stride, the COM tra- hm ¼ Am cos ð2pt=tp þ /m Þ þ Om (13)
jectory of the model during the entire climbing motion can be
quantitatively computed. where Am is the swing amplitude, and Om and /m are two offset
In addition to the vertical-climbing motion, the sequential com- angles with different purposes. The offset angle Om determines
position of the model behavior shown in Eqs. (9)–(12) can be uti- the offset of the swing central line with respect to the body’s
lized for the lateral climbing motion by incorporating its sprawl fore–aft direction, as shown in Fig. 1(b). The offset phase /m
DOF (hm ) with the arm length DOFs (uL , uR ). Figure 2(b) shows determines the relative phase between the arm sprawl angle (hm )

Journal of Mechanisms and Robotics AUGUST 2016, Vol. 8 / 044503-3

Downloaded From: http://mechanismsrobotics.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/journals/jmroa6/935090/ on 02/11/2017 Terms of Use: http://www.asme.o


Table 1 Physical parameters common to both initial and modi-
fied robot versions

m 5 kg
Rg 0.11547 m
(a,a) (0.117 m, 52.64 deg)
ln 0.111 m
l1 0.06 m
l2 0.105 m
N 54
Ns 540
Body size 0.142 m  0.215 m

and the arm length ðuL ; uR Þ, and its nominal value is set to p.
Because both the arm DOFs and the sprawl DOF should reach the
extreme simultaneously for effective vertical and lateral climbing
motions, the arm length uL is set as

uL ¼ ln þ l1 cosð2pt=tp  ðhm  Om ÞÞ
þ ðl22  l21 sin2 ð2pt=tp  ðhm  Om ÞÞ1=2 (14)

where ln , l1 , and l2 are the geometrical constants, as shown in


Fig. 1(b). In addition, the right-arm length uR has the same motion
profile but with a 180 deg phase difference.
After the two-arm model was quantitatively formulated, the
R
dynamic motion of the model was simulated using MATLABV, and
the manner in which the parameters of the model affected its
dynamic behavior could be investigated. The values of the model
parameters during simulation are listed in Table 1. The perform-
ance and behavior of the model are mainly accessed by the climb-
ing velocity of the model, including the speed and the direction.
Figure 3 shows the climbing speed (v) of the model versus the
resultant Coulomb friction force (F) and the resultant viscous
damping with the damping coefficient (C). The model is initially Fig. 3 The climbing speed (v ) of the model versus the result-
in the vertical posture (i.e., hb ¼ 0 deg) and operates with the ant Coulomb friction force (F) and the damping coefficient
settings Om ¼ 5 deg, Am ¼ 0 deg, and tp ¼ 1 s. Several observa- (C) shown in a 3D view in (a), a side view in (b), and a front view
tions can be made: (i) When the F becomes larger, the body has in (c)
less lateral swing motion because the larger portion of the gravity
potential would be used to compensate the friction force, not to
convert it to kinetic energy to initiate the swing motion of the marker is located on the right side of the origin, the model has lat-
model. (ii) The magnitude of the lateral swing motion has the eral movement. When the maker is located at the upper right cor-
opposite behavior of the climbing speed. When the lateral swing ner, the model has both vertical and lateral movements. The figure
motion is less, the hand of the arm in the aerial phase is easier to reveals the following trends: (i) The swing amplitude (Am ) has a
pivot toward the wall at a higher position with respect to the pivot critical effect on the motion direction of the model. When the
of the other arm. (iii) Though F and C have slightly different swing amplitude increases, the model has more lateral motion. (ii)
behaviors because the viscous damping is proportional to the In contrast, the offset angle (Om ) has a critical effect on the
velocity, in general, their effect difference is insignificant to the motion speed of the model. When the offset angle increases, the
model dynamics. Therefore, in the following development of model has less climbing speed. This effect is significant especially
the model, the effect of these two forces were treated as one when the model has a low period of 1 or 1.5 s. In some cases, the
“resistant force (F & C)” and varied together to represent the model even climbs down instead of climbing up. (iii) When the
resultant energy loss condition of the model. As this parameter model has a low period of 1 or 1.5 s, the climbing speed increases
increases, the climbing speed of the model increases, and the lat- as the resistant force increases. The trend of climbing speed versus
eral motion of the model decreases. resistant force is the same, as observed in Fig. 3. (iv) When the
Figure 4 shows the dynamic behavior of the model with differ- model period is larger than 2 s, the resistant force has less effect
ent initial body orientations (hb ) spanning between 40 deg and on the model behavior. In this case, the swing motion of the model
40 deg. The offset angle Om and the period tp were also varied for is damped before the hand switching, so the climbing motion is
evaluation. The figure shows that the dynamic behaviors of the more quasi-static. In addition, though the stride length of the
model in its first step (first half of the first stride) have a large vari- model increases when the period increases, the overall climbing
ation owing to the effect of the ICs, and the variation decreases velocity decreases.
beginning from the second step (second half of the first stride).
Moreover, the behaviors of the model with different ICs quickly
converge to the same behavior before the third stride. This phe- 4 Robot Design
nomenon indicates that the natural dynamics of the model domi- The morphology of the climbing robot basically follows that of
nate its behavior, and the model has a large attraction region to the two-arm model described in Sec. 2, and the detailed mecha-
converge its behavior to it natural dynamics. nism was designed to achieve the actuation requirement of the
Figure 5 shows the COM velocity of the model (ðvx ; vy Þ) with active DOFs ðuL ; uR ; hm Þ. Figure 6(a) plots the CAD drawing of
different resistant forces (F & C), periods (tp ), swing amplitudes the robot mechanism. For the DOF of the left arm (uL ), the slider
(Am ), and offset angles (Om ). In the figure, if the marker is located is driven by the left-arm motor through a bevel gear pair and the
above the origin, the model climbs vertically. In contrast, if the crank-and-slider mechanism, which is discussed in detail in the

