Scattering or Pushing for Object Singulation in Cluttered Environment: Case Study with Soma Cube

Round 1

Reviewer 1 Report

Dear authors
The paper presents a hybrid method for singulation of blocks. The duration time of the task has been chosen as the performance index for this algorithm. 
The topic is interesting and the paper is structured very well. 
The previous version of this paper missed discussion over reliability of this algorithm.
The authors have elaborated on this issue and have covered it.

Now the results and discussions are supporting enough the goal of the paper. 

With kind regards

Author Response

Thankyou for your kind remarks. The Table 1 as been further improved by adding coefficient of variation. English language and spelling errors are corrected. 

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper mentions a virtual world simulator in section 4 which has been used in the paper to simulate the singulation process. But the authors don’t provide any data that indicate the efficiency of the simulator with regard to the real world experiments. Does it provide sufficient and/or accurate information regarding the singulation process? This simulator doesn’t seem to be well connected to the rest of the experiments.

Furthermore, regarding the experiments, the number provided seems very small to assess the efficiency of the proposed methodology (section 5). What do the standard deviation numbers provided in Table 1 stand for? How do you interpret those values? A more meaningful would be the coefficient of variation.

Have the authors considered a certain application, for example, one where space is tight? Would the same algorithm be applied?

Comments for author File: Comments.pdf

Author Response

[Comment 2-1] The paper mentions a virtual world simulator in section 4 which has been used in the paper to simulate the singulation process. But the authors don’t provide any data that indicate the efficiency of the simulator with regard to the real world experiments. Does it provide sufficient and/or accurate information regarding the singulation process? This simulator doesn’t seem to be well connected to the rest of the experiments.

[Answer 2-1]

Thank you for pointing it out. The efficiency of the simulator has already been discussed and published in our previous work [1]. Physics engines like DAFUL and ADAMS are being widely used to perform virtual world experimentation. In this paper, a virtual world simulations have been done using a physics engine: a commercially available multi-body simulation software called ‘Daful’. These virtual world simulations have been performed by integrating the estimated physical properties such as mass ratios, coefficient of restitution and coefficient of frictions to perform the control singulation of objects. The colliding velocity of robot and direction of collision is updated in virtual world by employing a feedback routine that would minimize the error to achieve the desired singulation pattern. The simulator provides accurate information for singulation process. Once, the minimum distance among object is achieved in virtual world, the same velocities information is used in real world to perform the object singulation and all the experimentation is based using the same initial velocity information from the virtual world simulator. It is mentioned in section 4.3 and page number 9 of the manuscript.

[Comment 2-2] Furthermore, regarding the experiments, the number provided seems very small to assess the efficiency of the proposed methodology (section 5). What do the standard deviation numbers provided in Table1 stand for? How do you interpret those values? A more meaningful would be the coefficient of variation..

[Answer 2-2]

Thank you for pointing it out. We added data related to two more experiments for each case. In our application we have used Soma cube blocks which consists of 7 blocks. In order to validate our approach, we have started experimentation using three, four and 7 blocks with seven experiments for each case and it clearly shows that when full Soma cube with seven blocks is used, our hybrid approach is less time consuming as compared to rest of approaches for singulation.  We have modified the table and added the coefficient of variation in the table. In the previous Table 1 the time for one time scattering approach is constant and as a result the standard deviation for one time scattering approach is zero. However, in case of pure pushing and hybrid approach the time for each experiment is different. Standard deviation shows the amount of variation and dispersion of a set of values in the table.

[Comment 2-3] Have the authors considered a certain application, for example, one where space is tight? Would the same algorithm be applied?

[Answer 2-3]

Thank you for pointing it out. In the current case, we mostly considered an application where the blocks are tightly placed together inside workspace. In the case that the workspace is reduced for a different application, then our algorithm still works since it can use only pushing for singulation because scattering may scatter the blocks out of the workspace. This same algorithm having hybrid approach actually helps in singulation for variety of different cases by utilizing both scattering or pushing or any one of them based on the desired application. 

