Experimental verification of design automation methods for robotic finger

(1)

Experimental verification of design automation

methods for robotic finger

Mohammadali Honarpardaz, Mehdi Tarkian, Johan Ölvander and X. Feng

The self-archived postprint version of this journal article is available at Linköping

University Institutional Repository (DiVA):

http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-139386

N.B.: When citing this work, cite the original publication.

Honarpardaz, M., Tarkian, M., Ölvander, J., Feng, X., (2017), Experimental verification of design

automation methods for robotic finger, Robotics and Autonomous Systems, 94, 89-101.

https://doi.org/10.1016/j.robot.2017.04.011

Original publication available at:

https://doi.org/10.1016/j.robot.2017.04.011

Copyright: Elsevier

(2)

Robotics and Autonomous Systems

Experimental Verification of Design Automation Methods for

Robotic Finger

M. Honarpardaz

a, b

_{, M. Tarkian}

b

_{, J. Ölvander}

b

_{, X. Feng}

a

a _{ABB, Corporate Research Center, Forskargränd 7, Västerås 721 78, Sweden} b _{Linköping University, Division of Machine Design, Linköping 581 83, Sweden}

A R T I C L E I N F O Article history:

Received 17 January 17

Received in revised form 00 March 00 Accepted 00 May 00 Keywords: Grippers Fingers Design automation Robotics Experimental method A B S T R A C T

Design automation of industrial grippers is a hot research topic for robot industries. However, literature lacks a standard experimental method to enable researchers to validate their approaches. Thus, this paper proposes a generic experimental method to verify existing finger design approaches. The introduced method is utilized to validate the methods Generic Automated Finger Design (GAFD), Manually Designed Fingers (MDF) and the eGrip tool. Experimental results are compared and the strengths and weaknesses of each method are presented.

1. Introduction

Robot fingers play a crucial role in success and performance of workcells as fingers are the only interfaces that connect the robot to the physical working environment. Fingers are responsible for grasping and manipulating workpieces without dropping or damaging them (Engelberger et al., 1980). Therefore, designing fingers to accomplish assigned tasks is tremendously complex and requires high skills in robotics and designing at the same time (Causey, 1999).

Today, there is an obvious trend toward products with short lifecycles. As a result, many robot industries have been focusing on enhancing the competiveness of robotic automation in the agile market. SARAFun (SARAFun, 2016) and Factory-in-a-day (Factory-in-a-day, 2016) are two large European Commission projects which are formed to enable a non-expert user to integrate a robot system for an assembly task in one single day. Currently, functional fingers industrial grippers (e.g. parallel-jaw) are designed manually, a process that requires several exhaustive and time consuming trial and error iterations even for highly skilled specialists. The average iteration time is about three to four working days and the total time for designing fingers can amount to around two weeks depending on the complexity requirements.

The present iterative procedure of manual finger design is unable to fulfill the demands of “burst” production (i.e. ramp up to full volume in very short time, run production for 3-12 months, and then change to produce a new product). Thus, finger design automation has been increasingly attracting the attention of the robot industry. However, very few researchers have been studying finger design automation and unfortunately no one has validated the proposed approaches with a generic experimental method (Honarpardaz et al., 2017). In earlier work (Honarpardaz et al., 2016), Generic Automated Finger Design (GAFD) is proposed as a general approach to overcome drawbacks of the existing methods.

To this end, this paper proposes a generic experimental method in order to validate and benchmark GAFD. This work aims to encourage future studies to use the proposed generic experimental methods or further improvements of it, in order to enable scholars to verify their proposed frameworks and to be able to compare results. The GAFD method is benchmarked against manually designed fingers by specialists, and relevant and available finger design automation methods.

The remainder of the paper is divided into sections as follows: Relevant Work section reviews the related works and the Method section describes the utilized methodology. Results are presented in the Result section, and the stability and performance of the fingers designed by GAFD and existing approaches are compared in the Discussion.

(3)

2. Relevant Work

Although the importance of experimental verification has been highlighted by several scholars in the finger design research domain, very few studies have physically validated their proposed approaches (Balasubramanian et al., 2012; Kim et al., 2013; Weisz and Allen, 2012). One of the major reasons is the difficulty of accessing expensive robotic hardware such as manipulator, controller, vision system, etc. In addition, executing physical experiments requires high skills and knowledge in robot operation and experimental methodologies.

Unfortunately, almost none of the handful of studies that physically validated their approach, provide detailed information about the process of the experiment (Ala et al., 2015; Kim et al., 2013; Bone et al., 2008; Zhang and Goldberg, 2007; Velasco and Newman, 1998; Brown and Brost, 1997). As a consequence, it is impossible to replicate the experiments and compare the results. Therefore, this work tries to initiate a mainstream method to enable scholars to compare future studies in this field.

Antony (2014) points out important factors for conducting industrial experiments with quality characteristics. Quality characteristics, or response parameters, are features related to the requirements of an experiment. He encourages the following quality characteristics should be:

 Possible and simple to measure during the experiment.

 Continuous variables rather than Boolean variables with only two outcomes.  Measurable precisely and accurately with a correct method and equipment.

