Beruflich Dokumente
Kultur Dokumente
net/publication/304614906
CITATIONS READS
3 427
4 authors, including:
Tatiana G Kalganova
Brunel University London
83 PUBLICATIONS 783 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Mashood Mukhtar on 21 December 2017.
Abstract Robot hands have attracted increasing research interest in recent years
due to their high demand in industry and wide scope in number of applications.
Almost all researches done on the robot hands were aimed at improving
mechanical design, clever grasping at different angles, lifting and sensing of
different objects. In this chapter, we presented the detail classification of control
systems and reviewed the related work that has been done in the past. In
particular, our focus was on control algorithms implemented on pneumatic
systems using PID controller, Bang-bang controller and Backstepping controller.
These controllers were tested on our uniquely designed ambidextrous robotic
hand structure and results were compared to find the best controller to drive such
devices. The five finger ambidextrous robot hand offers total of 13 degrees of
freedom (DOFs) and it can bend its fingers in both ways left and right offering
full ambidextrous functionality by using only 18 pneumatic artificial muscles
(PAMs).
__________________________
Mashood Mukhtar, Emre Akyürek, Tatiana Kalganova
Department of Electronic and Computer Engineering, Brunel University, Kingston Lane,
Uxbridge, London, UK, e-mail: Mashood.Mukhtar@brunel.ac.uk
Nicolas Lesne
Department of System Engineering, ESIEE Paris, Noisy-le-Grand Cedex, France
1 Introduction
Robots play a key role in our world from providing daily services to industrial
use. Their role is in high demand due to the fact they offer more productivity,
safety at workplace and time reduction in task completion. Number of control
schemes are found in the literature [1] to control such robot such as adaptive con-
trol [2], direct continuous-time adaptive control [3], neurofuzzy PID control [4]
hierarchical control [5], slave-side control [6], intelligent controls [7], neural
networks [8], just-in-time control [9] ,Bayesian probability [10], equilibrium-
point control [11], fuzzy logic control [12], machine learning [13], evolutionary
computation [14] and genetic algorithms [15], nonlinear optimal predictive con-
trol [16],optimal control [17], stochastic control [18], variable structure control
[19], chattering-free robust variable structure control [20] and energy shaping
control [21], gain scheduling model-based controller [22], sliding mode control
[23], proxy sliding mode control [24], neuro-fuzzy/genetic control [25] but there
are five types of control systems (Fig.1) mainly used on pneumatic muscles.
Focus of our research was on the feedback control system and non-linear con-
trol systems. Feedback is a control system in which an output is used as a feed-
back to adjust the performance of a system to meet expected output. Control sys-
tems with at least one non-linearity present in the system are called nonlinear
control systems. In order to reach a desired output value, output of a processing
unit (system to be controlled) is compared with the desired target and then feed-
back is provided to the processing unit to make necessary changes to reach closer
to desire output. The purpose of designing such system is to stabilise the system
at certain target.
Tactical sensors are employed to investigate the grasping abilities of the ambi-
dextrous hand. These sensors have been used several times in the past on the ro-
bot hands driven by motors [26], [27] and robot hands driven by pneumatic artifi-
cial muscles [28] and [29]. The automation of the ambidextrous robot hand and
its interaction with objects are also augmented by implementing vision sensors on
each side of the palm. Similar systems have already been developed in the past.
For instance, the two-fingered robot hand discussed in [30] receives vision feed-
back from an omnidirectional camera that provides a visual hull of the object and
allows the fingers to automatically adapt to its shape. The three-fingered robot
hand developed by Ishikawa Watanabe Laboratory [31] is connected to a visual
feedback at a rate of 1 KHz. Combined with its high-speed motorized system that
allows a joint to rotate by 180 degrees in 0.1 seconds, it allows the hand to inter-
act dynamically with its environment, such as catching falling objects. A laser
displacement sensor is also used for the two-fingered robot hand introduced in
[32]. It measures the vertical slip displacement of the grasped object and allows
the hand to adjust its grasp. In our research, the aim of the vision sensor is to de-
tect objects close to the palms and to automatically trigger grasping algorithms.
Once objects are detected by one of the vision sensors, grasping features of the
ambidextrous robot hand are investigated using three different algorithms, which
are proportional-integrative-derivative (PID), bang-bang and Backstepping con-
trollers. Despite the nonlinear behavior of PAMs actuators [33], previous re-
searches indicates that these three algorithms are suitable to control pneumatic
systems.