[1].  A. Imran, S.-H. Kim, Y.-B. Park, I. H. Suh, and B.-J. Yi, "Singulation of Objects in Cluttered Environment Using Dynamic Estimation of Physical Properties," Applied Sciences, vol. 9, no. 17, p. 3536-3554, 2019.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The authors have provided suitable answers to previous mentioned issues.

I would still suggest to comment in the paper on the standard deviation and the coefficient of variation, to show the relevance of those numbers relative to the application

Author Response

Dear Reviewer

Thanks for your comment.

We added the following comment in the text.

The standard deviation and coefficient of variation for pure pushing were 2.69 and 8.88 respectively. The standard deviation and coefficient of variation for hybrid approach were 2.13 and 31.84 respectively. It is observed that hybrid approach has less standard deviation but that coefficient of variation is larger. That implies that pure pushing is more consistent in terms of task completion time, though it took more time for singulation of 7 objects.

Sincerely,

Byung-Ju Yi

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

It is an interesting paper that proposes a different (hybrid) technique for achieving objects singulation. It combines dynamic scattering and pushing alternatively. I would suggest a better title to express the final goal of the paper.

The paper follows closely another paper authored by almost same authors, reference 15, which is why the two papers have a close relation.

Referring to fig. 2, I would change the symbol (square) in the case where a decision needs to be taken (Desired Singulation) to a more traditional symbol: the diamond block. In the same figure prehensile manipulation is achieved before the non-prehensile one. Is it correct? Can it always be achieved?

What is a virtual world simulator? Did you mean virtual reality simulator? The authors apparently have developed such a simulator, is it possible to share the architecture and its main components (modules)?

I have observed that the mathematical model follows closely the one presented in ref. 15 (many eqs. are actually identical). Maybe just the reference would suffice.

Please have a look at eqs. 16 and 17, something seems to be wrong. In 16 the authors have used two times the “=” sign. Maybe using “≡” would be more appropriate. Please also check the signs.

Regarding the coefficients presented in fig. 5, how did the authors have obtained them?

The importance of figs. 6-9 is not fully emphasized or explained. Why so many different blocks formation? Why both fig. 8 and 9? What is the reason for all these figs? Please explain. Also, look at figs. 8, 9 and others where some pictures are hard to see.

What is the real sense at line 392?

A scheme to better explain the hybrid singulation would be welcomed. After all, this one of the main goals of this paper.

Please check the English. Some hints:

Line 130, some articles “the” are missing

Line 226: “can BE written as follows”

Line 352, eliminate “so”

Line 367: “takes A picture”

Line 396: add ARE at the end

Line 410: please check and reformulate

Comments for author File: Comments.pdf

Author Response

File is attached.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper presents an interesting research on hybrid manipulation approach for manipulating objects in cluttered environment. The paper follows a comparative study to show the performance of this approach over the pure pushing manipulation.

The paper is scientifically well written and well structured. The methodology is well defined. The only concern of the reviewer is about the evaluation part where he believes that the results are insufficient to evaluate the performance of the abovementioned manipulation approaches and eventually to provide a comparative analysis. The reviewer suggests that author need to perform more experiment and simulations in order to show the performance of the abovementioned approaches. Following you can find comments   

There are major comments mainly about the section 5 (Hybrid manipulation approach for singulation) and 6 (comparative analysis):

• In the section 5 authors represented a simulation and experiments on singulation following the two approaches i.e. hybrid and pure pushing. The reviewer believes that the repeatability is an important factor should be taken into account in this section. It is not clear that how many times the experiments and simulations on singulation tasks.
• In section 6, author just represented the time as a criteria for comparison between these two approaches. While the reviewer believe that it’s not enough to show whether the hybrid approach is performing better than the other one. For example, factors like “success rate” and “repeatability” or any other factor which related to reliability can change this judgement completely.
• I would like to suggest the authors to provide some more experiments in order to show robustness, repeatability and/or success rate of these two methods. Having these kind of measurements (in addition to task completion time (figure 15) ) would provide a way more clear evaluation over these two approach.   

