Você está na página 1de 11

832 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 19, NO.

4, JULY 2011

Position and Compliance Control of a Pneumatic


Muscle Actuated Manipulator for Enhanced Safety
Tae-Yong Choi, Student Member, IEEE, Byoung-Suk Choi, Student Member, IEEE, and Kap-Ho Seo, Member, IEEE

Abstract—The safety of humans who work with robots is an im- lator. Hirzinger et al. [5] developed a lightweight manipulator
portant issue. Many studies have addressed related methods, but DLR to decrease inertia. However, these attempts inevitably em-
fundamental limits to meet safety requirements have been encoun- ployed highly complex mechanisms and controllers with var-
tered owing to the absence of compliance in robot actuators. Pneu-
matic muscle is considered to be a basic actuator and offers the ious sensors due to the absence of a compliance property in the
advantage of intrinsic elasticity to achieve joint compliance. In this actuators. This is related to the use of motors as basic actuators.
study, joint compliance actuated by pneumatic muscle is actively There is a limit to the achievement of joint compliance when
utilized to enhance human safety during collisions. To this end, the using motors, which must use gears due to the low torque of
authors present a novel approach to control compliance and asso- these components [6].
ciated positions independently with no cross-performance effects
using pneumatic muscles. The proposed method is verified by sim- To overcome the drawbacks of conventional approaches,
ulation and experiments using a physical robot. pneumatic muscle (PM) has been investigated as an attractive
actuator that provides safe characteristics. PM offers a high
Index Terms—Human robot interaction, joint compliance, pneu-
matic muscle, safety. ratio of torque with respect to its weight and size, making it
possible to actuate machines without resorting to gears. Among
its various aspects, the intrinsic elasticity of PM is its most
I. INTRODUCTION attractive property with respect to achieving compliance in
various applications. For walking robots, the elasticity of PM
is utilized to allow natural walking [7]–[9]. PM has also been
V ARIOUS service and industrial robots have been devel-
oped to help people accomplish a range of tasks. Such
robots must interact closely with humans. Because of this close
applied to a robot arm on the basis of its torque performance
[10]. In the case of an exoskeleton-type wearable design, PM
interaction, the traditional metrics of robot performance must is used for the shoulder blades in humanoid robots [11]. Its
be augmented by an attention to safety in their design. Accord- compliance property is utilized in the exoskeleton-type wear-
ingly, a considerable amount of research related to safe robots able design of a recently proposed therapy machine [12]. Zhu
has been conducted. In order to avoid collisions with obstacles, et al. [13] used PM for the linear actuators in their parallel
Graham et al. [1] and Karlsson et al. [2] used a sensor fusion manipulator. Additionally, several studies involving the use of
method to detect high-risk obstacles in working spaces. Ohashi PM have been conducted to address safety issues. In [14] and
et al. [3] used the arms of a humanoid robot to avoid collisions [15], Bicchi et al. devised a soft arm with adjustable stiffness
in a manner similar to human motion. However, the most se- functions and examined its safety properties. Even with their
rious injuries or hazards result from collisions that occur after feasible idea of variable stiffness actuation (VSA), there was no
collision avoidance methods fail. specific consideration of the positioning problem. Positioning
Improvements to reduce injury in collisions can be realized is challenging problem for manipulators with PM owing to its
through proper mechanical design. This paper also addresses a sensitivity to external elements. Tonietti et al. [16] introduced
method to reduce injuries that result from collisions. Regarding an adaptive control method related to Bicchi’s [14] work. Clapa
this problem, Zinn et al. [4] proposed a new actuation mech- et al. [17] introduced an equilibrium point control method
anism in which the actuator is divided into low- and high-fre- using a 2-degrees-of-freedom (DOF) manipulator with PM.
quency terms to effectively reduce the stiffness of the manipu- However, the PI controller used in their work is not suitable
for PM in which the parameters can vary. Moreover, they did
not consider a stiffness control method. Van Damme et al.
Manuscript received September 28, 2009; revised January 11, 2010; accepted [18] used the proxy-based SMC (PBSMC) of Kikuuwe et al.
April 02, 2010. Manuscript received in final form June 02, 2010. Date of publi-
cation August 05, 2010; date of current version June 17, 2011. Recommended [19] to increase the safety of their manipulator, which was
by Associate Editor G. E. Stewart. actuated by pleated pneumatic artificial muscle (PPAM) with
T.-Y. Choi was with the Department of Electrical Engineering, Korea
Advanced Institute of Science and Technology, Daejeon 305-701, Korea. He
notable positioning performances. However, that solution has
is now with Samsung Electronics, Suwon 443-742, Korea (e-mail: taeyong- the significant disadvantage of low tracking performance for a
choi@kaist.ac.kr). large initial error owing to its delayed positioning.
B.-S. Choi is with the Department of Electrical Engineering, Korea Advanced
Institute of Science and Technology (KAIST), Daejeon 305-701, Korea (e-mail:
Due to the considerable number of challenges involved in
bs_choi@kaist.ac.kr). such work, few studies have considered both safety and po-
K.-H. Seo is with the Pohang Institute of Intelligent Robotics (PIRO), Pohang sitioning performance [17], [18]. Thus, in this study, a novel
790-784, Korea (e-mail: neoworld@kaist.ac.kr). method to enhance safety using PM is introduced. Here, an in-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. dependent joint compliance controller (IJCC) is added to a po-
Digital Object Identifier 10.1109/TCST.2010.2052362 sition controller for the purpose of controlling the position and
1063-6536/$26.00 © 2010 IEEE
CHOI et al.: POSITION AND COMPLIANCE CONTROL OF A PNEUMATIC MUSCLE ACTUATED MANIPULATOR 833