044503-4 / Vol. 8, AUGUST 2016 Transactions of the ASME

Downloaded From: http://mechanismsrobotics.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/journals/jmroa6/935090/ on 02/11/2017 Terms of Use: http://www.asme.o


Fig. 4 The dynamic behavior of the model with different body orientations

first paragraph of Sec. 3. The right arm has a mechanism symmet- our work was also designed to perform lateral motion, so the
rical to that of the left arm with respect to the sagittal plane of the passive claw was no longer a feasible solution. Therefore, two
robot. For the arm sprawl DOF (hm ), the rail of the slider (i.e., the attachment/detachment-active claws on each hand were adopted,
ground link of the four-bar linkage) is rotated by the sprawl motor and the claws were driven by a servo motor through a four-bar
through a shaft coupler, a worm and worm gear pair, and a pulley- linkage mechanism, as shown in Fig. 6(c), for two purposes: To
and-belt system. Note that the worm gears of the two arms are amplify the torque for attachment/detachment and to use the link-
driven by the same worm, so they have synchronized and opposite age mechanism to support the body weight, avoiding passing the
rotations, as designed in the model. load through the small servo motors. In addition, as shown in
The mapping from the left-arm motor, the right-arm motor, and Fig. 6(c), a perforated plastic sheet was used as the wall material
the sprawl motor, ð/L ; /R ; /S Þ, to the model inputs ðuL ; uR ; hm Þ for the claw to attach and detach to the wall as well as to preserve
can then be quantitatively formulated based on the mechanism the rotational freedom of the hand with respect to the wall.
details as shown in Fig. 6(a). The arm sprawl angle (hm ) is a func- Figure 7 shows the photos of the robot, and its specifications
tion of /S are listed in Table 1. The robot has a real-time embedded control
system (sbRIO-9602, National Instruments, Austin, TX). The
h_ m ¼ NS /_ S (15) left-arm, right-arm, and sprawl motors are brushed DC motors
(IG-36PGM TYPE1, Shayang, Taiwan) with encoders (HS-302-
where NS is the summation of the gear ratio of the motor gearbox, 512P-3A, Honest Sensor Corp., Taiwan), and the motors for claw
the worm and worm gear pair, and the pulley-and-belt system. motion are servomotors (HS-645MG, Hitec RCD, Poway, CA).
In contrast, the left-arm length (uL ) is a function of both /S and Hall-effect sensors (SS443A, Honeywell, Morris Plains, NJ) were
/L . The former one links to the uL by hm , as shown in Eq. (14), utilized for the initial configuration calibration of the arm with
and the latter one links to the uL in Eq. (14) by tp ¼ 2pN=/_ L . The respect to the robot’s body.
symbol N is the rotation speed ratio from the motor to the crank
of the four-bar linkage, and it is determined by the summation of
the gear ratio of the motor gearbox and the bevel gear pair. 5 Experiment Evaluation
The attachment/detachment mechanism of the hand to the wall The development of the model and the robot is tightly coupled.
is also an important issue to address. The hand of the model was On the one hand, the model was utilized to analyze and predict
assumed to have a rigid but rotatable contact to the wall, and in the dynamic behavior of the robot. On the other hand, the robot
the previous design, this was achieved by using passive claws to was also utilized to evaluate the effectiveness of the model. Most
engage the carpet wall [9]; however, the passive claw was only of the parameters (such as mass, inertia, and geometrical dimen-
functional when the robot climbed up because its attachment/ sions) of the model and robot were easy to map. The actuation
detachment mechanism was based on the unsymmetrical period tp selected for experimental evaluation was 1, 1.5, and 2 s.
“hooklike” design as shown in Fig. 6(b). In contrast, the robot in When the actuation period of the robot increased beyond 2 s, the

