An Advanced Rider-Cornering-Assistance System for PTW Vehicles Developed Using ML KNN Method

Questions

Answers

1 Define the acronyms, or put a reference: What is ESC? What is KTM?

Authors thank the reviewers. Now the meaning of ESC, KTM and all acronyms is reported in the text of the Manuscript.

2 In [9] or [10] the name is misspelled (it is Cali or Calì ?)

Authors thank the reviewers. Now the correct surname “Calì” is reported in the text of the Manuscript.

3 Take "Below" - line 112. Put “next” for instance;

Authors thank the reviewers. Now the sentence in line 112 starts with the word “next”.

4 What is Gr? What is Gf? (front and rear wheel center of gravity¿), but must be quoted.

Yes, authors confirm that G_r and G_f are the centers of gravity of the front and rear wheels (line 131-132).

5 Line 206 – remove “above” put “previous” for example;

Authors thank the reviewer. Now the word has been changed as indicated.

6 Line Define "rho"; defines "g" – equation (9);

I µ_lat, ρ, v_x, ϕ_{r , g} are respectively: coefficient of static friction, the curvature of the turn, the longitudinal speed, the tilt angle of the road, gravity.

See line 280-281.

7 line 241 - fix "precited";

Now this sentence has been changed. The correct word is predicted (estimated).

8 line 243 "is the response..."

Now this sentence has been changed as suggested.

9 Line 252 - I had defined only CVe h...

All coefficients introduced  in the study were defined in Section 3.

10 Correct line 259 "precited"?

Now the sentence has been changed. The correct word is predicted (estimated).

11 what is ci? Lowercase or uppercase?

Authors thank the reviewer. Now “ci” has been replaced with the  (lowercase) curve indexed according to the measure  à position = i (See table 2).

12 line 262 - is the response ...

Authors thank the reviewer. Now the sentence has been changed as indicated.

13 line 277 - is the response ...

Authors thank the reviewer. Now the sentence has been changed as indicated.

14 The notation (font) of mathematical symbols, when outside the equation, should be in italics;

Authors thank the reviewer. Now the mathematical symbols have been changed as indicated.

15 Make uniform notation. CDriver or Cdr (as in Fig. 10)?

Authors thank the reviewer. Now uniform notation for C_Dr has been used in the text.

16 Use uniform notation for equation ... equation 10;

Authors thank the reviewer. Now the mathematical symbols have been changed as indicated.

17 K from capital KNN always?

Yes, in our manuscript KNN can always be written with K capital.

18 Line 367 – standardize the call of equations...;

Authors thank the reviewer. Now call of equations were standardized.

19 Fig. 5 - prediction is exactly equal to the real value? Comment...;

As explained in the paragraph, the prediction of adherence coefficient was the subject of the authors previous paper. We meant to repeat the learning so that the new algorithm considers the theoretical values. Hence the similarity between predicted and real values.
As for the original prediction, here are the results:
As an output, the road grip is evaluated using test data as a function of velocity (Input). In fig.4 we can compare the prediction results given by the two technics (KNN and ANN *) with experimental data (○).

We found that the predicted road grip given by the KNN model is the closest one to the experimental values. More than that, ANN is instable because of its probabilistic behaviour. By calculating R² indicator, we found that the ANN algorithm gives different values that variate from 25% to 90%.

See line 290-306.

20 Fig. 10 does not show that the result K=2 is the best for all algorithms... In fact, there is little difference between the values of K. It would not be the case to say that, depending on the case, in which an algorithm is better, would the corresponding K be chosen? For example, if the objective was to test the driver, then what would be the best K?

In fact, the performance of the algorithms doesn’t behave the same way with the variation of K. Nevertheless, the only case which reacts significantly to the variation of K is that of C_ct prediction.

On the other hand, to optimize the algorithm resources we preferred to go for the least constraining.

21 You could test the Kalman filter, even linear, there is already a version with parameter estimation (covariances) in real time. There are also non-linear filters (Extended and Unscented). Kalman Filter tests for this dynamic performance function should at least be cited in future work.

Thank you for your suggestion, it will be a considerable added value to work with the filters (specially KALMAN). I believe that it could be complementary for our work on the deployment phase. Although given that the spirit of our project is to deploy less expensive instruments and to exploit more virtual estimators, hence our orientation towards ML technics.

Questions

Answers

1 Please explicitly indicate and clarify the challenges this study aims to address. What are the challenges and why? Why cannot the previous studies well address these challenges.