According to Antony (2014) and Costa et al. (2006), a methodology for design of experiments may involves four phases; planning, designing,

conducting and analyzing.

Planning Phase

Antony (2014) describes that selecting suitable responses for an experiment is critical to success of any experiment. Variables such as length, diameter, width, strength, viscosity etc. are generally better at providing information than attribute answers like pass/fail, yes/no and better/worse. The response value must be well formulated to provide valuable data for the aim of the experiment (Costa et al., 2006). In this phase, the methodology employed for interpreting and collecting experiment data should also be considered and selected to ease the designing phase of the experiment.

Designing Phase

This phase includes the following steps:

1. Designing experiments that concludes the defined response parameter. 2. Selecting the most appropriate designs for experiments.

3. Defining a design database which contains list of equipment, settings, order of running, etc.

4. Determining the principle of replication or iterations of experiment. Iterations of the experiment may eliminate certain deviations in the data that are caused by external parameters. By performing iterations, deviations in the data that are caused by external parameters may be eliminated and a statistical verification achieved (Antony, 2014).

Jørgensen et al. (2016) recommend to avoid complex experiments with large amounts of statistical tests. It is important to keep the design of the experiment simple and transparent and also utilize understandable variables.

Conducting Phase

In this phase, the experiment is conducted and results are gathered. Costa et al. (2006) recommends documenting a test plan initially in order to preparing for every essential step in the actual experiment. Antony (2014) mentions several significant considerations prior executing an experiment, such as environmental conditions, availability of materials, etc.

Analyzing Phase

After performing the experiment, the next phase is to analyze and interpret the results so a valid and correct conclusion may be made. There are certain objectives, according to Antony (2014), which can aid in the process:

 Clarifying the process variables that affect the mean process performance.  Obtaining the process variables that affects the viability of performance.

 Determining the relation between the quality of the results and number of iterations.

(4)

3. Proposed Method

This section describes the methodology that is utilized in this article to facilitate fair benchmarking of the proposed GAFD method against existing approaches in the robot finger design research field. As illustrated in Fig. 1, the proposed method begins by designing the fingers and measuring the total design process time, then stability of the fingers is measured by conducting force and torque experiments. In the next step performance of the fingers is evaluated by pick-and-place and assembly experiments. In the last step, the footprint of the fingers is measured.

Fig. 1: Flowchart of the proposed experimental method.

3.1. Design Process Time

The lead-time process of designing fingers plays an important role in comparing different finger design methods as the main purpose of design automation is to reduce the design lead-time. The design process time considers only the amount of time spend on designing fingers and it does not take the preparation and manufacturing time in to account.

3.2. Grasp Stability Verification

Grasp stability plays an essential role in the throughput of a robot workcell. Fingers with a more secure grasp can move workpieces with higher acceleration and deceleration, thus reducing cycle time and increasing throughput. The ideal finger design fully encompasses the workpiece (form-closure), yet it is infeasible in the most cases. Therefore, the stability of the grasp relies on upon friction (force-closure). In this work, two experimental methods are used to measure the stability of grasps. The first experiment measures the maximum disturbance force that grasps can resist without any slippage. The second method measures the maximum disturbance t that grasps can withstand before slipping.

3.2.1. Force Experiment

In order to determine the stability of grasps against disturbance forces, the maximum static friction-force at the contact point is required. According to Coulomb’s law of friction, static friction range acts up to the point of slipping and the direction of this frictional force always opposes the motion or impending motion. The magnitude of the frictional force can be described as,

𝐹𝑓≤ 𝑁. µ𝑠 (Eq. 1)

Where 𝐹𝑓 denotes the friction force, 𝑁 represents the normal force executed by fingers and µ𝑠is the static friction coefficient (see Fig. 2).

Fig. 2: Schematic diagram of force experiment.

In order to obtain the maximum resistant force (friction force), knowing the exact value of µ𝑠 is essential. However, determining coefficient of

friction in practice requires additional information (e.g. contact surface quality, contact temperature, etc.) which are usually not available. Therefore, experimentally measuring the maximum resistant force is critical for the verification of designed fingers (Causey, 1999). To obtain the maximum resistant force, an external force is exerted to the workpiece in X, Y and Z directions, and the magnitude of the external force is increased until the workpiece starts to slip between the fingers. The relation between external force and friction force is given as follows:

𝐹𝑒𝑥𝑡= − 𝐹𝑓 (Eq. 2)

Where 𝐹𝑒𝑥𝑡 denotes the external force. The maximum external force that the contact can resist just before slippage is considered as the stability of

(5)

3.2.2. Torque Experiment

As with the force experiment, knowing maximum static friction-force at the contact point is necessary to obtain the stability of the grasp against disturbance torques. As demonstrated in Fig. 3, the maximum resistant torque can be represented as,

𝑇𝑒𝑥𝑡= 𝐹𝑓. 𝑟 _{(Eq. 3)}

Where 𝑇_𝑒𝑥𝑡 denotes the external torque and 𝑟 is the distance between the contact point and the rotation center.

Fig. 3: Schematic diagram of torque experiment.