Journal of Mechanisms and Robotics AUGUST 2016, Vol. 8 / 044503-5

Downloaded From: http://mechanismsrobotics.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/journals/jmroa6/935090/ on 02/11/2017 Terms of Use: http://www.asme.o


Fig. 5 The COM velocity of the model with different resistant forces (F &C), periods (tp ), swing amplitudes (Am ), and offset
angles (Om )

motion of the robot gradually changed from dynamic behavior to swing motion of the robot also deteriorated the engagement qual-
quasi-static behavior. The resistant force of the model was deter- ity of the hand to the wall, and the claw often slid on the wall
mined based on the vertical-climbing behaviors of the model and before it could firmly attach to the wall. This phenomenon was
the robot. The trajectories of the model with different resistant observed when the robot climbed with the settings Om ¼ 0 (i.e.,
forces were generated and matched to the robot’s trajectory. the vertical segments of the COM trajectories), and this phenom-
Because the resistant force was designed to be applied to the enon was severe when the robot climbed with the settings Om ¼ 5
swing direction, the model with the least-squared trajectory error in which the robot could not climb up but maintained the swing at
to the robot trajectory in the lateral direction was selected as the the same position.
correct model setting, which yielded the resultant Coulomb fric- Figure 9 shows the COM trajectories of the two-arm model (a)
tion force F ¼ 2:5 (N) and the coefficient of the resultant damping and the robot (b) with three different actuation periods, tp ¼ 2,
force C ¼ 17:5. 1:5, and 1s. The three experiments were all conducted in eight
Figure 8 shows the COM trajectories of the two-arm model (a) total steps with the settings Om ¼ 5 and Am ¼ 0 . When the
and the robot (b) with different offset angles, Om ¼ 5, 0, and period decreased, the robot COM trajectories of the model and the
5deg. Three experiments were all conducted in eight total steps robot become sharper. This phenomenon resulted from the fact
with the settings T ¼ 2 and Am ¼ 0 . The robot motion trajectory that the arm motion was faster than the swing dynamics of the
was computed based on the sequential snapshots taken by a sta- model, so the former one dominated the COM trajectory, which
tionary HD camcorder (HDR-SR11, Sony) viewing the wall pulled the robot toward the engaged hand. Note that though the
where the robot climbed (i.e., “top view” of the robot). The robot zig–zag COM trajectories have similar sizes, because the actua-
showed the same motion trend as the model predicted. When the tion periods were different, the speed of the robot increased when
swing motion increased, the climbing speed decreased because the period decreased. The robot with tp ¼ 1 had a speed of about
the vertical displacement between the engaged hand positions in 0.16 m/s, which is twice as the speed of the robot with tp ¼ 2. The
consecutive swing steps decreased. The robot with Om ¼ 5 had speed of the robot was 7% faster than the model prediction of
the smallest swing motion, and its climbing speed was about 0.15 m/s.
0.087 m/s, 6% slower than the model prediction of 0.093 m/s. Figure 10 shows the COM trajectory of the robot with three
When Om ¼ 0, the swing motion of the model increased, and the different combinations of the swing amplitudes (Am ) and the
speed decreased to 0.062 m/s, 20% slower than the model predic- offset angles (Om ), Om ¼ 5; Am ¼ 10, Om ¼ 10; Am ¼ 15, and
tion of 0.078 m/s. The discrepancy between the model and the Om ¼ 15; Am ¼ 20. Unlike the results shown in Figures 8 and 9
robot increased because, empirically, we found that the large where the arm sprawl angle hm of the robot was not varied, the