The challenges this study aims to address are:

-  The complexity of the PTW (Powered Two Wheelers) dynamics;

-  The lack of driver assistance systems for this type of vehicle;

-  The real-time dynamic parameters sharing of information;

-  The lack of databases in this field;

These challenges are the main obstacles that obstruct the constitution of a development framework based on experience feedback and design an affordable driver assistance system for the majority of users of this type of vehicle.

The majority of studies that addressed this topic (curve assistance system) used an instrumentational approach based on sensors or automatic estimators, but we could cite the previous studies conducted in the same spirit (Ferhat Attal. “Classification de situations de conduite et détection des événements critiques d’un deux roues motorisé”. Mathématiques générales [math.GM]. Université Paris-Est. Français. NNT :2015PESC1003. (2015)) in which the author aimed to characterize critical curve situations using AI classification algorithms. Indeed, our work complements his study.

On the other hand, at this stage of the state of the art, the framework not yet being completely finalized, the works which arise complement each other. Notably the use of image recognition and AI for visual detection by camera, which is the main field analyzed by researchers currently.

See line 81-109.

2 At the end of section 1 add a table that summarizes the advantages and disadvantages of existing methods facing the same problem. This way the reader would rapidly appreciate novelty of the paper.

Thank you for your suggestion. A table that summarizes the advantages and disadvantages of the existing methods facing the same problem has been added at the end of section 1.

3 The introduction and related work are mixed into a long section. It would be more clearly for readers to separate them.

Authors thank the reviewer. Now a new sub-section was added in the Introduction.

4 Please enrich the captions of all figures and tables for clarification.

Authors thank the reviewer. Now the captions of all figures and tables have been enriched.

5 Figure 4 needs more explanation.

Authors thank the reviewer. Now contents of Figure 4 have been explained more in detail.

6 Why do you build on the K-Nearest Neighbor (KNN) Machine Learning technique? There are many state of the arts Machine Learning techniques that used with more popular and demonstrate better performance than KNN.

Our Choice is for KNN for several reasons, namely:

·        As explained in section 3.1 (methodology description), given the nature of outputs (numerical observations) we chose regression form of KNN.

·        We conducted a comparison of the different methods and we found that the ANN and KNN gives the best results, but the ANN is unstable given its probabilistic behavior. By calculating R² indicator, we found that the ANN algorithm gives different values that deviate from 25% to 90%.

·        it is optimal, uses less resources and gives good performance, simple to deploy and to interpret.

See line 81-109.

7 As shown in Figure 10, what is your explanation that the best result at K=2 for the KNN?

There is a slight difference in the performance of the algorithms with the variation of K (numbers of nearest neighbors), but we could observe a best performance with K=2 especially in the case of Cct prediction. This is explained by our database layout: the KNN algorithm exploits the correlation between the values, thus the results are limited to the extent of the database that we have used and the diversity of scenarios that it represents.

See line 376-381.

8 In the comparison to SOTA methods, more experimental results of other state-of-the-art methods should be given.

As explained before (question 6), The SOTA comparison was already performed on our previous paper ([26] Fakhreddine, J., Bekkay, H. and Abderrahim, M. Controlling Powered Two-Wheeled vehicles in bends using machine learning. International Conference on Electronic Engineering and Renewable Energy Systems (ICEERE'22), Saidia, Morocco, 2022.), but considering the characteristics of our dataset and the advantages offered by KNN, we preferred to use this technique.

9 I also find some grammar problems in this paper. Author needs to carefully check these low mistakes, which is very important for readers.

Authors thank the reviewers. The article has been reviewed by a native English speaker and all grammar errors in it have been corrected.

Questions

Answers

1 I appreciate your idea to estimate an appropriate value of the heeling angle of PTW by developing algorithms based on on-board measurements and agree that the sensors available to measure this variable are too bulky and expensive. Based on experience with sensor systems for autonomous vehicles I am afraid that single sensor(s) of one type might be insufficient to provide accurate information. The developed model based on ML and selected ML techniques appear to be promising although arguments for your choice (ANN, KNN) should be given. The parameters for the equilibrium control quantities are well defined and the accuracies of correction coefficients (the clue of your approach) well deduced and discussed. I took from the long list of references that a rigorous state-of-the-art assessment was made though most of the cited articles are surprisingly old. Finally, please clear the text for some minor faults and improve the clarity of Fig. 5 before publication.

The authors express their gratitude to the reviewer.
The text has been reviewed by a native English speaker. All language errors have been corrected and the clarity of Fig. 5 improved.