 I have some minor comments:

• The reviewer believes that the prehensile and consequently the non-prehensile manipulation are well defined in the robotics. So I would suggest to skip the first paragraph of the introduction.
• Figure two is described the approach in a so generic ad vague way. In order to increase the clarity of this approach, I would suggest to provide some example (snapshots like figure 3) in order to show the possible steps that the robot would face.
• Please provide the explanation for the acronym PCA.
• In line 156, it’s mentioned that “Impulse propagation can be seen in Figure 4, where manipulator scatters the blocks” while the figure 4 doesn’t show such a thing. You may need to change this figure.
• Please define parameter T in equation (1).
• ?_a in equation (2) represents torque while torque usually is represented in literature by Tau (the Greek letter). I would suggest to use Tau as well in order to be consistent with literature and avoid to mistaking torque with the parameter T presented in equation (1).
• Please define what “a” is representing in equation (2) onward.
• Please define the parameter G in equation (4) onward. Please explain the superscript “env”.
• Figure 8 (c, d and e) and the figure 9 (c and d) are not clear. The same as figures 12 and 13.
• I would suggest to provide a table of parameters have been used in this paper since there are so many parameters.

Author Response

File is attached.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have addressed successfully all indicated issues in the previous review process.

Minor improvements are related to the language: please check all definite articles and punctuation.

Author Response

 [Comment 1-1] Minor improvements are related to the language: please check all definite articles and punctuation.

[Answer 2-1]  

Minor improvements are related to the language have been done and manuscript has been updated with track changes related to the corrections.

Author Response File: Author Response.docx

Reviewer 2 Report

Most of the comments are carefully answered, applied and integrated in the current version of the paper. It makes paper way more clear and easier to understand for readers. 
the only concern of the reviewer is about the results which as was reflected in the first review, those results are not sufficient to support the authors claim.
Authors have provided some answers to my comments and I wrote them back with more explanation and clarifications. I hope my comments are clear now.

You can find my comments attached to this message 

With kind regards

Comments for author File: Comments.docx

Author Response

 [Comment 2-1] In the section 5 authors represented a simulation and experiments on singulation following the two approaches i.e. hybrid and pure pushing. The reviewer believes that the repeatability is an important factor should be taken into account in this section. It is not clear that how many times the experiments and simulations on singulation tasks.

[Answer 2-1]

Thank you for pointing it out. Repeatability is usually used for getting the same outcome under same conditions. However, in our study the target is to singulate the soma cube blocks. The outcome in our case for different number of blocks is different and our approach of hybrid singulation is based on achieving singulation in less time. Nonetheless, we conducted simulations more than 5 times until the desired singulation is achieved and conducted experimentation 2 or 3 times for each formation. The experiment results were almost identical since the robot is controlled very precisely.

[Comments_rev 2-1] Your comment about repeatability test is true, but if you take “desired singulation” as your desired outcome so you can carry out the repeatability test. The reviewer’s comment was on “taking repeatability into account” which means to show weather the task is repeatable or not. For example you have repeated the simulations 5 times in order to get to the “desired singulation”. I there any evidence that after 5 time simulation you’ll get the same results? (you’ll get the same duration time or the same motion planning will be obtained) . For experimentations also you have repeated 2 or 3 times which is nice and my comment is just to show the duration time, average and/or standard deviation for each of these 2or 3 trials. That’s the only way to show the repeatability of the algorithm in practice.