The work presented in this chapter aim to validate the possibility of controlling
a uniquely designed ambidextrous robot hand using PID controller, Bang-bang
controller and Backstepping controller. The ambidextrous robot hand is a robotic
device for which the specificity is to imitate either the movements of a right hand
or a left hand. As it can be seen in Fig.2, its fingers can bend in one way or an-
other to include the mechanical behavior of two opposite hands in a single device.
The Ambidextrous Robot Hand has a total of 13 degrees of freedom (DOFs) and
is actuated by 18 pneumatic artificial muscles (PAMs) [58].
2 Implementation of Controllers
2.1 PID Controller
PID controller is a combination of proportional, integral and derivative controller
[34].Proportional controller provides corrective force proportional to error pre-
sent. Although it is useful for improving the response of stable systems but it
comes with a steady-state error problem which is eliminated by adding integral
control. Integral control restores the force that is proportional to the sum of all
past errors, multiplied by time [35]
To imitate the behaviour of human finger correctly, some of the pneumatic arti-
ficial muscles must contract slower than others. This prevents having medial and
distal phalanges totally close when the proximal phalange is bending. Thus, ex-
periments are done to collect PAMs’ pressure variation according to the fingers’
position. As the finger is made of one proximal phalange and two other phalanges
for which the movement is coupled, the finger has nine extreme positions. These
positions are numbered from 1 to 9. Position 1 refers to the proximal and medi-
al/distal phalanges reaching both the maximum range on their left side. Position 2
refers to the proximal phalange reaching its maximum range on its left side,
whereas the medial/distal phalanges are straight. Position 3 refers to the proximal
phalange being on its maximum range on its left side, whereas the medial/distal
phalanges are on their maximum range on the right side. Position 4 to 6 refers to
the finger when the proximal phalange is straight. Last position, position 9, is
when the proximal and medial/distal phalanges reach both the maximum range on
their right side. The experiment is summarized in Table 1 and Fig.3.
4
Pressure variation (bars)
Proximal left
3 PAM
2 Medial left
PAM
1 Medial right
PAM
0
1 3 5 7 9
Fingers' position number
Fig.3. Pressure variations at extreme finger positions.
When the fingers’ close from vertical to left position (position 5 to the average
of positions 4 and 2), it is observed that the pressure of medial left PAM varies
thrice less than the pressure of the proximal left PAM, whereas the pressure of
the two medial PAMs vary from the same ratios but in opposite ways. Therefore,
the proportional constant gains are consequently defined as:
1.1
1.05
Forefinger
1
Force (N)
0.95 Middle
finger
0.9 Ring
finger
0.85
Little
0.8 finger
0.75
0.1 0.15 0.2 0.25 0.3
Time (sec)
Fig.4. The Ambidextrous Robot Hand holding a can (a grasping movement implemented with
PID controllers) and a graph representing data collected after every 0.05 seconds.
From Fig.4, it can be seen that grasping of a can of soft drink began at 0.15 ap-
proximately and it became more stable after 0.2 sec. An overshoot has occurred on
all fingers but stayed in limit of 0.05N where system has automatically adjusted at
the next collection. These small overshoots occurred because different parts of the
fingers get into contact with the object before the object actually gets into contact
with the force sensor. This results bending of fingers slower when phalanges
touches the object. Since the can of soft drink does not deform itself, it can be de-
duced that the grasping control is both fast and accurate when the Ambidextrous
Hand is driven by PID loops.
Anthropomorphic po-
Ambidexterity
Robot hand
# fingers
grasping
# DOFs
ror (%)
Gifu Hand II [39] 5 Motors 16 PID 0.55ab 16%ab
High-speed hand [40] 3 Motors 8 PD 0.05a N/A
ITU Hand [41] 2 SMAs 1a N/A 3.76 N/A
I. Yamano and T. Maeno
5 SMAs 20 N/A N/A 6%a
[42]
L. Zollo et al. [43] 3 Motors 10a PD N/A N/A
S. Nishino et al. [29] 5 PAMs 13a PID, Cascade 1.0a 6%a
D. Gunji et al. [32] 2 Motors 1 PD 0.6a N/A
T. Yoshikawa [44] 2 Motors 4a PID 0.55a 22%a
N/
J.Y. Nagase et al. [45] 4 PAMs Fuzzy logic 0.9a 9%ad
A
DLR Hand II [46] 4 Motors 16a N/A 28.7 N/A
DLR Hand [47] 5 Motors 19 Cascade 0.1a N/A
Shadow Hand [26] 5 Motors 20 PID 0.15a N/A
ACT Hand [48] 5 Motors 23 N/A N/A N/A
T. Nuchkrua et al. [49] 3 PAMs 3 N/A N/A N/A
TU Bionic Hand [50] 5 Motors 15 PID 0.208 N/A
DEXMART Hand [51] 5 Motors 20 NN 0.25a 25%ad
Ambidextrous Hand [52] 5 PAMs 13 PID 0.20 8%
a
Estimations made from curves, pictures or videos of the robot hands
b
Experiments do not concern grasping but contact tasks
c
Only for grasping
d
For a target of 1 N ± 10%, whereas the error can exceed 30% for lower force
targets
Table 2. Comparison of various robot hands with our ambidextrous robot hand.