compliance of the manipulator simultaneously. The added IJCC TABLE I


provides enhanced safety in collisions through online control of DETERMINED PM’S PARAMETERS FROM EXPERIMENTS
the joint compliance. To enable control of joint compliance, a
VSA mechanism is applied. The total controller is composed of
the IJCC and the position controller. Compliance from the IJCC
has no direct effect on the position of the manipulator, but the
vibration caused by joint compliance results in position errors.
To suppress vibration, Tzafetas et al. [20] and Park et al. [21]
used adaptive impedance control and a robust control approach.
Katsura et al. [22] proposed a method to reduce vibration in
collisions in which the environment of the robot is taken into and a nonlinear contraction, as in
account numerically. However, suppressing vibration upon PM
actuation is nearly impossible in practical terms owing to the nu- (1)
merous unpredictable perturbations of position control, such as
variation and errors in the PM model parameters. In particular, inflation
the PM characteristics can be affected by temperature, humidity, deflation
and operational stress. In conventional studies related to PM, a
(2)
sliding mode controller (SMC) has been considered to be the
most suitable position controller with respect to minimizing the where means the amount of PM contraction, with
effort related to complex error dynamics [18], [23], and [24]. corresponding to the fully deflated or relaxed PM. means
In this work, a boundary layer augmented SMC (BASMC) pressure. is a hanging mass. are spring
is used for position control with mitigated chattering while constant, damping constant, and elemental contraction force by
securing reasonable positioning performance, even with large PM without spring and damping characteristic, and modeled as
joint compliance from an IJCC. Finally, the safety of both (2). The coefficient of friction B depends on whether the PM
humans and robots during collisions is advanced by increased is being inflated or deflated. By rearranging (1), an equation of
joint compliance. The advantages of the proposed method are motion for a mass is given like
as follows.
1) Online control of the joint compliance is assured indepen- (3)
dent of the position control.
2) An easy method to meet the safety requirements is offered where is a given force to a mass . or is the
without resorting to complex algorithms or mechanisms. acceleration term of the mass .
3) Safety is enhanced without a loss of positioning perfor- Here, the form of (3) is used as the base model. Most re-
mance. searchers agree on PM’s fundamental elements, such as the
4) No use of expensive force/torque feedback sensors is force, spring and damping terms. The process to determine pa-
needed. rameters of PM model such as ,
This paper is arranged as follows. Section II derives the and is necessary before manipulator modeling. Each value
dynamics of a two-link manipulator based on a numerical PM of PM model parameters was determined as in Table I by
model. In addition, the joint compliance characteristic is studied. following the procedure of [28] and [29].
Section III describes the control method for a robot actuated by
PM. A BASMC is used for positioning with an IJCC. Section IV B. Modeling of a Joint
discusses a simulation of position control as well as collisions The antagonistic configuration of PM will now be considered.
with obstacles for the two-link manipulator. Section V gives the PM can only contract, so if one side of a pair of PMs is con-
experimental results from a case similar to that considered in tracted, it has no way to return unless the other side of the pair
the simulation. Experiments demonstrating user safety during contracts to stretch it. This characteristic requires a pair of PMs
collisions are also presented. Section VI discusses the limit of for one degree of freedom, as shown in Fig. 1. Final actuator
the proposed idea, and Section VII concludes this work. torque from a PM pair is the difference between two PMs if the
torque of each muscle is defined as in (5). Mathematically, the
II. SYSTEM MODELING relation is described as (6). Reference [23] has used the model
from [28], and [23] was referred to here to get the final joint
A. Numerical Modeling of PM torque equation

A PM’s mathematical model is necessary to adapt it to use (4)


in a robot manipulator. The PM’s nonlinearity has been a chal-
lenging problem and some advanced models such as [25]–[28] where is from to with clockwise positive direc-
have been suggested. Among them, Reynolds et al. [28] is an tion
easy model to apply to robots with PMs. The dynamic behavior
of a PM hanging vertically and actuating a mass has been mod-
eled as a combination of nonlinear friction, a nonlinear spring (5)
834 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 19, NO. 4, JULY 2011

Fig. 1. Basic configuration of robot arm with PMs. (a) and (b) are PMs of the Fig. 2. Two-link manipulator schematic. (a) and (c) are PMs of bicep. (b) and
bicep and tricep. (d) are PMs of tricep. A pair of (a) and (b) actuate a joint of J1 and a pair of (c)
and (d) actuate a joint of J2.

where and mean the torque of bicep and tricep muscle. where and are described in
The parameters of , and is from (2). and are the
contraction length of bicep and tricep. is the radius of joint

(10)

