Comparison fingers. By controlling the wrist pose and orientation

Comparison of
Grasping Technique using a Sensor-Based Reflex Mechanism and using an Electric
Field Pretouch System.

 

Sandra Susan Mathew

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

Divya D

Department of Computer Science and Engineering

Department of Computer Science and Engineering

Mahatma Gandhi University

Mahatma Gandhi University

Kottayam

Kottayam

Adi Shankara Institute of Engineering and Technology,
                                           
Kalady

Adi Shankara Institute of Engineering
and Technology,
                                            Kalady

 

                Abstract – This paper presents comparison of two
grasping techniques. The first grasping technique uses a sensor based reflex
mechanism. A soft robotic hand that is a Pisa/IIT soft hand with IR sensors
attached to the finger tips is used to achieve the objective. A
feedback received from the sensors used at the finger tips of the hand helps in
the correction of hand-object posture so as to enhance the probability that the
object is approached uniformly by all the fingers. By controlling the wrist
pose and orientation and hand closure, the distance between the hand’s finger
pads and the object is minimised through the algorithm. The second technique
uses an electric field sensing mechanism. A Barrett hand with transmit and
receive electrodes at its fingers are used to achieve the objective. The analog
front end, a current amplifier, measures the current induced at the receiver,
which gets modified when an object is to be grasped.

 

 

                Index Terms – List
key index terms here. No mare than 5.

 

I.                   
Introduction

        The problem of
autonomous robotic grasping has been in the focus of robotic research community
for the past three decades. Intelligent and effective grasp plans have been
developed that allow robotic hands to perform grasping tasks close to that of
an object. In this paper we discuss and compare two different grasping
techniques.  One among them uses the
electric field sensing mechanism. A Barrett hand with transmit and receive
electrodes at its fingers are used  along
with an analog front end, a current amplifier,which measures the current
induced at the receiver, which gets modified when an object is to be grasped.
Here, electric field sensing is employed for non- contact measurements with
good performances in improving the reliability of manipulation. Such techniques
rely on finding optimal fingerpad placement on the object, while the
surrounding environment has to be avoided as an obstacle.However these
approaches are limited by the hand rigidity and fragility, and the manipulation
strategies are very far from those a human would perform in real scenarios. In this work, an
approach to grasp refinement for the Pisa/IIT Softhand using IR sensors is
presented. The idea was inspired by the concept of “caging” an object, i.e., to
prevent an object from escaping from the “cage” that is formed by robots or
parts of robots. However, even if reminiscent of the transition from caging to
grasping, we want to clearly state that our approach simply consists in keeping
all the fingers equally close to the object as the hand closes, by using IR
sensor information.

 

II.                 
Grasping Technique Using a
Sensor Based Reflex Mechanism

    The main aim of this approach is to reduce
uncertainties and perform a successful grasp. The complete algorithm has 3
phases:

1)    The pre-grasp phase: In this phase the
robot arm is m  oved to the pre-grasp
location

2)    The sensor checking phase: In this phase it
is checked if the object is visible to all the sensors and if not the fingers
are positioned accordingly.

3)    The grasp refinement phase: In this phase
optimization of hand to object position is done before the final execution of
the grasp.

 

A.                  
Pre-grasp
Phase

 

    The algorithm used in this phase includes 2
methods M1 and M2. The 1st method, M1 is used when the object is not
previously known and the grasp almost guessed. In this method a rough estimate
of the object center of mass which uses only a RGB-D cue of the object is
calculated to position the hand accordingly. The 2nd method, M2 is
used when the object is previously known which means the grasp is already known
to be nearly successful. This method includes an object recognition phase using
a vision sensor and it also depends on a pre-computed grasp database 2
consisting of hand postures associated with the object 3. Also, after each
successful grasp the relative hand to object position is recorded in the grasp
database and is evaluated in the Gazebo simulator 4 and checked for
collisions and reachability in the current scenario.

 

B.                  
Sensor
Checking Phase

 

    In this
phase, it is made sure that the object is visible to all the IR sensors at the
finger tips. 

 

Fig. 1.    The Pisa/IIT Softhand with IR sensors mounted on the thumb, index and
ring fingers

 

 

 

    
Due to the particular shape of the fingers, the middle finger lags behind
the others and its contribution to the grasp becomes important only when the
other fingers are already in contact. If we consider the distance between the
middle finger tip and the object in this phase, it would lead to wrong
conclusions when the objective is to center the fingers around the object.  Contact with the little finger rarely happen
when objects are very small. So, only three out of five fingers (thumb, index
and ring) are considered for mounting sensors to achieve a good grasp.

 

TABLE I

All
Possible Situations of Sensor Visibilities

 

Case #

Visible Sensors(s)

c1

None

c2

Thumb

c3

Index

c4

Ring

c5

Thumb,
Index

c6

Thumb,
Ring

c7

Index,
Ring

 

TABLE II

Recovering
Actions for Different Cases of Sensor Visibilities

Action #

Recovering action

?x

Magnitude

Case#

a1

translational motion

-dx, -dy

4 mm

c1

a2

translational motion

dx, -dy

2 mm

c2

a3

translational motion

-dy

2 mm

c3

a4

translational motion

-dx, -dy

2 mm

c4, c6

a5

translational motion hand closing

dx, d?