𝐹𝑡𝑚𝑖𝑛 = 𝑊𝑜 + ∑ 𝐹𝑓 (5)
𝑓=1
1.5
1.4
Force (N)
Forefinger
1.3
Middle finger
1.2
Ring finger
1.1
1 Little finger
0.9
0.1 0.15 0.2 0.25 0.3
Time (sec)
Fig.6. Graph shows force against time of the four fingers while grasping a drink can with bang-
bang controllers. The ambidextrous robot hand holding a can with a grasping movement imple-
mented with bang-bang controllers. It is seen the can has deformed itself.
It is noted during experiment that speed of fingers does not vary when it touches
the objects. That explains why we got higher curves in Fig.6 than the previously
obtained in Fig.4. This makes bang-bang controller faster than PID loops. The
bang-bang controllers also stop when the value of 1 N is overreached but, without
predicting the approach to the setpoint, the process variables have huge over-
shoots. The overshoot is mainly visible for the middle finger, which overreaches
the setpoint by more than 50%. Even though backward control is not implemented
in the bang-bang controller, it is seen the force applied by some fingers decreases
after 0.2 sec. This is due to the deformation of the can, which reduces the force ap-
plied on the fingers. It is also noticed that the force applied by some fingers in-
crease after 0.25 sec, whereas the force was decreasing between 0.2 and 0.25 sec.
This is due to thumb’s adduction that varies from 7.45 to 15.30 N from 0.1 to 0.25
sec. Even though the fingers do not tighten anymore around the object at this
point, the adduction of the thumb applies an opposite force that increases the forc-
es collected by the sensors. The increase is mainly visible for the forefinger and
the middle finger, which are the closest ones from the thumb. Contrary to PID
loops, it is seen in Fig.6 that the force applied by some fingers may not change be-
tween 0.15 and 0.2 sec, which indicates the grasping stability is reached faster with
bang-bang controllers. The bang-bang controllers can consequently be applied for
heavy objects, changing the setpoint of 𝐹𝑓 defined in (5).
as an outer loop
Controller used
Ambidexterity
bang control
# actuators
# DOFs
(sec)
trol
Modular
R. Van Ham et N/
part of a PAMs Pressure control PID 2 0.2a N/A
al. [57] A
leg
B. Vanderbor- Generate a joint tra- 4% for pressurea
Two legs PAMs PI 3 6 0.4a
ght et al. [64] jectory 0.27% for anglea
B. Vanderbor- Generate a joint tra- 3% for pressure
Two legs PAMs PID 3 6 0.2a
ght et al. [65] jectory 4.5% for anglea
17% for pressurea
B. Vanderbor- Generate a joint tra-
Two legs PAMs Delta-p 6 12 0.15a 17.5% for posi-
ght et al. [63] jectory
tiona
Z. Xu et al. Index of a Air cylin- Evaluation of speed
None 2b 4b 0.33 N/A
[62] hand ders capabilities
Ambidextrous 53% for
Hand PAMs Force control Proportional 9b 14b 0.15
Hand [37] the force
a
Estimations are made from curves
b
These numbers do not correspond with the actuators of the robotic structure but with the ones used for the bang-bang control
Table 3. Comparison of bang-bang controls’ characteristics between the Ambidextrous Hand and
other robotic models.
The force provided by the hand is assumed not being strong enough as long as
𝑒12 /2 exceeds a minimum grasping force defined as 𝐹𝑔𝑚𝑖𝑛 . In (8), it is noted
that 𝑒̇1 (𝑡) ≠ 0 as long as 𝐹𝑓 (𝑡) keeps varying. Therefore, 𝑉1̇ (𝑒1 ) cannot be stabi-
lized until the system stops moving. Thus, a stabilizing function is introduced.