[Answer_rev 2-1]

Thank you for pointing it out, in order to singulate objects the colliding velocity of robot and direction of collision is updated in virtual world by employing a feedback routine that would minimize the error to achieve the desired singulation pattern. The error is calculated by comparing the minimum distance of objects in each iteration in virtual world with the minimum desired distance among objects. Once, the minimum distance among object is achieved in virtual world, the same velocities information is used in real world to perform the object singulation. However, for large pile of objects such as for case of seven objects, it is not always possible to achieve the desired minimum distance among objects within predefined workspace. In those cases, the velocity of the manipulator is continuously increased, until the blocks leave the predefined workspace. That maximum threshold velocity is applied in real world to perform the singulation. The blocks, which remained in contact even after scattering, are singulated by using the pushing technique without using the virtual world simulator.

 In most of our cases we have seen that on average after 5 iterations in virtual world desired singulation is acheieved, here desired singulation doesnt mean that after each experiment in virtual world we get same singulation pattern, it actually means the it took 5 iterations until we get the singulation where the desired minimum distance among objects is achieved within predefined workspace and after that in the real world we have performed the 2-3 experiments for each formation and every time the achieved patern of singulation was different having desired minimum distance among objects.

[Comment 2-2] In section 6, author just represented the time as a criteria for comparison between these two approaches. While the reviewer believe that it’s not enough to show whether the hybrid approach is performing better than the other one. For example, factors like “success rate” and “repeatability” or any other factor which related to reliability can change this judgement completely.

[Answer 2-2]

Thank you for pointing it out. This question is along the same line as the previous question of the reviewer. Our goal is to carry out singulation in less time successfully. So the parameter of time is selected because in all cases both pure pushing and hybrid singulation approach can achieve singulation. We conducted 2 or 3 experiments for each formation we tested, and for all cases, the singulation is achieved either with pure pushing or hybrid approaches and have complete success rate in both approaches. Therefore, the success rate is not an important criterion in this paper. However, in case of hybrid approach, less time is consumed because it utilizes both scattering and pushing techniques together. It means that the hybrid approach is desirable with respect to the criteria of singulation time.

[Comments_rev] Considering your goal “carry out singulation in less time successfully” means that the duration time of te task is your benchmark. But there are still different factors that can affect on the duration time. For example : if you change the orientation of the blocks (when they are all together) you definitely will get different motion planning outcome from your simulations which means you’ll have different duration time (example 1). Or as another obvious example: if you increase the number of the blocks you will have different motion planning as the outcome of your simulations and you definitely will have different duration time (example 2).

So now if I ask my question again it would be as follow: is the algorithm able to successfully singulate the blocks for the conditions in example 1 and example 2? What would be the results if you combine this two factors? (having different numbers of blocks and changing orientation of whole blocks ).

Moreover, these two are just example that randomly came to my mind and authors should search and find more factors that can potentially/intuitively can affect the time duration of the task.  

[Answer_rev 2-2]

Thank you for pointing it out. Our hybrid approach carries out singulation in less time successfully both for blocks formation with different orientations and for the cases with maximum number of blocks which in our case is 7 blocks.

We have seen that when we try to singulate less than 7 blocks we only need scattering. However, when we have 7 blocks arranged in random formation we have seen that in some cases only scattering alone cannot be able to achieve desired singulation and in those cases our hybrid approach does the task of achieving desired singulation in less time.

Yes, orientation and number of blocks are the factors that affect the time of singulation that’s why we have used various formations starting from 3 blocks soma cube formation which only needs single scattering to get desired singulation to 7 blocks formation that in some random formations need hybrid approach of scattering and pushing for achieving desired singulation.

Pushing alone takes a lot of time for achieving the desired singulation that’s why hybrid approach is used to reduce the time of singulation.

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

There were two major comments on the paper about the results of the paper. What reviewer was asking was simply "considering" the factors that affects duration time and then show them them in results.
Authors have seems that have considered the most of the mentioned factors (by having an argument) but these factors are not reflected in the results.
Unfortunately, the paper cannot be accepted with the current results.  
Good luck