(6)
C. Compliance Property of a Joint
Until now a link depends on two input of and . A link The key idea in modifying joint compliance is to adjust the
is actuated by the torque difference of two muscles, so only the nominal pressure of the bicep and tricep upon PM actuation.
pressure difference can be the unique input. Each pressure input A PM’s nominal pressure dominates the joint stiffness prop-
can be set like (7) to match only one input to a link erty without affecting the joint angle at a constant . The
torque of the joint’s spring force terms is derived as in (11) by
extracting the spring terms from (10)
(7)
(8)

where and are nominal pressures. is the max- (11)


imum pressure that compressor can generate in real. Then,
is the pressure difference between two muscles where is the spring torque at a joint. and are spring
at given time, and only become the controllable unique torque from bicep and tricep. is the stiffness value.
input. is the nominal pressure difference between bicep By applying (4) and (8)–(11), spring torque equation is de-
and tricep. is derived by setting equal to at given rived like
nominal position. Final torque from PM pairs is described like
(9) from (6) to (7)

(9) (12)
CHOI et al.: POSITION AND COMPLIANCE CONTROL OF A PNEUMATIC MUSCLE ACTUATED MANIPULATOR 835

Fig. 4. Whole manipulator controller. (a) Joint compliance controller(IJCC).


(b) Position controller (BASMC).

Fig. 3. Proposed system’s safety improvement strategy.


damage that occurs during the collision or during the collision
Finally the joint compliance of a joint is derived as in recognition period.
Post-collision action is beyond the scope of this work. De-
termining appropriate post-collision action is another complex
and considerable challenge that must be met in order to enhance
(13) robot safety. However, this study uses the basic control-stop ac-
tion as a post-collision action. The measure to estimate a colli-
D. Dynamics of Two-Link Manipulator With PMS
sion is as follows:
Two-link manipulator’s dynamics is a general approach and
is well described in many textbook as a form of Collision detected: (18)

(14) where is the joint angle error threshold and is the joint
velocity error threshold due to the collision. This condition is
where is a robot inertia matrix, is a matrix of Coriolis and continuously inspected once the transient condition is satisfied
centrifugal effects and is a matrix of gravity loading. is a after the initial movement. By doing this, manipulator can pre-
torque. vent faulty operation owing to large initial joint errors. The tran-
We can derive the dynamics equation of (15) which has the sient condition is set as follows:
input of pressure and the output of joint acceleration by re-
placing torque of (14) by (9)
Transient condition: and
(15)
(19)
where where is the sampling number. and are user param-
eters.
(16) Before collision recognition, the manipulator is controlled by
the proposed method to enhance safety in the event of sudden
(17) collisions. The complete proposed manipulator controller con-
sists of a position controller and a compliance controller, as
For the above equations, and are for each link1 shown in Fig. 4. Joint compliance is controlled independently
and link2. and are for each link1 and link2. and of the position controller with no effect on the position control.
are for each link1 and link2. If we set the nominal pressure
as the and , then and matrix is directly defined by B. Position Control
current joint value and joint velocity. A focus on achieving manipulator elasticity inevitably results
in low joint stiffness, which induces low precision in the end-ef-
III. WHOLE MANIPULATOR CONTROL fector due to vibration during movement. Moreover, the and
matrix of (15) are imprecise terms, as they include PM model
A. Control Strategy parameters with unavoidable estimation errors. The model pa-
Supposing a collision between an object and a robot, the main rameters of PM are very changeable terms. They depend on fac-
damage that occurs during the recognition period is presented in tors such as the temperature, humidity, and operational stress
Fig. 3. Most robot will stop control or take some post-collision of the PM. SMC may be the best solution in this condition. It
actions once the collision situation is recognized. However, the guarantees a bounded error of the trajectory tracking problem,
collision recognition process expends some time, even if that though it cannot assure optimal tracking with almost zero error
time is less than one second. Objects can be damaged or hu- [30]. There have been many studies related to SMC, such as
mans injured during this very short time, and the robot cannot [19], [23] and [31]. Among them, Lilly et al. [23] considered a
avoid this recognition delay. Hence, the most important factor specific PM model at a BASMC and is referred to here. Stability
to enhance robot safety in the event of a collision is to reduce the proofing for the position controller can also be seen in [23]. A
836 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 19, NO. 4, JULY 2011

BASMC is a type of SMC that can alleviate chattering by re- TABLE II


placing the discontinuous ‘signum’ function with a continuous TWO-LINK MANIPULATOR KINEMATIC INFORMATION
“saturation” function in the feedback term.
For each joint , assume estimate of from (15)
for known parameter such that

(20)

Assume the control gain is also unknown but upper and


lower bound is known like (21) joint compliance are directly determined from (29) without
the compliance sensor feedback
(21)
(29)
and define
(30)
(22)
such that and ,
where and are determined as (31) by applying
Let the estimation of as
(30) to (28). are at and are at
. are limited by (30) from rear-
(23)
ranged (7) for both bicep and tricep
and consider the sliding surface as with

(24)
(31)
where are the positive real numbers set by user.
In conclusion we can set the control law or of (15) as 2) Compliance Control for the Safety Characteristic: Here,
we aim to maximize the joint compliance for the safety purpose,
so the joint compliance control law are determined as
(25)
Safety constraint: (32)

