An Algorithm for Calculating Apple Picking Direction Based on 3D Vision
Round 1
Reviewer 1 Report
This paper presents An Algorithm for Pose Prediction in Robotic Apple Picking using 3D Visualization. It seems not much contribution if the paper only discusses the pose prediction since the apple can be considered a static target.
Figure 2 shows the mobile robot as the base for the arm robot. This is the interesting topic on how to integrate the arm robot motion control with mobile base motion and how to overcome the overwhelming DOFs of both the arm robot and mobile base and ensure stability.
The current manuscript presentation does not give any new contribution.
Figures 3 and 4 are common knowledge.
The mathematical models are also not sufficient for insight analysis.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Present paper entitled “An Algorithm for Pose Prediction in Robotic Apple Picking us- 2 ing 3D Visualization” aims to find the optimal picking posture. In my opinion, this paper is applicable and attractive because how to harvest fruits is one of the most important factors affecting the quality and storage period. Scratches on the skin of apples during picking lead to storage disease. Therefore, designing a high-efficiency robot that can pick the fruit in the best position is the main desire. Their proposed method predicted the optimal posture relative to the fruit and branches. ; 62.658% of the predictions erred by ≤10°, and 85.021% erred by ≤20°. The average time 25 for estimating the pose of an apple was 0.543 s. Although the estimated time for commercial applications is a bit long, it is currently satisfactory considering that the research work is a typical work and needs to be further improved for commercial application. Nevertheless, the idea was interesting and scientific. A few things that need to be edited in the text of the article are:
Sentences are usually not expressed in the form of the subject, for example line 81 “We then extract the point cloud within the bounding box and use the RANSAC algorithm to fit a sphere”. It should be stated as “the point cloud within the bounding box were extracted …..”
Figure 14 (a) has been written in non- English format
Conclusion should be modified. Line 457-467 should be omitted because is a kind of abstract
It is strongly suggested that an algorithm be added to the present study that identifies mature fruit while finding the optimal position for picking. The result of combining these two algorithms will be excellent
Please re-edit references. The format of all of them should be uniform and according to the guide of the journal.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
I think this paper addresses an important problem and solves it with a wise and interesting approach. The manuscript needs a moderate revision.
Moderate English language editing is required.
Sometimes apple tree canopy can be very dense, therefore it should be explained that what makes the researcher so sure that the nearest branch to the target apple is the branch to which the apple is attached?
It is recommended to explain what kind of orchard trellising system, the present detection system is appropriate for.
Lines 342-362 should be stated in section 2 (materials and methods) since it reports the approach of an experiment.
Figure 14(a) and Figure 14(b) should be renamed Figure 14 and Figure 15, respectively. Obviously, the follow-up figures shall be renamed as well.
The symbol “M” was used to show the point clouds of the target apple in section 2.2.3 and used the same symbol for point clouds of the branches and leaves around the apple in section 2.2.4. It causes confusion and it =is recommended to use different symbols.
In line 254, the symbol “M” has not defined clearly. It seems that it is the space between the two spheres mentioned in lines 252 and 253, Right?
The Space described in lines 252 and 253 and the space described in lines 262 and 263 is not exactly the same. Which one of these spaces is “M”; the space in lines 252 and 253 or the space in lines 262 and 263?
Due to grammatical errors in line 269, it is not clear what was meant by the words ‘distance’ and ‘straight line’. It seems that it was intended to state “ … and then connect them by a straight line. Next, the distance from the straight line to the center of the apple must be calculated.”?
Within the whole paper, the ‘pose (or posture) has been used as a ‘prediction’ phrase. Since it is not really a prediction, the word ‘determination’ should be used instead of ‘prediction’. Obviously, changing verb format is required for the aforementioned words too.
Line 285 should not be a paragraph by itself.
In lines 312-316 the word ‘if’ is used twice in a sentence and makes the sentence incorrect grammatically.
In line 318 please replace ‘can are obtained’ with ‘can be obtained’.
In Figure 14(a) please remove the non-English text.
In lines 231-235 several inappropriate terms are used instead of robotic technical terms. In line 332 the term ‘robot base coordinate system’ is used to mention the end-effector’s initial coordinate. Actually, according to Figure 2(a), the base of the robot is attached to the floor of the robotic platform which still and really does not affect the path planning process in picking an apple. First, the initial position and orientation of the end-effector are totally different from the orientation and position of the target apple. Through the path planning process, the robot moves towards the target and gradually changes the gripper’s (end-effector’s) position and orientation. The aim of your research is to provide the robotic system with an appropriate target position and orientation, especially focus is on the orientation. Within the whole paper, you should use the terms position and orientation instead of ‘posture’ or ‘pose’, since ‘posture’ and ‘pose’ do not express the meaning properly.
According to comment number 15, it is needed to briefly explain the path planning process of apple harvesting and its relations to the present paper to help the reader understand the paper’s needs.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
See the file.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
This paper discusses An Algorithm for Orientation Determination in Robotic Apple 2 Picking using 3D Visualization. The authors implement YOLOv3 for target detection and RANSAC for sphere fitting by considering that the apple is a sphere. The authors applied a mobile manipulator to pick the apple (object); however, the authors do not consider the motion and degree of freedom of this mobile manipulator. The object is considered stationary, raising doubt about whether the proposed method works in this research.
The authors only give the distance from the robot to the object. This manuscript lacks modeling analysis of mobile manipulators and overly simplified the algorithm to pick the apple. Fig 12 does not show the actual condition of the apple without leaves and other backgrounds. Therefore, this manuscript does not show that the system is applicable in actual agriculture/farming.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 4 Report
The authors gave the fruit branch morphology of artificial pruning that they considered. Also, the authors transferred the main topic from the the Pose Prediction to the Orientation Determination. However, the research highlight is yet not clear, and has serious flaws.
1. Maybe this research should be highlighted as the issue of the avoidance of branch interference during robotic picking apples. This issue includes the identification problem and determination of grasping direction. That means, the title is not accurate (only mention the orientation determination) to highlight the problem, and it is easy to be misunderstood.
2. In the text, the authors always use the word “predict”. But maybe only the Apple Identification can be called “predict”, because it has the priori model. The line fitting and the the orientation are only calculation, not “prediction”.
3. During the robotic picking apples, what the robot has to face is the relationship between different apples and the nearest branch, rather than the different positions of the same apple and the nearest branch. Different positional relationships are just calculations, while apples and branches need to be predicted by the model trained by the author (YOLOv3). The errors mainly come from the points cloud extraction and the fitting process. This causes that the design of experimental work is not reasonable. I want to say, in the experiment, the authors should use many different apples, not only a single apple.
4. In the text, the authors never explain and mention the Figure 15. Also, in the Figure 15, why does the value of axis y convergence to 1% ?
Author Response
Please see the attachment.
Author Response File: Author Response.docx