In order to determine the maximum torque, an external moment is applied to the workpiece in X, Y and Z directions, its magnitude is increased until the workpiece begins to slip between the fingers. Thus, grasp stability can be regarded as the maximum external torque that the contact can withstand without slippage.

3.3. Performance (repeatability) Verification

The recent introduction and acceptance of vision-based parts feeding systems and modular manufacturing concepts have considerably increased the expectations and requirements of grippers performance. Therefore, two of the most common applications of industrial robots, i.e.

pick-and-place and assembly, are selected in this work to verify the performance of the GAFD method and benchmark it against existing approaches. The

parameter that is measured during these experiments is the repeatability of the designed fingers. This response parameter is defined by the number of failures in a certain number of operations (i.e. pick-and-place and assembly).

3.3.1. Pick-and-Place Experiment

Pick and place is the most common operation that industrial robots are utilized for (International Federation of Robotics, 2016). In this work, a similar pick-and-place procedure to what Fantoni et al. (2013) and Hladowski et al. (2016) present is used to verify the performance of the fingers by measuring their repeatability. As shown in Fig. 4, the Fantoni et al. (2013) and Hladowski et al. (2016) procedure may be summarized as follows:

1. Approaching the workpiece which is placed on a known position from an initial robot pose. 2. Closing the gripper jaw until the fingers contact the workpiece.

3. Increasing the contact force. 4. Securing the grasp.

5. Lifting the workpiece and manipulating it over a finite time period. 6. Releasing the workpiece at a predefined position.

Fig. 4: Pick-and-place experiment procedure, inspired by Fantoni et al. (2013) and Hladowski et al. (2016).

3.3.2. Assembly

Experiment

Assembling multiple workpieces is a complicated tasks in robot automation. Thus, verification of the repeatability of the designed fingers in assembly operations helps to measure the performance of fingers in complex applications. The most common types of assemblies in industry are snap-fit, push-in, tilt-in and screw (Nof, 1999), as illustrated in Fig. 5.

(6)

Fig. 5: Schematic diagram of typical assembly types in industry.

Among these assemblies, screwing is the most complex, as it requires high accuracy to center workpieces along the screwing axis and high precision to perform the rotation and translation simultaneously. For this reason, a screw-fit assembly was used to verify the fingers in this study. Fig. 6 presents an example of screwing assembly and the experiment procedure may be stated as follows:

1. Approaching workpiece A, which is placed in a known position from an initial robot pose. 2. Closing the gripper jaw until the fingers are in contact with the workpiece.

3. Increasing the contact force. 4. Securing the grasp.

5. Lifting the workpiece and manipulating it over to workpiece B located in a known position. 6. Centralizing workpiece A and workpiece B.

7. Rotating and translating workpiece A toward workpiece B. 8. Releasing workpiece A.

Fig. 6: Assembly experiment procedure.

3.4. Footprint

Another essential factor that should be considered when comparing different finger design methods is the footprint of the fingers. This is the required area around the workpiece to enable the fingers to grasp the workpiece without any collision. Finger footprint plays a significant role in determining the throughput of a robot workcell. The smaller the footprint, the greater the number of parts that can be fed to the system, and the higher the throughput. Fig. 7 illustrates the methodology used in measuring finger footprint. To measure the footprints, the height of the workpiece (H) is determined and projection of the finger at H is obtained (solid red line). The area of the projection is considered as the footprint.

(7)

4. Case Study

In this section, existing methods for designing robot fingers are described and then utilized to generate fingers. The fingers are then used as test objects for the experimental method introduced in the previous section.

4.1. Existing Finger Design Approaches

There are two general approaches available for generating robot fingers, manually designed fingers (MDF) and automatically designed fingers. According to a recent comprehensive review on finger design automation for industrial robots, existing approaches may be classified as modular design, re-configurable design and customized design (Honarpardaz et al., 2017). Customized design approaches, unlike the other ones, provide a dedicated solution for every workpiece. Furthermore, these methods have high design flexibility and reliability in comparison with modular design and re-configurable design approaches. While the customized design methods are highly generic and are considered most practical for automation of the finger design process, few scholars attempt to use these methods in finger design automation (Velasco et al., 1996; Velasco and Newman, 1998; Velasco et al., 1998; Pedrazzoli et al., 2001). Velasco and Newman (1998) introduce an algorithm which designs fingers to fit and envelop the workpiece surfaces. The method begins by assuming fingers to be solid blocks. Then the geometry of the object is subtracted from the blocks to build the customized fingers. Recently, a commercial tool, eGrip (eGrip, 2016), was launched by SCHUNK (SCHUNK GmbH & Co. KG, Lauffen am Neckar, Germany) which follows a similar approach as the one Velasco and Newman (1998) propose.

Even though the existing customized design approaches are computationally inexpensive and can handle complex geometries, they are unable to provide a solution for workpieces with a specific geometrical property (e.g. axi-symmetric). Besides, these approaches are incapable of designing fingers with internal grasp which is contacting with inner geometry of the workpiece (e.g. a hole) and a motion that is directed outward. This type of grasp is crucial in some applications.