044503-6 / Vol. 8, AUGUST 2016 Transactions of the ASME

Downloaded From: http://mechanismsrobotics.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/journals/jmroa6/935090/ on 02/11/2017 Terms of Use: http://www.asme.o


Fig. 7 The photo of the robot

Fig. 6 The detailed CAD drawings of the robot arm (a) and the
claw (c), and (b) the CAD drawing of the passive claw utilized in
Ref. [9]

arm sprawl angle was varied to initiate the lateral climbing


motion. Because the physical robot has a minimum achievable
sprawl angle of hm ¼ 5, when the swing amplitude (Am ) Fig. 8 The COM trajectories of the two-arm model (a) and the
increased, the offset angle Om had to increase simultaneously to robot (b) with three different offset angles (Om )
avoid the collision of the two robot arms. The robot in these three
testing combinations had the same minimum sprawl angle,
hm ¼ 5. Figure 10 reveals that when the Am increased, the lateral 6 Conclusion
climbing motion of the robot increased. When the Am increased to We report on the model-based development of a climbing robot
20 deg, the robot moved almost horizontally. The figure also that is capable of performing vertical and lateral climbing
shows that the COM trajectories of the model had a greater lateral motions. The robot was designed based on the two-arm vertical-
climbing motion behavior than that of the robot (i.e., had a larger climbing template inspired by the dynamic climbing motion of
lateral/vertical displacement ratio). The difference mainly results cockroaches and geckos with the extension of introducing the arm
from the mechanism backlash of the robot, which yields about sprawl motion to initiate the lateral climbing motion. The quanti-
65 deg error of the sprawl angle, and the backlash decreases the tative formulation of the model was derived based on Lagrangian
effective sprawl angle of the robot in this specific lateral climbing mechanics, and the numerical analysis of the model reveals sev-
motion. The experimental results confirm that the robot could eral facts. (i) The two-arm model has a self-stabilizing mechanism
indeed perform the lateral motion by using the swing motion of in which the model with different initial conditions (i.e., body ori-
the arms, which also matched the prediction of the model, as entation, hb ) converges to the same dynamic behavior in one or
shown in Fig. 5. two strides. (ii) The resultant Coulomb friction force (F) and the

Journal of Mechanisms and Robotics AUGUST 2016, Vol. 8 / 044503-7

Downloaded From: http://mechanismsrobotics.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/journals/jmroa6/935090/ on 02/11/2017 Terms of Use: http://www.asme.o


eliminates the necessity of extensive trial-and-error on the physi-
cal robot.
The experimental results also indicate that better robot perform-
ance and the precise mapping between the model and the robot
would require improvements in both the model and the robot. For
the model, the arm mass can be included, and a better energy loss
model can be investigated so that the model dynamics match the
empirical robot behaviors better. For the robot, the attachment/
detachment failure is the major cause for behavior discrepancies
between the model and the robot, and a better hand mechanism
should be investigated. Perhaps, the arm motion of the robot can
also be redesigned with the goal of reducing the maximum inter-
action forces between the hand and the wall. A model-based feed-
back control strategy capable of failure recovery would definitely
improve the robot’s climbing performance, and in this case, the
robot must have a body status sensory system and extra DOFs to
jointly generate this real-time hand reattachment motion. This is
not a trivial task and is currently under investigation.