In that case, the joint compliance are estimated as and


the final pressure inputs to each side are determined as (33) by
(26) (7) and (32)

where is the boundary layer thickness of the sliding surface


for each joint and the is usually defined like (33)

if This is the special one-side actuation control strategy for safety


(27)
if constraint. In case for the need of high stiffness, can be
selected as a more higher value within operational range.
C. Compliance Control
IV. SIMULATION STUDIES
1) Independent Joint Compliance Control: The proposed
controller structure in Fig. 4 make it possible to control the The simulator of two-link manipulator actuated by PM
manipulator’s joint compliance independently of the position was implemented. The manipulator was modeled as shown
control. It is assumed that typically by in Fig. 2, representing a newly designed reverse actuating
setting the nominal position of the manipulator to a zero angle. architecture that reduces the manipulator size in the actual
Joint compliance is then determined via (28) from (13) for each implementation. The kinematic parameters of the manipulator
joint are given in Table II. Link1 and Link2 are the length of the
each link. LinkC1 and LinkC2 denote the length from the joint
to the center of the mass point of each link. The Simulator is
(28) implemented using . The open dynamics engine (ODE)
and OpenGL wrapped by were used for the dynamics
At a moment, the joint compliance depend on only , calculation and 3-D display. An Intel Core2Duo 2 GHz system
because are from joint sensors and are determined by the with 2 GB of RAM was used in this simulation. The sampling
position controller. Then, the control laws for the desired time in all of the simulation was set to 1/60 s. A randomized
CHOI et al.: POSITION AND COMPLIANCE CONTROL OF A PNEUMATIC MUSCLE ACTUATED MANIPULATOR 837

maximum parameter error of 20% for the A and G matrix of


(15) was set during the operation of the manipulator to represent
unpredictable modeling errors

Joint
Joint (34)

In order to show that the proposed method meets the safety re-
quirements, collisions with obstacles were simulated. Initially,
the manipulator was commanded to follow the simple desired
trajectory of (34). A rigid beam was then suddenly placed in a
temporary position with a constraint that the trajectory of the
manipulator must meet the beam, as shown in Fig. 6. Here, the
properties of the beam, including its position, material, and size, Fig. 5. Overview of PBSMC: there are two virtual connections, the virtual cou-
have no special meaning. The manipulator collided with the pling and the proxy, between the end effector and the object.
beam during the simulation and the collision impulse was trans-
ferred to the beam. The measured impact impulses were used as
safety criteria. The nature of the impact impulse is described by
(35) and is the sum of forces transferred to the object

(35)

where means the impulse during collision and means force


delivered to the object. is the collision moment and is the
end of the collision by stopping control of the manipulator.
Different conventional safety criteria exist, including those of
HIC, GSI, and TTI [15], [32]. These indices are based on the ve-
locity or acceleration of a moving object. GSI uses the acceler-
ation of the object and HIC uses its velocity properties. In other
words, they depend on the status and properties of the object. Of
Fig. 6. Simulator with obstacle on the way of manipulator’s moving with
course, a heavy person will be safer than a lighter person when counter clockwise direction.
subjected to the same impact impulses. However, it is also true
that a lower impact impulse from the manipulator decreases the shows increased errors after the moment of collision. The joint
danger for humans. The present study focuses on such manip- torque values were also increased by the controller to reduce
ulator impacts. A robot is deemed to be safer after the impact tracking error, as shown in Fig. 8. Fig. 9 shows an increase in
impulse it can deliver to a static object is decreased. Whatever compliance at the safety constraint when using the IJCC com-
the static object is, the lower the impact impulse from the ma- pared to the normal operational configuration( 225 kPa)
nipulator is, the higher the safety index is. Thus, measuring im- and PBSMC. The collision situation is ended when the manip-
pact impulse serves as an immediate method for determining the ulator stops after recognizing a collision at 9.13 s, as indicated
safety of a robot. by the vertical line labeled as “Detection” in Fig. 7. Impulses
The nominal pressure of static 225 kPa common to all joints during the collision recognition period are used to represent the
was used as a comparison condition. The maximum pressure of impact impulses during collisions. Impact impulses are calcu-
450 kPa was allowed for both the bicep and tricep PMs, and lated from the joint torque differences before and after the colli-
a wide operational range is acquired when denotes the sions. Joint torque values are converted to forces against an ob-
mid-point of the maximum pressure. This is the conventional stacle by the kinematics of the manipulator. The final impulse
method that is used to control PMs regardless of the joint com- is the sum of horizontal forces from joint1 and joint2 after as-
pliance control. In addition, the PBSMC was compared to the suming the horizontal movements of a human. These impulses
proposed method. The core idea of the PBSMC is represented were derived as 6.17 Ns at the low compliance of ( 225
in Fig. 5. A PBSMC is type of SMC that increases joint compli- kPa), 5.02 Ns at PBSMC and 4.92 Ns at the safety constraint.
ance. PBSMC assumes an imaginary object, known as a proxy, The PBSMC also showed feasible safety performance. How-
connected to the end effector of the robot by means of virtual ever, the PBSMC offset the tracking performance. PBSMC al-
spring-coupling. Essentially, a conventional SMC controls the ways had joint errors in the tracking problem owing to the in-
position of the proxy and the real end effector is forced to move tentionally delayed control due to the virtual spring-coupling.
to the proxy by the virtual spring-coupling. Virtual spring-cou- The long convergence time due to large initial position errors is
pling increases the compliance of the joints. a particularly critical issue, as can be seen in the trajectory of
The moment after the collision with the beam at 8.95 s, as in- joint2, which is depicted in Fig. 7.
dicated by the vertical line labeled as “Collision” in Fig. 7, is the In other words, impact impulses were decreased by roughly
focus for finding the impulse transferred to the obstacle. Fig. 7 22% with the proposed safety constraint when using the IJCC.
838 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 19, NO. 4, JULY 2011

Fig. 9. Compliance variations at collision simulations. The compliance at the


Fig. 7. Position variations of the collision simulations. The proposed method safety constraint shows more increased joint compliance for all working range
shows good tracking performance. However, it takes long time for the PBSMC contrary to the others.
to converge owing to the large initial position errors at joint2.