As a result, the GAFD method is proposed to overcome the downsides of the existing customized finger design approaches. This method consists of connecting several key processes (i.e. grasp planning, grasp analysis, finger design and collision detection) which are essential for designing robust fingers. The proposed method begins by generating a point cloud model of the workpiece using its Computer Assisted Design (CAD) model. Then possible grasp sets are determined and analyzed in the grasp synthesis and analysis stage to find grasp sets which have the highest quality (stability) against resistant wrenches. Once the grasp set with the highest quality is known, fingers are generated based on the position of the grasp sets on the surface of the workpiece. This method designs the fingers in such a way that fingertips imitate the surface contour of the workpiece at the location of contact. To ensure feasibility of the designed fingers, the collision detection section of the algorithm checks the possibility of unwanted collisions between the fingers and the workpiece (Honarpardaz et al., 2016).

4.2. Case Studies

As shown in Fig. 8, four different industrial components with complex geometries are selected to assess fingers stability, and Table 1 shows designed fingers for the selected workpieces. Columns of the table present 3D model of the left and right fingers using different methods (rows). All three design methods managed to design fingers for the lamp cap (a) and lamp base (b), while the eGrip tool was unable to propose a design solution for the clip (c) and board (d). The clip necessitates internal grasping and the tool is incapable of designing fingers with internal grasps. Further, the large external dimensions (50x45) of the board prevent the tool from designing fingers that envelop the geometry. In this study, fingers are designed for ABB Smart Hand yet any robot gripper may be used.

(8)

Table 1: Designed fingers using GAFD, MDF and eGrip methods.

Once designed, all fingers are manufactured. Many of the conventional manufacturing methods (e.g. injection moulding, machining, casting, etc.) may be utilized to produce robot fingers. However, additive manufacturing methods have recently become interesting for robot industries due to their short time and low cost production (Velasco et al., 1996). Thus, in this work, fingers are manufactured by additive manufacturing methods using acrylonitrile butadiene styrene (ABS) plastic on a Stratasys U-Print SE Plus 3D printer with solid material fill, and 0.25 mm layer resolution. Table. 2 presents the manufactured fingers designed by GAFD, MDF methods and the eGrip tool for handling the selected industrial components. Table 2: Manufactured fingers using the 3D printer.

(9)

5. Result

The presented experiments in this work are executed based on the process introduced by Antony (2014) and Costa et al. (2006) which consists of four phases; planning, designing, conducting and analyzing. In this study, the experimental verification is conducted based on the experiments introduced in section 3 to measure the design process time, grasp stability, performance and footprint of manually designed fingers (MDF) by specialists at ABB and fingers designed automatically using the GAFD method and the eGrip tool (eGrip, 2016). The latest product of ABB Robotics, YuMi, is used as a platform in this investigation. However, any robot platform may be used for replication of the stated experiments. Details of the implementation and execution of the experiments are provided in Appendix A and the results of the experiments are presented in the following sections.

The total time and number of iterations required for designing functional fingers for grasping and manipulating the lamp cap and lamp base components are presented in Table 3. As the MDF uses a trial and error based approach, it takes three and four iterations to design functional fingers for the lamp cap and lamp base, respectively. However, GAFD and eGrip methods generated fingers in the first iteration.

Table 3: Lead-time and number of iteration for designing the lamp cap and lamp base using MDF, GAFD and eGrip methods.

Lamp Cap Lamp Base

Design lead-time [min] No. iterations Design lead-time [min] No. iterations GAFD 23 1 38 1 MDF 3360 3 4320 4 eGrip 16 1 17 1

As stated in the method section, two experiments (i.e. force and torque) are defined to measure the stability of three different sets of designed fingers. Each set of fingers is examined in X, Y and Z direction to measure the stability of the fingers in three-dimensional space. Fig. 9 illustrates the reference coordinate system used in these experiments. The results of force and torque experiments on the designed fingers are presented as follows.

Fig. 9: Schematic diagram of the reference coordinate system

5.2.1. Force Experiment

This experiment is executed in order to measure the maximum resistant force [N] in X, Y and Z directions at the threshold of slippage as the workpiece is grasped by the fingers. Therefore, an external force is applied to the workpiece in a certain direction while it is grasped by the designed fingers and then its magnitude is gradually increased until the workpiece starts to slip. As shown in Fig. 10, the experiment setup consists of a robot arm, force sensor, analog-to-digital convertor, cable, spring and a test control computer. Cables are used to connect the sensor to the grasped workpiece, and the spring is used to prevent impact forces on the workpiece. The signal convertor is setup between the sensor and the computer to enable collecting the results. Appendix A presents details of the force experiment.

(10)

Fig. 10: Setup of the force experiment.

It should be noted that the slippage occurs only in directions perpendicular to the force closure direction of the grasp. The directions for which the workpiece is in form-closure have not been examined in this experiment due to the fact that a very large amount of force is required to move the workpiece between the fingers and this force may damage the robot hardware (i.e. gripper and arm). To be able to properly compare the stability of the designed fingers, the maximum force that the robot arm or gripper can resist without any damage is used for directions with form-closure grasps. In this case, the maximum allowed force on the robot gripper used for form-closure grasps is 20 N.

To reduce possible external errors caused by operator and equipment, five iterations for each experiment are conducted. The mean values of iterations and standard deviation of each experiment in X, Y and Z directions are presented in Fig. 11.