Acknowledgment
The authors would like to thank the undergraduate students of
the Bio-inspired Robotics Laboratory for their help in building the
prototype of the robot.
This work was supported by the National Science Council
(NSC), Taiwan, under Contract Nos. NSC 101-2815-C-002-103-
E, NSC 102-2815-C-002-074-E, and NSC 102-2221-E-002-214-
Fig. 9 The COM trajectories of the two-arm model (a) and the MY3.
robot (b) with three different actuation periods (tp )
Nomenclature
Am ¼ swing amplitude of the sprawl DOF
C ¼ coefficient of the resultant damping force acting on
the model COM
F ¼ resultant Coulomb friction acting on the model
COM
g ¼ gravity constant
h ¼ relative position vector between the two hands at
hand switching
I ¼ moment of inertia of the two-arm model, and
I ¼ mR2g , where Rg is the radius of gyration
l1 , l2 , ln ¼ geometrical parameters of the arm mechanism
m ¼ mass of the two-arm model
Fig. 10 The COM trajectories of the two-arm model (a) and the
N ¼ gear ratio between the motor rotation and the arm
robot (b) with three different combinations of the offset angles sprawl motion
(Om ) and the swing amplitudes (Am ) NS ¼ gear ratio between the motor rotation and the arm
crank rotation
Om ¼ offset angle of the sprawl DOF
viscous damping (C) acting on the model COM have very similar ri ¼ distance between the hand and the COM, and
effects on the model dynamics. When the values of these two i ¼ R or L indicates right or left
terms increase, the model has less swing motion and faster climb- t ¼ time
ing speed. (iii) The swing amplitude (Am ) has a critical effect on T, V ¼ kinetic energy and potential energy of the two-arm
the motion direction of the model. When the swing amplitude model
increases, the model has a more lateral climbing motion. (iv) The tp ¼ actuation period of the arm and sprawl DOFs
offset angle (Om ) has a critical effect on the motion speed of the ui ¼ arm length of the two-arm model, and i ¼ R or L
model. When the offset angle increases, the model has a slower indicates right or left
climbing speed. hb ¼ body orientation of the two-arm model with respect
The robot was then built and controlled based on the findings of to the inertia frame
the model dynamics. Three motors were installed to provide arm hi ¼ orientation configuration of the COM with respect
stretch/retract and sprawl motions. The experimental results of the to the hand, and i ¼ R or L indicates right or left
robot reveal that the climbing speed of the robot decreases when hm ¼ arm sprawl angle
the swing magnitude increases, which was the same trend in the wi ¼ angle included by the ui and the ri vectors
model. In addition, the large swing of the robot may also result in v ¼ climbing speed, including lateral and vertical
the slip behavior when the hands try to engage to the wall. When components ðvx ; vy Þ
the arm sprawl motion was introduced, the robot could indeed per- /m ¼ offset angle between the sprawl DOF and the arm
form a lateral climbing motion, as predicted by the model. The length DOF
qualitative behaviors between the model and the robot match quite /L ; /R ; /S ¼ rotation of the left-arm motor, the right-arm motor,
well, and this further suggests that the behavior development of and the sprawl motor
the robot can indeed be explored and evaluated using a simple ða; aÞ ¼ geometrical parameters defining the relative position
reduced-order model in the simulation environment, which of the shoulder joint (Si ) and the COM

044503-8 / Vol. 8, AUGUST 2016 Transactions of the ASME

Downloaded From: http://mechanismsrobotics.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/journals/jmroa6/935090/ on 02/11/2017 Terms of Use: http://www.asme.o