Fig. 10. Control system configuration for PMR.

Fig. 8. Applied torques around collisions both at a low compliance state at


(P = 225 kPa), a high compliance state at the safety constraint with IJCC
and PBSMC.

It is clear that the high compliance of the joint at the safety con-
straint with the IJCC can decrease the impact impulse during
collisions. As expected, the tracking performance was also se-
cured, in contrast to the results when using PBSMC. A max-
imum tracking error below 0.025 rad was determined for the
safety constraint with IJCC for all joints. This is feasible per-
formance for the tracking problem, though it is slightly greater
than the maximum error of 0.019 rad at the normal condition.

V. EXPERIMENT
The proposed method was examined on the real robot of Fig. 11. Schematic of the PMR. The left is the front view and the right is the
pneumatic muscle robot (PMR). PMR applies the reverse actu- side view. (a) is link1 and (b) is link2. (c) and (d) are encoders of the joint1 and
joint2. (e) and (f) are PMs of bicep and tricep. (g) is an air hose.
ation mechanism of Fig. 2. Joint1 of (c) is actuated by the PMs
of part (a) and Joint2 of (d) by the PMs of part (b) in Fig. 11.
Two encoders are attached at each joint for position feedback, four external proportional pressure regulators for individual
and there are four pneumatic muscles for two revolute joints, pneumatic actuators, and one air compressor. The FESTO
CHOI et al.: POSITION AND COMPLIANCE CONTROL OF A PNEUMATIC MUSCLE ACTUATED MANIPULATOR 839

Fig. 14. Measured forces at collision experiments. Proposed the safety con-
Fig. 12. Still shots of the collision experiment. Manipulator collide against to
0
the loadcell at 11.95 s through T ( = 10:95 s). (a) T ( 0:41; 0:16). (b) Colli-
straint with IJCC and PBSMC show low forces after collisions.
sion at 11.95 s.

Fig. 15. Estimated compliance variations at collision experiments. The com-


pliance at the safety constraint shows more increased joint compliance (3.0–3.4
times of the case of (P = 225 kPa) depending on the position).
Fig. 13. Position variations at collision experiments with the safety constraint
and IJCC. It shows good tracking performance before collisions contrary to
PBSMC. on the floor near the projection of joint1. At that point, the hori-
zontal direction impulses were only measured using a load-cell
company’s 20 mm diameter pneumatic air muscles are used as with the general assumption of the horizontal movement of an
actuators. Also, a control board was developed that consists of obstacle such as a human. The direction of the impact impulses
an AVR MCU with a 10 bit, four-channel DAC whose outputs and the direction of the movement of the obstacle are identical.
are inputs to the proportional pressure regulators. The control The standard value of a load-cell for a unit weight was decided
frequency was set to 60 Hz, as it was during the simulation by hanging a 10 kg dumbbell.
experiments. The motion of the manipulator was given in (34). It could
Two kinds of experiments were carried out. The first presents not reach the desired position because it collided against the
collisions with an object to address the enhanced safety char- load-cell at 11.92 s. The impulses were measured for collision
acteristic with the IJCC at the safety constraint. A load-cell was recognition periods (11.92–12.66 s) under a normal condition of
used to measure the impulse at collision. The second kind of ex- ( 225 kPa) using the safety constraint and the PBSMC.
periment involves a human participant pushing a moving robot The collision detection delay was longer than that of the simula-
manipulator by hand as an alternative form of human-robot col- tion because of the communication delay in the physical system.
lision. The measured forces are shown in Fig. 14. The forces are in-
creased after collision because of joint errors that occur after
A. Collision Experiment reaching the predefined upper limit within a few iterations. The
The experimental environment pictured in Fig. 12 was con- forces are kept as static values due to the controller stop after the
structed to measure impact impulses. A load-cell was positioned collision detection. Impact impulses were decreased to 20.41 Ns
840 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 19, NO. 4, JULY 2011

Fig. 16. Still shots of the pushing experiment. “T n( ; )” indicate a moment at T n when joint1 angle is and joint2 is . T n is a time index of Fig. 17. T2
0
indicate the end of the push action. (a) T 1( 0:31; 0:35), (b) T 2(0:15; 0:46), (c) T 2 , (d) T 3(0:28; 0:48), (e) T 4(0:25; 0:41), (f) T 5(0:27; 0:47).