Fig. 11: Results of the force experiment in X, Y and Z direction.

5.2.2. Torque

Experiment

In this experiment, the maximum disturbance torque that can be exerted on the workpiece in X, Y and Z direction, without sliding between fingers is measured. Similar to the previous experiment, an external torque is applied to the workpiece grasped by the fingers, and then magnitude of the external torque is gradually increased until the workpiece starts to slip between the fingers. As presented in Fig. 12, setup of this experiment is similar to the force experiment yet workpieces are directly connected to the torque sensor. Detail of the experiment setup is described in Appendix A.

(11)

Fig. 12: Setup of the torque experiment.

As mentioned in the force experiment section, the stability of the grasp against external toque is assessed only for directions in which the workpiece is in force closure. In case of form-closure grasps, the maximum torque that the robot hardware can withstand is considered as the maximum resistant torque (0.5 [N.m]). Fig. 13 demonstrates the average value of five iterations of each experiment in X, Y and Z directions.

Fig. 13: Results of the torque experiment in X, Y and Z direction.

The results of pick-and-place and assembly (i.e. screwing) experiments are presented as follows.

5.3.1. Pick

and

Place

Experiment

In this experiment, fingers pick workpieces from the feeder tray and place them in the dedicated workpiece container. Fig. 14 demonstrates the setup that is utilized to examine the performance of the fingers. Positions of feeders and containers are known to the robot. For each set of fingers, 100 iterations are conducted to minimize the possible external errors. Table 4 illustrates the repeatability of the fingers using different approaches.

(12)

Fig. 14: Setup of the pick-and-place experiment. Table 4. Results of the pick-and-place experiment.

No. of Success/Total Iterations

GAFD 100/100 100/100

MDF 100/100 100/100

eGrip 99/100 97/100

5.3.2. Assembly Experiment

In the assembly experiment, the fingers are utilized to execute a screw-assembly using the lamp cap and lamp base components. Fig. 15 demonstrates the sequences for executing this assembly and the setup of the experiment. In keeping with the previous experiment, 100 experiment-iterations are executed for each set of fingers to reduce possible external errors. As the repeatability of each set of fingers is presented in Table 5, GAFD, MDF, and eGrip accomplished the task with, 99, 100 and 86 percent repeatability respectively.

(13)

Fig. 15: Setup of the assembly experiment. Table 5: Results of the assembly experiment.

No. of Success/Total Iterations

GAFD 99/100

MDF 100/100

eGrip 86/100

5.4. Footprint

To be able to compare the effect of the footprint on the throughput of the workcell, spatial efficiency which is the maximum number of workpieces that can be fed to the system in a matrix arrangement with a certain area (200x200 mm) is determined. Table 6 demonstrates the matrix feeder with maximum feeding capacity for the lamp cap and the lamp base, respectively. As demonstrated in Table 7, the lam cap feeder trays for fingers designed by GAFD and MDF contain 30 workpieces while the tray for eGrip fingers has 25 workpieces. The MDF, GAFD and eGrip can feed 36, 30 and 20 lamp bases per tray respectively.

(14)

Table 7: Footprint and maximum number of workpiece per feeder tray for fingers designed using MDF, GAFD and eGrip methods.

Footprint [mm2_] Max No. of

Workpieces/Tray

Footprint

[mm2_] _{Workpieces/Tray}Max No. of

GAFD 1055 30 925 30

MDF 1050 30 573 36

eGrip 1472 25 1536 20

6. Discussion

In this section, first, limitations of proposed experimental methods are discussed. Following this, the results of the design process time, grasp stability, performance verification and footprint experiments utilized GAFD method are compared with the other existing methods in the field of finger design research and strengths and weaknesses of each method are presented.

According to the process lead-time of designing fingers using different methods presented in section 5.1, manually designed fingers have the longest design lead-time which is due to the many trial-and-error iterations. One should notice that the time for producing fingers has been deducted from the design lead-time. In addition, the design process time for the MDF method presented here is accumulated from a single internal source at ABB and may differ from other robot industries. Moreover, no information regarding design process time of MDF are published, yet an obvious trend in this research area is established to develop finger design automation commercial tools (e.g. eGrip) in order to reduce the process time (Velasco et al., 1998; eGrip, 2016; SARAFun, 2016). Therefore it can be deduced that MDF is slow enough to justify the field of automated finger design automation.

Both automated finger design methods, GAFD and eGrip, managed to design functional fingers with only one iteration. However, the eGrip tool has shorter design process time in comparison to GAFD methods, yet the short difference in the design time (11 minutes) has limited implications.

The results illustrate that the designed fingers using each method generate grasps with high stability in different directions. Therefore, a score based method is utilized to allow a fair comparison of the fingers. The fingers with the highest and second highest resistant force or torque are respectively scored with two and one points; and obviously the fingers with the lowest resistant force or torque get no point. Summary of the experimental results is displayed in Table 8.

