A Novel Inverse Kinematic Solution of a Six-DOF Robot Using Neural Networks Based on the Taguchi Optimization Technique

Round 1

Reviewer 1 Report

In this paper authors report a typical implementation of a robust design artificial neural networks (RDANN) methodology based on Tagushi method with application to inverse kinematics of a robotic arm. The paper is correctelly organized, strating with an introduction which reviews the main contributions followed by a paragraph where they detailed their material and methods, then results are deailed in paragraph 3. The conclusions are presented in paragraph 5, a paragraph 4 is missing. 

The use of artificial neual networks for inverse kinematics is not a new contribution, especially with the classical feedforwad architecture. Authors should explain their motivations about the use of this architecture which needs a re-training ant time the main task is changed.... RNN for inverse kinematics should be included in authors review : 

- Liu, R., & Liu, C. (2020). Human motion prediction using adaptable recurrent neural networks and inverse kinematics. IEEE Control Systems Letters, 5(5), 1651-1656.

- Reinhart, R. F., & Steil, J. J. (2009, December). Reaching movement generation with a recurrent neural network based on learning inverse kinematics for the humanoid robot iCub. In 2009 9th IEEE-RAS International Conference on Humanoid Robots (pp. 323-330). IEEE.

 

It looks like the target is a 3D printed robot, a real application is missing in the results paragraph related to what real implmentation results compared to simulations? 

The ANN topology used is presnted in a late manner, it comes in figure 10 page 16, which may make the contribution difficult to understand for readers missing ANN background. 

The ANN activation function used in this investigation, need to be clearly indicated. 

The max desired topology of fig 11-page 16, should be better explained and detailed. 

Comparative analysis toward related works, cited by authors is also missing, such a compatative my help in assessing the method toward its challengers.  

Please check paragraph numbering, it looks like paragraph 5 should stand for (4), or that paragraph (4) is missing. 

A real application on the 3D printed robot is missing, and sould be added. Some comments regarding simulation results VS real application results are a must, especially if the 3D printed robot is available. 

Author Response

Dear reviewer, thank you for reviewing our manuscript, we attached a document with the suggested changes.

Best regards,

The authors

Author Response File: Author Response.docx

Reviewer 2 Report

Overall this paper is interesting in that it presents a framework of training Neural Networks for increased robustness while minimizing the training time and human-decision investment. The paper is well-written with few errors. The results are clearly presented and the outcome discussed in sufficient detail. Specific comments include:

1. The first sentence of the abstract is weak, and could be deleted. The second sentence is much stronger.

2. lines 19 & 20 need commas (19: after Taguchi, applied / 20: open source, 6 degrees)

3. line 35: first word should be singular (-work, not -works)

4. Citations 1-4 are odd choices for what is being cited. Consider replacing with any/all of the following:

Random search for hyper-parameter optimization, by James Bergstra and Yoshua Bengio (2012). by James Bergstra and Yoshua Bengio

Practical Bayesian optimization of machine learning algorithms, by Jasper Snoek, Hugo Larochelle, and Ryan Adams

Vasyl Teslyuk "Optimal Artificial Neural Network Type Selection Method for Usage in Smart House Systems", Sensors 2021, 21(1), 47

5. line 56: you have not identified what a BPNN is, so spell it out

6. line 58: delete the word ordinary at the end of the line

7. line 59: add space between [13] and "a"

8. lines 72-77: these summarize the important contribution of the paper and should be folded into the first paragraph.

9. line 78: "Experimental design" is not a statistical technique itself, maybe you need "Factorial experimental design" for this to make sense.

10. lines 136-144: this paragraph is out of place here, possibly move to after line 71.

11. line 197: remove "the" in front of equation (3)

12. Fig 2: make larger so that it is more easily to identify important components

13: Fig 4,6,7,8: split to be vertical instead of horizontal to fit within paper margins (without reducing size)

14. line 439: remove space between comma and word momentum

15. Ref [17] is probably the most important reference for this paper and should be higher in the list and discussed in the text in more detail.

16. Replace ref [14] with [34] and add Malik, "Multi-Objective Swarm Intelligence Trajectory Generation for a 7 Degree of Freedom Robotic Manipulator," Robotics 2021, 10(4), 127.

Author Response

Dear reviewer, thank you for reviewing our manuscript, we attached a document with the suggested changes.

Best regards,

The authors

Author Response File: Author Response.docx

Reviewer 3 Report

This paper proposes an inverse kinematics solution method of a 6-DOF robot using neural networks based on Taguchi optimization technique, and the results show that the proposed method can achieve high accuracy. The work is very interesting and meaningful, and it can be published after revision. Some suggestions are given as follows:

1. How about the efficiency of the proposed method? Which is very important in practice.

2. Can the method be applied in real-time controller of robot system in industry?

3. The quality of the figures in this paper should be improved, such as Figs. 3, 4, an 11.

4. Compared with some traditional optimization methods, such as GA and PSO, why the author chooses Taguchi method?

5. The conclusion part is too long, which should be summarized.

6. Besides the serial robot studied in this paper, the kinematics of parallel robot is more complex, and the proposed method can also be applied for parallel robot. This potential can be mentioned, and the following papers of parallel robot can be cited:

1) Minimum-time trajectory planning and control of a pick-and-place five-bar parallel robot, IEEE/ASME Transactions on Mechatronics, 2015, 20(2): 740–749.

2) A postprocessing strategy of a 3-DOF parallel tool head based on velocity control and coarse interpolation. IEEE Transactions on Industrial Electronics, 2018, 65(8): 6333-6342.

Author Response

Dear reviewer, thank you for reviewing our manuscript, we attached a document with the suggested changes.

Best regards,

The authors

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Authors responded my comments fairly. 

Reviewer 3 Report

The paper is well revised, and it can be published.