This stabilizing function is noted as a second error 𝑒2 (𝑡):
𝑒2 (𝑡) = 𝑘 ∗ 𝑒̇1 (𝑡) (9)
Fig.7. Basic Backstepping controller is shown.
with 𝑘 a constant > 1. 𝑒2 (𝑡) indirectly depends on the speed, as the system cannot
stabilized itself as long as the speed is varying. Consequently, both the speed of
the system and 𝑒2 (𝑡) are equal to zero when one finger reaches a stable position,
even if 𝐹𝑓 (𝑡) ≠ 𝐹𝑡 . 𝑘 aims at increasing 𝑒̇1 (𝑡) to anticipate the kinematic moment
when 𝑒̇1 (𝑡) becomes too low. In that case, the BSC must stop running as 𝐹𝑓 (𝑡) is
close to 𝐹𝑡 . Both of the errors are considered in a second Lyapunov function:
1 2
𝑉2 (𝑒1 , 𝑒2 ) = (𝑒 (𝑡) + 𝑒22 (𝑡)) < 𝐹𝑠 (10)
2 1
𝑉2̇ (𝑒1 , 𝑒2 ) = 𝑒1 (𝑡) ∗ 𝑒̇1 (𝑡) + 𝑒2 (𝑡) ∗ 𝑒̇2 (𝑡) (11)
where 𝐹𝑠 refers to a stable force applied on the object. This second step allows sta-
bilizing the system using derivative control. Using (9), (11) can be simplified as:
𝑉2̇ (𝑒1 , 𝑒2 ) = 𝑒̇1 (𝑡) ∗ (𝑒1 (𝑡) + 𝑘𝑒̈1 (𝑡)) (12)
1.1
Forefin
ger
1
Middlef
0.9 inger
Force (N)
Ring
0.8 finger
0.7 Little
finger
Fig.8. Graph shows force against time of the four fingers while ambidextrous hand grasping a
can of soft drinks using backstepping controllers.
Algorithms
# actuators
Error max.
Ambidex-
Execution
time (sec)
combined
actuators
structure
to which
the BSC
Robotic
Type of
# DOFs
Aim of
(in %)
terity
BSC
it is
C.-Y Su and Y. Ste- Trajectory
Manipulator Motors None N/A N/A 0.3a 10%a
panenko [74] control
PAM’s
N/A, single
P. Carbonell et al. [67] PAM length None N/A 1 2.4a 8%a
PAM
control
PAM’s
N/A, single
P. Carbonell et al. [68] PAM length Fuzzy logic N/A 1 7a 1%a
PAM
control
Force and
D. Nganya-Kouya et al.
Manipulator N/A position None 4 N/A 9a N/A
[75]
control
Trajectory
Lotfazar et al. A. [76] Manipulator Motors None 5 5 2a N/A
control
Force and
S.-H. Wen S.-H. Wen
Manipulators N/A position NN N/A N/A 1.9ab 9%ab
and B. Mao. [77]
control
H. Aschemann and D. Parallel ro- Position
PAMs None 2 4 0.3a 4%a
Schindele [71] bot control
M.R. Soltanpour and Trajectory
Manipulator Motors SMC 2 N/A 7.6ab <10-3%b
M.M. Fateh [78] control
Parallel ro- Position
X. Liu and A. Liadis [79] Motors Fuzzy logic 2 2 N/A N/A
bot control
Trajectory
L. Qin et al. [72] Arm N/A SMC 6 N/A 0.45a 2%a
control
H. Aschemann and D. Position
Axis PAMs None 1 2 4a 1.4%a
Schindele [73] control
Ambidextrous Hand, Force
Hand PAMs None 9c 14c 0.37 4%
2014 [80] control
a
Estimations are made from curve
b
Results are obtained through a simulation
c
Four DOFs and four actuators are unused for the BSC
Table 4. Comparison of BSCs’ characteristics between the Ambidextrous Hand and other robot-
ic mode
CONCLUSION
PID, Bang-bang and BSC are tested and their performance in controlling an ambi-
dextrous robotic hand is analyzed. PID controller was found the best when applied
as compare to Bang-bang and Backstepping control. Backstepping control tech-
nique was validated for the first time on an ambidextrous robot hand. By combining
PID controllers and force sensors, this research proposes one of the cheapest solu-
tions possible. In future, PID controller could be used with artificial intelligent con-
trollers to further improve the controls.
ACKNOWLEDGMENT
The authors would like to cordially thank Anthony Huynh, Luke Steele, Michal
Simko, Luke Kavanagh and Alisdair Nimmo for their contributions in design of
the mechanical structure of a hand, and without whom the research introduced in
this paper would not have been possible.
References