Based on the results, the manually designed fingers have the highest grasp stability in all case studies. And between the automatic finger generator methods, fingers designed using GAFD have the highest grasp quality for the Lamp cap and the eGrip tool designed fingers with highest stability for the lamp base. Therefore, both automated finger design methods produced fingers with highest grasp stability for one case study each. However, the eGrip tool could not design fingers for two of the selected components.

Table 8: Summary of the grasp quality experiments.

Fx Fy Fz Mx My Mz Total Point Value [N] Score Value [N] Score Value [N] Score Value [N.m] Score Value [N.m] Score Value [N.m] Score Lamp Cap GAFD 20 2 20 2 5.32 1 0.5 2 0.5 2 0.131 1 10 MDF 20 2 20 2 20 2 0.5 2 0.5 2 0.5 2 12 eGrip 20 2 20 2 2.09 0 0.5 2 0.5 2 0.043 0 8 Lamp Base GAFD 20 2 20 2 4.25 0 0.5 2 0.5 2 0.045 0 8 MDF 20 2 20 2 4.78 2 0.5 2 0.5 2 0.5 2 12 eGrip 20 2 20 2 4.29 1 0.5 2 0.5 2 0.074 1 10 Clip GAFD 2.74 1 20 2 2.51 1 0.5 2 0.029 1 0.5 2 9 MDF 3.52 2 20 2 3.38 2 0.5 2 0.030 2 0.5 2 12 eGrip 0 0 0 0 0 0 0 0 0 0 0 0 0

(15)

Board

GAFD 20 2 20 2 4.84 2 0.5 2 0.5 2 0.057 1 11

MDF 20 2 20 2 2.74 1 0.5 2 0.5 2 0.5 2 11

eGrip 0 0 0 0 0 0 0 0 0 0 0 0 0

To measure the repeatability of designed fingers using GAFD, MDF and eGrip methods, two sets of experiments are conducted based on a pick-and-place and an assembly application.

As presented in Table 4 and Table 5, both GAFD and MDF methods managed to execute the defined pick-and-place task with a success ratio of 100 %, while fingers designed using eGrip have 99 and 97 successful operations for lamp cap and lamp base, respectively. The failures arising using eGrip fingers may be due to the design approach of this tool. As this tool generates fingers by subtracting the geometry of workpiece from a set of initial finger blocks, the designed fingers can only fully envelop workpieces with the exact dimensions of the CAD model that was provided as input to the eGrip design tool. In other words, the design approach of the eGrip tool mimics every small detail on the surface of the workpiece which causes fingers to have difficulty fully enveloping the workpiece if any small variation occurs in the dimensions of the workpiece. As a result, the designed fingers by eGrip were unable to secure the grasp and pick the workpiece in some cases.

Based on the results of the assembly experiment, MDF accomplished the screwing assembly with 100 % success ratio and the fingers designed by GAFD method executed the task with 99 % repeatability. The 1 % failure was due to a lower tolerance factor compared to MDF. The fingers designed using the eGrip tool completed the task with 86 % repeatability. The main reason for the relatively high failure ratio using eGrip fingers is due to misalignment of the components in space upon execution of the screwing task. One possible explanation would be that the components slide between the fingers just before they make contact for screwing.

6.4. Footprint

Based on the results presented in Table 7, the manually designed fingers can have the maximum spatial efficiency. After MDF, the fingers designed by GAFD and eGrip methods have respectively smaller footprints and consequently higher spatial efficiencies. The higher spatial efficiency, the higher is the throughput of the workcell. Therefore, GAFD method and MDF fingers can significantly increase the throughput of the workcell in comparison to the fingers design by eGrip tool.

6.5. Experimental Method

In the experimental methods proposed for stability verification of fingers, external sensors are utilized to measure force and torque applied to the workpieces. Some robot manipulators have sensors built into their structures which would facilitate the experiment directly without introducing external errors. Another alternative is attaching fingers to a fixture to hold the workpiece during force and torque experiments. This may be useful for cases where robot hardware is not available. In the conducted study the experiments are conducted manually, both in terms of applying the load and recording the maximum torque. Hence there is a possibility to improve and automate the experimental process by for example using a sort of tensile testing machine. However, as can be seen by the statistical bars in figures 11 and 13, the accuracy and repeatability are more than sufficient for the assessment of the different fingers. In addition, the method proposed for performance verification is based on screwing, which is one of the most complex, yet most common, assembly methods in industry. On the other hand, screwing is one of many alternative assembly techniques, therefore, the result may not be generalized to all applications in reality. Furthermore, the main focus of the proposed methods is verifying the stability and performance of fingers for industrial parallel-jaw grippers. As a result, further investigations on applicability of the method to other types of grippers is necessary.

7. Conclusion

This paper presents a generic experimental method for benchmarking methods for industrial robot finger design. Even though automatic finger generation is a hot topic, there is a lack of formal experimental methods to validate the proposed methods. The experimental method proposed in the paper is used to validate and benchmark the methods generic automated finger design (GAFD), and manually designed finger (MDF) as well as the tool eGrip. The proposed experimental method can enable scholars to validate their future approaches in the finger design research area and benchmark them against existing methods.

The experimental methods consist of stability and performance verification of fingers which measure the following qualities:

 Design process time: the lead-time for designing the fingers.

 Grasp stability: the maximum disturbance force and torque that fingers can resist without any slippage.