2 mm, 20 tick

c5

a6

translational motion

-dx, dy

2 mm

c7

 

   The algorithm used in this
phase considers both the above tables. In this, first it is checked if the
object is visible to any of the sensors, if no, table 3.1 is referred for the
case. Once the case is identified table 2 is referred for the necessary actions
to be performed. For example, if case 4 (c4) is identified, ie, the object is
visible to only the ring finger. Now, in table 3.2, for c4 the corresponding
action is a4, which is translational motion in two directions {-x, -y} with a
displacement of 2 mm in each direction. This process is repeated until the
object is visible to all the 3 IR sensors.

 

C.                  
Grasp
Refinement Phase

    

     In this
phase, optimization is done by computing cost function, if the value is more
than the minimum expected one, then necessary corrections are computed and hand
movements are performed as required.

 

III.              
Grasping Technique Using an
Electric Field Pretouch System

     In this approach, an electric field
sensing mechanism is used to perform the manipulation for grasping by a robotic
arm 5. Here, Barrett hand is used as an end effector 6. The fingers of this
hand are having transmit and receive electrodes. When an AC voltage is applied
to the transmit electrode an AC current is induced in the receive electrode. If
an object comes in between the fingers the there will be a variation in the
induced current. The changes are measured using sensor boards placed at the
fingers as well as the palm. The sensor results received at the fingers indicates
distance between the fingers and the object which helps in the pre-grasp
configuration of the hand.

 

            

            Fig.
2.    Barrett Hand labeled with coordinate axis

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

                                            TABLE
III

                                        Comparison

Comparison based on

 “An
Electric Field Pretouch System For Grasping And Co-Manipulation ”
 

 Enhancing
Adaptive Grasping Through A Simple Sensor-Based Reflex Mechanism

End
effector
                      

Barett
hand – 3 finger hand

Softhand
with IR sensor

Uncertainties
and their avoidance
 
 
 

These
are developed for robotic arms with rigidity.
 
 
 

Uncertainties
can be avoided using sensor feedback and this technique is employed for soft
robotic arms
 

Comparison with real world
scenario

Manipulation strategies are very far from those a human
would perform in real scenarios

Manipulation strategies are very close to those of a human would perform in real scenarios

Advantages

·   This
technique helps in improving the reliability of manipulation.   
·    Here, robotic hands with sensible rigidity
and high degrees of actuation were used.
·   This
technique addresses the problem of grasping refinement for complaint hands.
 

·    The success rate is considerably increased
from 16 percent to 83 percent with this algorithm.
·    Due to the simplified vision acquisition
and fast response of IR sensors, the overall time consumption is much less.
·    In other cases, where IR-guided strategy
is not used, the successful grasp drop to 0 when the object is deliberately
moved from its position. But in this case, the arm was able to recapture the
lost object and recover 70 percent of the failures without knowing the new
locations.
·    An effective method to refine the final
grasping pose of a softhand with respect to an object is obtained using low
cost IR sensors.
 

Disadvantages

·   The system would certainly fail for
objects that are different in size from those, the system was tuned for.
·   More general approaches to interpreting
the sensor data are needed to allow the system to succeed.
 

·    It requires extra time and experiments to
construct the databases used for object  
recognition and grasps.
·    Procedures need to be repeated for new
objects.
·    IR sensors become insensitive to distance
variations below a higher threshold distance.
 

 

IV.               
Discussion and Future Work

     The grasping technique which uses an
electric field pretouch system will succeed for objects it will be tuned for
and also objects of similar sizes. But the system would certainly fail for
objects that are drastically different in size from those it was tuned for.
More general approaches to interpreting the sensor data are needed to allow the
system to succeed with much wider range of object geometries. The grasping technique which uses a sensor
based reflex mechanism will succeed for objects it will be tuned for and
also for objects which weren’t tuned. Grasping for Objects which were displaced
from their original position were also successful. This technique can be applied on various applications in the fields of
surgical medicine, industries and agriculture. There is scope for refinement to
achieve more accurate results with the goal of employing it as a soft probe for
perceiving unstructured environments and enhancing its autonomous grasping
ability. Further it can be improved to achieve hundred percent grasp for new
objects and also sensors which does not become insensitive to distance
variations below a higher threshold distance can be employed.

 

References

1 Emanuele Luberto, Yier Wu, Gaspare Santaera, Marco Gabiccini, and
Antonio Bicchi, “Enhancing Adaptive Grasping Through a Simple

Sensor-Based Reflex
Mechanism”, Ieee Robotics And Automation Letters, Vol. 2,
No. 3, July 2017,pp. 1664-1668.

2 Dual manipulation grasp database. Online.
Available: https://bitbucket.org/dualmanipulation/grasp_db

3 A.Miller and P.Allen, “Graspit! a versatile
simulator for robotic grasping,” IEEE Robot. Autom. Mag., vol. 11, no. 4, pp.
110–122, Dec. 2004.

4 Gazebo simple grasp utility. Online.
Available:        https://bitbucket.org/hamalMarino/gazebo_simple_grasp_utility

5  B.
Mayton, L. LeGrand, and J. Smith, “An electric field pretouch system for
grasping and co-manipulation,” in Proc. Proc. IEEE Int. Conf. Robot. Autom.,
May 2010, pp. 831–838.

6 M R Hasan, R Vepa, H Shaheed; H Huijberts, “Modelling
and Control of the Barrett Hand for
Grasping”, in UKSim
15th International Conference on Computer Modelling and Simulation,  2013, pp.  230 – 235.