TABLE III
MEASURED IMPACT IMPULSE AND IAE AT COLLISION EXPERIMENTS

with the proposed method compared to 22.39 Ns with PBSMC


and 30.12 Ns at a normal configuration of ( 225 kPa).
This experiment showed that nearly 33% of the impact impulses
were decreased at the proposed safety constraint when using the
IJCC, compared to the normal operational condition. The safety
performance of the PBSMC was nearly identical to that of the Fig. 17. Joint trajectories with pushing actions at T2. Manipulator’s motion
proposed method, but its tracking performance was lower than at each time T n((T 1; T 2; T 3; T 4; T 5) = (5.4 s, 6.7 s, 7.0 s, 7.2 s, 9.1 s)) is
shown in Fig. 16.
that of the proposed method owing to the virtual spring-coupling
of the PBSMC, as depicted in Fig. 13. The impact impulses at
the safety constraint were decreased by virtue of the increased
joint compliance shown in Fig. 15. This result is in good agree- B. Pushing a Moving Manipulator by Hand
ment with the simulation results, despite the minor differences The fundamental objective of this work is to enhance the
due to measuring noise and minor dimension errors. The posi- safety of humans who are working with robots. An experiment
tion control performance was secured with a low rate of error was devised in which a human pushes the manipulator back-
below 0.028 rad, which is similar to the simulations. ward. This is an alternative version of the collision situation be-
The midpoints of 150 and 300 kPa of were also tween a human and a robot. Such an interaction would be unde-
accessed using an interval of 25 kPa, which is displayed in sirable for robots using motors for actuation. The resilience of
Table III. The range of for comparison was determined by PM-actuated manipulators results in increased safety for both
(30) and the usual experimental results of , the robot and its human operator.
but 225 kPa only applies for the It is clear that higher joint compliance results in more elas-
full swing of . The ticity and greater safer compared to lower joint compliance. A
experimental results from the various showed that lower condition satisfying the safety constraint with the IJCC was ap-
led to lower impact impulses in all collisions. The pro- plied in this pushing experiment. The motions of the manipu-
posed method maximized the joint compliance resulting in the lator at moments T1–T5 in Fig. 17 are shown as a series of still
smallest impulses. In addition, the proposed method can be shots in Fig. 16. T2 is in the point at which the human pushed the
used with a full swing of . Integral of absolute errors manipulator, as the manipulator was moving. A collision with
(IAE) during [1 s (settling time) 11.95 s (collision time)] was the human caused a sudden backward motion of the manipulator
derived to measure only the tracking performance in Table III. at both links, as shown in Fig. 17. This backward motion is pos-
Notably, there were not much increase of IAE of joint posi- sible as a consequence of the increased compliance of the PM.
tions as decrease with IJCC contrary to PBSMC’s large This momentary impact causes oscillations, unlike the sustained
errors. In conclusion, the safety characteristic was enhanced collision in the previous experiments. However, the manipulator
with the proposed method without a great loss of positioning stopped a few moments later because the controller stopped.
performance. During the experiments, the joint compliance was maintained
CHOI et al.: POSITION AND COMPLIANCE CONTROL OF A PNEUMATIC MUSCLE ACTUATED MANIPULATOR 841

collision, and the second is after the initial collision (post-colli-


sion). The initial collision refers to the collision that first oc-
curs between the objects. The initial collision causes unpre-
dictable robot motion, which can then result in other collisions,
or post-collisions. A human can easily be injured by unpredicted
robot motions after the initial collision. To build a genuinely safe
robot, functions to prevent both types of collisions must be con-
sidered.
In this study, only a method to reduce the impact impulse
that occurs upon an initial collision is addressed. Addressing the
safety problems after a collision requires additional study. An
initial collision can damage the sensor system or the mechan-
ical parts of a robot. It is difficult for a controller to accurately
determine environmental information during such an instance
(e.g., the robot’s pose, the relative position of the robot to the
object, or information about the object) even if it is strongly as-
sumed that the system experiences no malfunction caused by
Fig. 18. Estimated compliance variations around collision moments T2 and T5 the initial crash. Moreover, unexpected controllable and uncon-
of Fig. 17. trollable parameters may be added to the main control systems
owing to a completely changed environment.
It is true that post collision hazards are a very important issue
that requires additional study to improve safety between humans
and robots. However, the assurance of safety after the initial col-
lision is beyond the scope of this work, and a basic stop motion
was implemented here to prevent post-collisions.

B. Comparison Study in the Human Demonstration


Demonstrations involving a human participant were executed
and the results are presented in Section V-B. The PBSMC and
BASMC were also tested in a manner similar to that applied for
the IJCC. However, these kinds of experiments are difficult to
regularize. It is difficult to measure experimental data and im-
possible to exactly repeat experiments when human subjects are
involved. This experiment cannot measure the force applied by
the human at the moment of the push and cannot precisely re-
peat the same pushing motion at the same point of contact on
the robot with constant force over the range of various nominal
pressures. Nevertheless, the subjects who pushed the manipu-
1
Fig. 19. Control laws p around collision moments T2 and T5 of Fig. 17. lator were able to observe that use of the IJCC results in more
It shows the sudden increase after T2 followed by zero control law after the
collision detection.
elasticity and greater safety than use of a BASMC. Of course,
it was difficult for the human subjects to observe the compli-
ance difference between using the IJCC and PBSMC. The use
of the IJCC allowed positioning performance superior to that
at high values, as shown in Fig. 18. A joint angle error of almost
when using PBSMC.
0.18 rad ensued directly after the collisions at T2, and the con-
troller subsequently stopped. The control laws in Fig. 19 show VII. CONCLUSION
zero values after the collision detection, following steep increase
It is a fundamental proposition that a robot must never injure
around the times of the push in an attempt to reduce errors.
a human, and robots are considered to be safer when they offer
It is notable that a PMR with a safety constraint determined by
greater joint compliance. In relation to this, a joint compliance
the IJCC can enhance human safety with high joint compliance
control method without a significant loss of precision perfor-
without a significant loss in performance precision.
mance in the manipulator was addressed in this work. PM with
intrinsic elasticity was used as a basic actuator in an antago-
VI. DISCUSSION nistic configuration. The proposed unified control framework,
which uses an IJCC and a SMC for the manipulator actuated by
A. Scope of This Work PM, was utilized to meet acceptable safety characteristics. The
proposed model offers enhanced safety relative to conventional
Briefly, there are two cases in which a human can be injured works in both simulations and experiments involving collisions
by a collision with a robot. The first of these is during the initial with obstacles.
842 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 19, NO. 4, JULY 2011