 Repeatability (performance): the ratio of number of successfully accomplished assembly tasks.

 Footprint: the required area around the workpiece to enable the fingers to grasp the workpiece without any collision.

According to the comparison outlined in the discussion section, manually designed fingers (MDF) provide the highest grasp quality, repeatability and spatial efficiency in comparison to GAFD and eGrip methods. On the other hand, this method has the longest design process time which makes it impractical for product development which follow agile processes.

(16)

Among the considered automated methods, the shortest design process time is for the eGrip tool. Furthermore, this tool produces fingers with similar grasp quality to GAFD method. However, eGrip tool has the least grasp quality, performance and spatial efficiency. Thus, this tool would be effective for simple pick-and-place applications.

Following MDF closely, fingers designed by GAFD deliver highest grasp quality, repeatability and spatial efficiency. Also, this method has a very fast design process (a reduction of two orders of magnitude) in comparison to MDF with only small reduction in stability and performance. As a consequence, this method is suitable for conventional assembly applications as well as pick-and-place applications.

While it has been shown that MDF is better than other methods and applicable to all applications, the proposed GAFD method can provide a foundation for designers to manually improve the fingers. Meaning that the design process can start with GAFD in order to quickly provide design alternatives to the engineer which then can proceed to enhance and optimize the design with MDF.

Based on the conclusions drawn it is obvious that MDF can handle more complex requirements and thus the finger design automation community needs to investigate new approaches that takes complex manipulation (e.g. in-hand manipulation, re-grasp, multi-functional fingers etc.) into consideration.

Acknowledgements

The research leading to these results has received funding from the European Community’s Framework Programme Horizon 2020 – under Grant Agreement No. 644938 – SARAFun. Mikael Hedelind and Jonas Larsson at ABB are gratefully acknowledged for technical discussions and various supports. Special thanks to Joakim Elf and Rasmus Sjögren for improving the GAFD algorithm from various aspects and implementing the algorithm on various objects in order to verify the proposed approach.

Reference

Ala, R., Kim, D.H., Shin, S.Y., Kim, C., Park, S.-K., 2015. A 3D-grasp synthesis algorithm to grasp unknown objects based on graspable boundary and convex segments. Inf. Sci. 295, 91–106. doi:10.1016/j.ins.2014.09.062

Antony, J., 2014. Design of Experiments for Engineers and Scientists, Second. ed. Elsevier.

Balasubramanian, R., Xu, L., Brook, P.D., Smith, J.R., Matsuoka, Y., 2012. Physical Human Interactive Guidance: Identifying Grasping Principles From Human-Planned Grasps. IEEE Trans. Robot. 28, 899–910. doi:10.1109/TRO.2012.2189498

Bone, G.M., Lambert, A., Edwards, M., 2008. Automated modeling and robotic grasping of unknown three-dimensional objects, in: IEEE International Conference on Robotics and Automation, 2008. ICRA 2008. Presented at the IEEE International Conference on Robotics and Automation, 2008. ICRA 2008, pp. 292–298. doi:10.1109/ROBOT.2008.4543223

Brown, R.G., Brost, R.C., 1997. A 3D modular gripper design tool, in: , 1997 IEEE International Conference on Robotics and Automation, 1997. Proceedings. Presented at the , 1997 IEEE International Conference on Robotics and Automation, 1997. Proceedings, pp. 2332–2339 vol.3. doi:10.1109/ROBOT.1997.619310

Causey, G.C., 1999. Elements of agility in manufacturing.

Costa, N.R.P., Pires, A.R., Ribeiro, C.O., 2006. Guidelines to help practitioners of design of experiments. TQM Mag. 18, 386–399. doi:10.1108/09544780610671057

eGrip, 2016. eGrip [WWW Document]. URL http://www.egrip.schunk.com/Account/LogOn?ReturnUrl=%2f (accessed 12.8.15). Engelberger, J.F., Lock, D., Willis, K., 1980. Robotics in Practice: Management and Applications of Industrial Robots. AMACOM. EU Project: Factory-in-a-day, 2016.

Fantoni, G., Gabelloni, D., Tilli, J., 2013. Concept design of new grippers using abstraction and analogy. Proc. Inst. Mech. Eng. Part B J. Eng. Manuf. 227, 1521–1532. doi:10.1177/0954405413489967

H2020 SARAFun EU project, 2016.

Hladowski, L., Galkowski, K., Nowicka, W., Rogers, E., 2016. Repetitive process based design and experimental verification of a dynamic iterative learning control law. Control Eng. Pract. 46, 157–165. doi:10.1016/j.conengprac.2015.10.007

Honarpardaz, M., Tarkian, M., Feng, X., Sirkett, D., Ölvander, J., 2016. Generic Automated Finger Design V05BT07A071. doi:10.1115/DETC2016-60514 Honarpardaz, M., Tarkian, M., Ölvander, J., Feng, X., 2017. Finger design automation for industrial robot grippers: A review. Robot. Auton. Syst. 87, 104–

119. doi:10.1016/j.robot.2016.10.003

International Federation of Robotics, 2016. URL http://www.worldrobotics.org/# (accessed 12.5.16).