[15] Murphy, M. P., Tso, W., Tanzini, M., and Sitti, M., “Waalbot: An Agile Small-
References Scale Wall Climbing Robot Utilizing Pressure Sensitive Adhesives,” IEEE/RSJ
[1] Luk, B. L., Cooke, D. S., Galt, S., Collie, A. A., and Chen, S., 2005, “Intelligent International Conference on Intelligent Robots and Systems (IROS), Beijing,
Legged Climbing Service Robot for Remote Maintenance Applications in Oct. 9-15, pp. 3411–3416.
Hazardous Environments,” Rob. Auton. Syst., 53(2), pp. 142–152. [16] Koo, I. M., Trong, T. D., Lee, Y. H., Moon, H., Koo, J., Park, S. K., and Choi,
[2] Wei, T. E., Quinn, R. D., and Ritzmann, R. E., 2005, “A CLAWAR That Bene- H. R., 2013, “Development of Wall Climbing Robot System by Using Impeller
fits From Abstracted Cockroach Locomotion Principles,” Climbing and Walk- Type Adhesion Mechanism,” J. Intell. Rob. Syst., 72(1), pp. 57–72.
ing Robots, Springer, Berlin, pp. 849–857. [17] Tavakoli, M., Viegas, C., Marques, L., Pires, J. N., and de Almeida, A. T.,
[3] Krosuri, S. P., and Minor, M. A., 2005, “Design, Modeling, Control, and Evalu- 2013, “OmniClimbers: Omni-Directional Magnetic Wheeled Climbing Robots
ation of a Hybrid Hip Joint Miniature Climbing Robot,” Int. J. Rob. Res., for Inspection of Ferromagnetic Structures,” Rob. Auton. Syst., 61(9),
24(12), pp. 1033–1053. pp. 997–1007.
[4] Nabulsi, S., Montes, H., and Armada, M., “Roboclimber: Control System [18] Greuter, M., Shah, G., Caprari, G., T^ache, F., Siegwart, R., and Sitti, M.,
Architecture,” Climbing and Walking Robots, Springer, Berlin, pp. 943–952. “Toward Micro Wall-Climbing Robots Using Biomimetic Fibrillar Adhesives,”
[5] Bretl, T., 2006, “Motion Planning of Multi-Limbed Robots Subject to Equilib- 3rd International Symposium on Autonomous Minirobots for Research and
rium Constraints: The Free-Climbing Robot Problem,” Int. J. Rob. Res., 25(4), Edutainment (AMiRE 2005), Fukui, Japan, Sept. 20-22, Springer, Berlin, pp.
pp. 317–342. 39–46.
[6] Kim, S., Spenko, M., Trujillo, S., Heyneman, B., Santos, D., and Cutkosky, [19] Kim, H., Kim, D., Yang, H., Lee, K., Seo, K., Chang, D., and Kim, J., 2008,
M. R., 2008, “Smooth Vertical Surface Climbing With Directional Adhesion,” “Development of a Wall-Climbing Robot Using a Tracked Wheel Mechanism,”
IEEE Trans. Rob., 24(1), pp. 65–74. J. Mech. Sci. Technol., 22(8), pp. 1490–1498.
[7] Spenko, M., Haynes, G. C., Saunders, J., Cutkosky, M. R., Rizzi, A. A., Full, [20] Wang, H. Q., Yamamoto, A., and Higuchi, T., 2014, “A Crawler Climbing
R. J., and Koditschek, D. E., 2008, “Biologically Inspired Climbing With a Robot Integrating Electroadhesion and Electrostatic Actuation,” Int. J. Adv.
Hexapedal Robot,” J. Field Rob., 25(4–5), pp. 223–242. Rob. Syst., 11, p. 191.
[8] Provancher, W. R., Jensen-Segal, S. I., and Fehlberg, M. A., 2011, “ROCR: An [21] Nam, S., Oh, J., Lee, G., Kim, J., and Seo, T., 2014, “Dynamic Analysis During
Energy-Efficient Dynamic Wall-Climbing Robot,” IEEE/ASME Trans. Mecha- Internal Transition of a Compliant Multi-Body Climbing Robot With Magnetic
