An End-to-End Grasping Stability Prediction Network for Multiple Sensors

Round 1

Reviewer 1 Report

This manuscript describes an end-to-end method to predict grasping stability, based on the use of ML as CNN.

First of all, an extensive language editing is needed, preferably performed by an English native speaker.

One aspect of the paper which is somehow weird to me is that the authors claim that one of the major contributions is the release of a dataset which has not been released so far; future works are not contributions, at most they will be future contributions. Moreover, the process of data collection described in the manuscript hasn't introduce any special feature. In my opinion, this claim should be removed as the data set is not public domain up to date. 

Fig. 1 looks poorly, and portrays almost no relevant information.

Eq. 1 is misplaced

Results included in Table 2 are so tight that a statistical hypothesis testing is probably worth. It should be provided to support the affirmation that 12x12 array yields a better performance.

The same stands for some results in Table 3

Some sections as Authors contributions are still to be fulfilled 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

 

An End-to-end Grasping Stability Prediction Network for Multiple Sensors

 

Major comments.

The paper describes a method of classifying the phenomenon of grasping stability using a variant of convolutional neural network. The issue is interesting and very important, especially for application in industrial robots, manipulators, etc. 

1. The aim of the paper is not exactly stated, we can only guess that it relates to the three main contributions listed – please update this.
2. One of them is the description and test of the new classification method, called MM-CNN. However, its description is not clear. The essential element of this is Figure 5, which is COMPLETELY ILLEGIBLE (too small fonts).
   An unclear description makes it difficult to say to what extent the proposed method is better than the methods known in the literature and how innovative the method is.
3. Poor presentation of output results is another problem. The author should consider reducing the size of the tables or reorganizing them (e.g. Table 3). They should only present data relevant to the main goal of the article (which should be defined).
4. The reviewer has the impression that the article is carelessly written, why none of the co-authors checked if the section “Author Contributions” includes right content! (only a copy of the “instruction for authors” is inserted).

Generally, the main disadvantage of this article is the lack of scientific accuracy in describing the problem.

 

Minor (detailed) comments .

1. Many papers use the term "grasping stability" in the common sense. Try to define this term strictly. This is important when creating data sets, their interpretation and so on.

2. Improve the quality of Figure 1
3. Figure 3. is unreadable (too small fonts).
4. Please provide more details concerning the software used (Adam), what exactly modules have been utilized? What hardware components in “Self-made Tactile Sensing Array” ase used? (amplifiers, type of A/D converters etc) What computer system is utilized?
5. 240: “..We resize the input to the unified size of 16*12 by padding 239 zeros in some experiments”. Does this operation interfere with the statistical results of the experiments?
6. 94:   “Compared with the Kitronyx tactile sensing array, our array is much more sensitive to the change of ….”   This statement is very imprecise, please estimate (very roughly) the sensitivity of the two sensor systems.
7. 108: “image of a single channel.” Please specify what exactly does this term mean here?
8. 110-116: How were these 1562 and 3478 samples created, do we have 9 objects(Fig.2)?
   Have different manipulator movement parameters been used?
9. 186: “ a vector of 2 in length …”  The terms length and dimension of a vector do not mean the same.
10. Equation (1) Please reformat this expression.

 

The article is difficult to read, mainly because of a strange selection of words. However, we can correctly guess what the authors meant. In my opinion, the language used in the article can be treated as a kind of dialect of international English. In this sense it is acceptable.

Nevertheless, I strongly recommend improving article language.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Most of the issues highlighted in the past review have been fixed. Nevertheless, it would be great to see a statistical hypothesis testing associated with result tables, as average and standard deviation do not provide the complete information to judge the comparison between sets.

However, I think that the manuscript reaches the standard quality to be published.

Reviewer 2 Report

The description of the problems presented in the article has been corrected. The article is suitable for publication.