Jørgensen, M., Dybå, T., Liestøl, K., Sjøberg, D.I.K., 2016. Incorrect results in software engineering experiments: How to improve research practices. J. Syst. Softw. 116, 133–145. doi:10.1016/j.jss.2015.03.065

Kim, J., Iwamoto, K., Kuffner, J.J., Ota, Y., Pollard, N.S., 2013. Physically Based Grasp Quality Evaluation Under Pose Uncertainty. Trans Rob 29, 1424– 1439. doi:10.1109/TRO.2013.2273846

Nof, S.Y. (Ed.), 1999. Handbook of Industrial Robotics, 2 edition. ed. Wiley, New York.

Pedrazzoli, P., Rinaldi, R., Boer, C.R., 2001. A rule based approach to the gripper selection issue for the assembly process, in: Proceedings of the IEEE International Symposium on Assembly and Task Planning, 2001. Presented at the Proceedings of the IEEE International Symposium on Assembly and Task Planning, 2001, pp. 202–207. doi:10.1109/ISATP.2001.928990

Velasco, J., V.B., Newman, W.S., 1998. Computer-assisted gripper and fixture customization using rapid-prototyping technology, in: 1998 IEEE International Conference on Robotics and Automation, 1998. Proceedings. Presented at the 1998 IEEE International Conference on Robotics and Automation, 1998. Proceedings, pp. 3658–3664 vol.4. doi:10.1109/ROBOT.1998.681393

Velasco, V.B., Jr., Newman, W.S., 1996. An Approach to Automated Gripper Customization Using Rapid Prototyping Technology. Velasco, V.B., Newman, W.S., Zheng, Y., Choi, S., 1998. Automated Gripper and Fixture Customization via Rapid Prototyping.

Weisz, J., Allen, P.K., 2012. Pose error robust grasping from contact wrench space metrics, in: 2012 IEEE International Conference on Robotics and Automation (ICRA). Presented at the 2012 IEEE International Conference on Robotics and Automation (ICRA), pp. 557–562. doi:10.1109/ICRA.2012.6224697

(17)

Appendix A: Details of experiments setup

A.1. Experiment Conduction:

Based on the conducting phase describes in the previous section, conducting of the experiments and required equipment are presented here.

A.1.1. Stability Verification Experiment Setup

As shown in Fig. 10 and Fig. 12, the following equipment are used to measure the resistant force and torques of the fingers designed by GAFD, MDF and eGrip methods:

• Torque/force sensor (MAGTROL SA – TMB 306/411) measures the torque resistance with analog voltage signal as output (see Fig.

A1 (a)).

• Analog-to-digital convertor (PicoScope 2000) that converts the analog signal from the torque sensor to a laptop through a USB

connection (see Fig. A1 (b)).

• Robot (YuMi) is used to grasp the workpiece (see Fig. 14).

• Work table enables an attachment of equipment on it (see Fig. 12).

• Spring: Adjusting the component through a pull force (will only be used for the force experiment). The spring is attached between the

component and the sensor with the purpose to give a certain elasticity to the pull force and prevent impact forces (see Fig. 12).

• Cables are used to attach the component to the sensor (see Fig. 12).

• Computer with the following software and specification:

–Robot Studio (6.03) is a software delivered from ABB to be able to control the robot. Robot Studio provides offline programming that feeds instructions to the robot. These instructions may be actions like moving, grabbing or manipulating depending on the setup and type of robot. The program will provide the pick-and-place and assembly operations during the experimental verification. – PicoScope 6: The software displays the voltage resistance as a continuous graph with parameters voltage over time. – Intel i7 at 2.4 GHz and 8 GB of RAM.

Fig. A1: a) PicoScope 2000; b) Torque/force sensor

In the experiment, the robot arm and gripper are used as fixtures that enable the fingers to grasp the workpieces. The external force and torque are applied to the workpiece by manually applying a torque on the input shaft of the torque sensor. The output shaft is directly connected to the workpiece for the torque experiments, whereas for the force experiments a wire is attached to a drum to obtain a linear force. While the robot arm could be utilized to exert the external torque and force, it increases the complexity of the experiment as uncertainties in accuracy and repeatability of the manipulator should be taken in to consideration in execution and analysis of the experiment. Fig. A2 demonstrates an example of the measured force as a function of time for the stability verification experiment. According to Eq. 1, the maximum friction force that an object can withstand is precisely when the object starts moving (slippage). Therefore, the maximum external load (i.e. the peak) in Fig. A2 illustrates the maximum resistant load.

(18)

Fig. A2: An example of output data of the torque experiment

A.1.2. Performance Verification Experiment Setup

The equipment that are used to examine the quality and performance of the designed fingers are as follow:

• Robot (YuMi) is used as a platform (see Fig. 14).

• Feeder that provide workpieces to the system (see Fig. A3).

• Container are utilized to collect picked/assembled workpieces (see Fig. A3).

Fig. A3: Schematic setup of the performance verification experiment.

Appendix B: Additional Experiment Results

(19)

Fig. B1: Pick-and-place performance experiment using fingers designed by a) GAFD, b) MDF and c) eGrip methods.