tron., 16(5), pp. 897–906. Adhesion,” J. Mech. Sci. Technol., 28(12), pp. 5175–5187.
[9] Lynch, G. A., Clark, J. E., Lin, P. C., and Koditschek, D. E., 2012, “A Bioins- [22] Zhang, H., Gonzalez-Gomez, J., Chen, S., Wang, W., Liu, R., Li, D., and
pired Dynamical Vertical Climbing Robot,” Int. J. Rob. Res., 31(8), Zhang, J., “A Novel Modular Climbing Caterpillar Using Low-Frequency
pp. 974–996. Vibrating Passive Suckers,” IEEE/ASME International Conference on
[10] Dickson, J., and Clark, J. E., 2012, “The Effect of Sprawl Angle and Wall Incli- Advanced Intelligent Mechatronics (AIM), Zurich, Switzerland, Sept. 4-7.
nation on a Bipedal Dynamic Climbing Platform,” International Conference on [23] Autumn, K., Hsieh, S., Dudek, D., Chen, J., Chitaphan, C., and Full, R., 2006,
Climbing and Walking Robots and the Support Technologies for Mobile “Dynamics of Geckos Running Vertically,” J. Exp. Biol., 209(2), pp. 260–272.
Machines (CLAWAR), Baltimore, MD, July 23-26. [24] Goldman, D. I., Chen, T. S., Dudek, D. M., and Full, R. J., 2006, “Dynamics of
[11] Guan, Y. S., Zhu, H. F., Wu, W. Q., Zhou, X. F., Jiang, L., Cai, C. W., Zhang, Rapid Vertical Climbing in Cockroaches Reveals a Template,” J. Exp. Biol.,
L. M., and Zhang, H., 2013, “A Modular Biped Wall-Climbing Robot With 209(15), pp. 2990–3000.
High Mobility and Manipulating Function,” IEEE/ASME Trans. Mechatron., [25] Lynch, G. A., Rome, L., and Koditschek, D. E., 2011, “Sprawl Angle in Simpli-
18(6), pp. 1787–1798. fied Models of Vertical Climbing: Implications for Robots and Roaches,” Appl.
[12] Osswald, M., and Iida, F., 2013, “Design and Control of a Climbing Robot Bionics Biomech., 8(3), pp. 441–452.
Based on Hot Melt Adhesion,” Rob. Auton. Syst., 61(6), pp. 616–625. [26] Ko, W. H., Chiang, W. H., Hsu, Y. H., Fang, I. L., Lin, H. S., Yu, M. Y., and
[13] Degani, A., Long, A. W., Feng, S. Y., Brown, H. B., Gregg, R. D., Choset, H., Lin, P. C., 2014, “A Dynamic Vertical Climbing Robot,” International Confer-
Mason, M. T., and Lynch, K. M., 2014, “Design and Open-Loop Control of the ence on Advanced Robotics and Intelligent Systems (ARIS), Taipei, Taiwan,
ParkourBot: A Dynamic Climbing Robot,” IEEE Trans. Rob., 30(3), June 6-8, p. 248.
pp. 705–718. [27] Yu, M. Y., Lin, H. S., Ko, W. H., Chiang, W. H., Fang, I. L., Hsu, Y. H., and
[14] He, B., Wang, Z. P., Li, M. H., Wang, K., Shen, R. J., and Hu, S. Q., 2014, Lin, P. C., 2014, “Analysis of a Dynamic Climbing Two-Arm Model,” Interna-
“Wet Adhesion Inspired Bionic Climbing Robot,” IEEE/ASME Trans. Mecha- tional Conference on Advanced Robotics and Intelligent Systems (ARIS),
tron., 19(1), pp. 312–320. Taipei, Taiwan, June 6–8, p. 250.

Journal of Mechanisms and Robotics AUGUST 2016, Vol. 8 / 044503-9

Downloaded From: http://mechanismsrobotics.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/journals/jmroa6/935090/ on 02/11/2017 Terms of Use: http://www.asme.o

You might also like