REFERENCES [24] H. Aschemann and D. Schindele, “Sliding-mode control of a high-


[1] J. H. Graham, J. F. Meagher, and S. J. Derby, “A safety and collision speed linear axis driven by pneumatic muscle actuators,” IEEE Trans.
avoidance system for industrial robots,” IEEE Trans. Ind. Electron., Ind. Electron., vol. 55, no. 11, pp. 3855–3864, Nov. 2008.
vol. IA-2, no. 1, pp. 195–203, Feb. 1986. [25] C.-P. Chou and B. Hannaford, “Measurement and modeling of McK-
[2] B. Karlsson, N. Karlsson, and P. Wide, “A dynamic safety system ibben pneumatic artificial muscles,” IEEE Trans. Robot. Autom., vol.
based on sensor fusion,” J. Intell. Manuf., vol. 11, pp. 475–483, 2000. 12, no. 1, pp. 90–102, Feb. 1996.
[3] E. Ohashi, T. Aiko, T. Tsuji, H. Nishi, and K. Ohnishi, “Collsion avoid- [26] G. K. Klute and B. Hannaford, “Accounting for elastic energy storage
ance method of humanoid robot with arm force,” IEEE Trans. Ind. Elec- in McKibben artificial muscle actuators,” J. Dyn. Syst., Meas., Control,
tron., vol. 54, no. 3, pp. 1632–1641, Jun. 2007. vol. 122, pp. 386–388, Jun. 2000.
[4] M. Zinn, O. Khatib, B. Roth, and J. K. Salisbury, “Playing it safe [27] D. Trivedi, A. Lotfi, and C. D. Rahn, “Geometrically exact models
[human-friendly robots],” IEEE Robot. Autom. Mag., vol. 11, no. 2, for soft robotic manipulators,” IEEE Trans. Robot., vol. 24, no. 4, pp.
pp. 12–21, Jun. 2004. 773–780, Aug. 2008.
[5] G. Hirzinger, A. Albu-schäffer, M. Hähnle, I. Schaefer, and N. Sporer, [28] D. B. Reynolds, D. W. Repperger, C. A. Phillips, and G. Bandry,
“A new generation of torque controlled light-weight robots,” in Proc. “Modeling the dynamic characteristics of pneumatic muscle,” Annals
Int. Conf. Robot. Autom., Seoul, Korea, 2001, pp. 3356–3363. Biomed. Eng., vol. 32, pp. 310–317, 2003.
[6] G. Pratt and M. Williamson, “Series elastic actuators,” in Proc. IEEE [29] Y. Mao, J. Wang, S. Li, and Z. Han, “Energy-efficient control of pneu-
Int. Conf. Intell. Robots Syst., 1995, vol. 1, pp. 399–406. matic muscle actuated biped robot joints,” in Proc. 6th World Congr.
[7] R. Q. van der Linde, “Design, analysis, and control of a low power joint Intell. Control Autom., Dalian, China, Jun. 2006, pp. 8881–8885.
for walking robots, by phasic activation of McKibben muscles,” IEEE [30] K. D. Young, V. I. Uikin, and Ü. Özgüner, “A control engineer’s guide
Trans. Robot. Autom., vol. 15, no. 4, pp. 599–604, Aug. 1999. to sliding mode control,” IEEE Trans. Control Syst. Technol., vol. 7,
[8] M. Wisse, A. L. Schwab, R. Q. van der Linde, and F. C. T. van der no. 3, pp. 328–342, May 1999.
Helm, “How to keep from falling forward: Elementary swing leg action [31] Y. Yildiz, A. Sabanovic, and K. Abidi, “Sliding-mode neuro controller
for passive dynamic walkers,” IEEE Trans. Robot., vol. 21, no. 3, pp. for uncertain systems,” IEEE Trans. Ind. Electron., vol. 54, no. 3, pp.
393–401, Jun. 2005. 1676–1685, Jun. 2007.
[9] I. Mizuuchi, H. Waita, I. Mizuuchi, Y. Nakanishi, T. Yoshikai, M. [32] S. Haddadin, A. Albu-schäffer, and G. Hirzinger, “Safe physical
Inaba, and H. Inoue, “Desing and implementation of reinforceable human-robot interaction: Measurements, analysis & new insights,” in
muscle humanoid,” in Proc. IEEE Int. Conf. Intell. Robots Syst., Sep. Proc. Int. Symp. Robot. Res., Hiroshima, Japan, 2007, pp. 439–450.
2004, pp. 828–833.
[10] K. Kawashima, T. Sasaki, A. Ohkubo, and T. Kagawa, “Application of
robot arm using fiber knitted type pneumatic artificial rubber muscles,”
in Proc. IEEE Int. Conf. Robot. Autom., Apr. 2004, pp. 4937–4942. Tae-Yong Choi (S’10) was born in Busan, Korea, in
[11] Y. Sodeyama, I. Mizuuchi, T. Yoshikai, Y. Nakanishi, and M. Inaba, “A 1977. He received the B.S. degree in electronic and
shoulder structure of muscle-driven humanoid with shoulder blades,” electrical engineering from Pohang University of Sci-
in Proc. IEEE Int. Conf. Intell. Robots Syst., Apr. 2005, pp. 1077–1082. ence and Technology (Postech), Pohang, Korea, in
[12] T. G. Sugar, J. He, E. J. Koeneman, J. B. Koeneman, R. Herman, H. 2003, and the Ph.D. degree in electrical engineering
