**5. ANFIS control of an intelligent robotic gripper**

92 Fuzzy Inference System – Theory and Applications

spaced cluster centers, we typically choose rb =1.25 ra (Chopra et al. , 2006).

as the second cluster center. In general, after the K'

revise the potential of each data point by the formula

cluster center into a fuzzy rule for identifying the class.

*<sup>i</sup> x* } then class is c1.

is a constant defined by Equation 10 .

Where xk\* is the location of the K'

Suppose cluster center xi

provides the rule: Rule : *i* If {x is near \*

Where

highest degree of fulfillment.

center P1\*. Typically, one can use pk\* < 0.15P1

The degree of fulfillment of {x is near \*

 <sup>2</sup> 4 *br*

and rb is a positive constant. Thus, we subtract an amount of potential from each data point as a function of its distance from the first cluster center. The data points near the first cluster center will have greatly reduced potential, and therefore will unlikely be selected as the next cluster center for the group. The constant rb is effectively the radius defining the neighborhood which will have measurable reductions in potential. To avoid obtaining closly

When the potential of all data points in the group has been reduced according to Equation 11, we select the data point with the highest remaining potential as the second cluster center. We then further reduce the potential of each data point according to their distance potential

of acquiring new cluster center and reducing potential repeats until the remaining potential of all data points in the group is below some fractions of the potential of the first cluster

Each cluster center found in the training data of a given class identifies a region in the feature space that is well populated by members of that class. Thus, we can translate each

*<sup>i</sup> x* } is defined as

By applying subtractive clustering to each class of data individually, we thus obtain a set of rules for identifying each class. The individual sets of rules can then be combined to form the rule base of the classifier. For example, suppose we found 2 clusters centers in class c1 data, and 5 cluster centers in class c2 data, then the rule base will contain 2 rules that identify class c1 members and 5 rules that identify class c2 members. When performing classification, the output class of the classifier is simply determined by the rule with the

2 \* *<sup>i</sup> x x*

<sup>2</sup> \*

\* *i k x x P P Pe i ik* (13)

\* as the stopping criterion (Chiu, 1997).

*<sup>i</sup> e* (14)

\* was found in the group of data for class c1; this cluster center

th cluster center and pk\* is its potential value. The process

Where

 

\* 1

<sup>2</sup> \*

<sup>1</sup> *<sup>i</sup> x x P P Pe i i* (11)

(12)

th cluster center has been obtained, we

The effectiveness of the Fuzzy Inference control will be illustrated here by applying the method to control the operation of a robotic gripper. The robotic gripper will be first described, its operation principle will be illustrated, then the application of the Adaptive Network Fuzzy Inference System control to the gripper system will be presented.

Generally, the main goal of robotic gripper during object grasping and object lifting process is applying sufficient force to avoid the risk of a difficult task or sometimes a task that could not be achieved. The problem can be posed as an optimization problem (Ottaviano et al.,2000; Bicchi & Kumar,2000). Sensory systems are very important in this field. Two types of sensing are most actively being investigated to increase robot awareness: contact and non-contact sensing. The main type of non-contact sensing is vision sensing where video camera is processed to give the robot the object information. However, it is costly and gives no data concerning forces (Lorenz et al.,1990). Tactile sensing, on the other hand, has the capability to do proximity sensing as well as force sensing, it is less expensive, faster and needs less complex equipment (Choi et al.,2005). The basic principle of the Slip-Sensitive Reaction used in this work is that, the gripper should be able to automatically react to object slipping during grasp with the application of greater force. A lot of researches have been focusing on fingertip sensors development to detect slippage and applied force (Dario & De Rossi ,1985; Friedrich et al., 2000), which requires complicated drive circuit and suffers from difficult data processing and calibration. Polyvinylidene fluoride (PVDF) piezoelectric sensors are presented in (Barsky et al.,1989) to detect contact normal force as well as slip. Also, an array 8x8 matrix photo resistor is introduced in (Ren et al.,2000) to detect slippage. A slip sensor based on the operation of optical encoder used to monitor the slip rate resulting from insufficient force is presented in (Salami et al, 2000). However, it is expensive and have some constrains on the object to be lifted. Several researchers handle finger adaptation using more than one link in one finger to verify stable grasping (Seguna & Saliba, 2001; Dubey & Crowder, 2004). This results in complicated mechanical system leading to difficulty in control and slow response. Fuzzy controllers have been very successful in solving the grasping problem, as they do not need mathematical model of the system (Dominguez-Lopez & Vila-Rosado, 2006). In this study, a new design and implementation of robotic gripper with electric actuation using brushless dc servo motor is presented. Standard sensors adaptation in this work leads to maintaining the simplicity of the mechanical design and gripper operation keeping a reasonable cost. The gripper control was achieved through two control schemes. System modeling had been introduced using ANFIS approach. A new grasping scenario is used in which we collect information about the masses of the grasped objects before starting the grasping process without any additional sensors. This is achieved through knowledge of object pushing force that allows applying an appropriate force and minimizing object displacement slip through implementation of the proposed fuzzy control.

#### **5.1 Gripper design and configuration**

A proper gripper design can simplify the overall robot system assembly, increase the overall system reliability, and decrease the cost of implementing the system. Hence, the design of the gripping system is very important for the successful operation.

Control of Efficient Intelligent Robotic Gripper Using Fuzzy Inference System 95

permanent magnet brushless dc motor (BLDC). It has the advantage of high power density, ease of control, high efficiency, low maintenance and low rotor inertia. BLDC servo motor used is an internal rotor motor "BLD3564B" from Minimotor inc. with its drive circuit

The design of the gripper fingers must take some restrictions into consideration. Long fingers require high developed torque and short fingers impose restrictions on object dimensions. Hence fingers are selected to be 15 cm long. Also, a contact rubber material area between the fingers and the object of 25 mm by 25 mm is used to decrease the pressure on the object, increase the friction, and avoid deformation from centric concentrated force. With this gripper configuration, we succeeded to verify all previous design guidelines except guideline no.4 as our proposed gripper doesn't fully encompass the object in order to be able to grasp a greater variety of objects, although this imposes more difficulty in the control

To build the proposed controller, we need to get information about the system characteristics for use in simulation and experimental work. Hence, input/output variables of the system are measured and processed. The input variable to the system is the speed control command to the servo motor drive expressed as reference voltage Vref . The applied force on the object is the output variable Fapp. The deformable compliant rubber material covering the contact area of the fingers, as shown in figure 4, is important to allow a wide range of force control for solid objects as well as decreasing the pressure on the object and increasing the friction. Hence, we need to model the variation of the applied force Fapp by

Experimentally, and due to the mechanism constraint according to the gripper design, the applied force by the gripper fingers Fapp on the objects could not decrease if the reference voltage control command Vref is decreased. To verify the proposed controller, a model was built using MATLAB software package considering the mechanical constraints, which in turn lead to the accumulation of the applied force when Vref is changed. For practical control, a maximum limit was set to the applied force Fmax.app , figures 5 & 6. From this simulation model, the set of training data, checking data and testing data to be used for

the gripper finger with time at different reference voltage control commands Vref .

"BLD5604-SH2P" .

during gripping.

**6. Robotic gripper modeling** 

Fig. 4. Gripper configuration.

ANFIS model training were prepared.