Huang, R. S. Schultz, D. E. Herring, J. Wanberg, S. Balasubramanian, from Korea Advanced Institute of Science and Tech-
P. Swenson, and J. A. Ward, “Design and control of rupert: A device for nology (KAIST), Daejeon, Korea, in 2010.
robotic upper extremity repetitive therapy,” IEEE Trans. Neural Syst. Currently, he is a Senior Engineer with Samsung
Rehabil. Eng., vol. 15, no. 3, pp. 336–346, Sep. 2007. Electronics, Suwon, Korea. His current research in-
[13] X. Zhu, G. Tao, and B. Yao, “Integrated direct/indirect adaptive robust terests include safe robot, robotics, machine vision,
posture trajectory tracking control of a parallel manipulator driven by and machine learning.
pneumatic muscles,” IEEE Trans. Control Syst. Technol., vol. 17, no.
3, pp. 576–588, May 2009.
[14] A. Bicchi, S. L. Rizzini, and G. Tonietti, “Compliant design for intrinsic
safety: General issues and preliminary design,” in Proc. IEEE Int. Conf. Byoung-Suk Choi (S’08) was born in Busan, Korea,
Intell. Robots Syst., Maui, HI, 2001, pp. 1864–1869. in 1981. He received the B.S. and M.S. degree in
[15] A. Bicchi and G. Tonietti, “Fast and “soft-arm” tactics,” IEEE Robot. electrical engineering from Pusan National Univer-
Autom. Mag., vol. 11, no. 2, pp. 22–33, Jun. 2004. sity, Busan, Korea, in 2004 and 2006, respectively.
[16] G. Tonietti and A. Bicchi, “Adaptive simultaneous position and stiff- He is currently pursuing the Ph.D. degree from
ness control of a soft robot arm,” in Proc. IEEE Int. Conf. Intell. Robots Korea Advanced Institute of Science and Tech-
Syst., Lausanne, Switzerland, 2002, pp. 1992–1997. nology (KAIST), Daejeon, Korea.
[17] D. J. Clapa, E. A. Croft, and A. J. Hodgson, “Equilibrium point control Currently, he is taking part in a research project of
of a 2-DOF manipulator,” J. Dyn. Syst. Meas. Control-Trans. ASME, the Unmanned Technology Research Center (UTRC)
vol. 128, pp. 134–141, Mar. 2006. supported by the Agency for Defense Development
[18] M. Van Damme, B. Vanderborght, R. Van Ham, B. Verrelst, F. Daerden, (ADD), Korea. He is also taking part in a research
and D. Lefeber, “Proxy-based sliding mode control of a manipulator project of the National Robotics Research Center for Robot Intelligence Tech-
actuated by pleated pneumatic artificial muscles,” in Proc. IEEE Int. nology supported by the Ministry of Knowledge Economy of Korea. His re-
Conf. Robot. Autom., Roma, Italy, Apr. 2007, pp. 4355–4360. search interests include mobile robot.
[19] R. Kikuuwe and H. Fujimoto, “Proxy-based sliding mode control for
accurate and safe position control,” in Proc. IEEE Int. Conf. Robot.
Autom., Orlando, FL, May 2006, pp. 25–30.
[20] C. S. Tzafestas, N. K. M’Sirdi, and N. Manamani, “Adaptive Kap-Ho Seo (M’09) was born in Pusan, Korea,
impedance control applied to a pneumatic legged robot,” J. Intell. in 1974. He received the B.S. degree in electrical
Robot. Syst., vol. 20, pp. 105–129, 1997. engineering from Korea University, Seoul, Korea, in
[21] N.-C. Park, H.-S. Yang, H.-W. Park, and Y.-P. Park, “Position/vibra- 1999, and the M.S. and Ph.D. degrees in electrical
tion control of two-degree-of-freedom arms having one flexible link engineering from Korea Advanced Institute of
with artificial pneumatic muscle actuators,” Robot. Autonomous Syst., Science and Technology (KAIST), Daejeon, Korea,
vol. 40, pp. 239–253, 2002. in 2001 and 2009, respectively.
[22] S. Katsura, J. Suzuki, and K. Ohnishi, “Pushing operation by flex- Since 2009, he has been a Senior Researcher with
ible manipulator taking environmental information into account,” IEEE the Pohang Institute of Intelligent Robotics (PIRO),
Trans. Ind. Electron., vol. 53, no. 5, pp. 1688–1697, Oct. 2006. Pohang, Korea. He has worked in developing robotic
[23] J. H. Lilly and L. Yang, “Sliding mode tracking for pneumatic muscle system for the elderly and disabled. His current re-
actuators in opposing pair configuration,” IEEE Trans. Control Syst. search interests include embedded system, neural network, vision-based mobile
Technol., vol. 13, no. 4, pp. 550–558, Jul. 2005. robot, and rehabilitation robot system.

Você também pode gostar