MIM_SLAM: A Multi-Level ICP Matching Method for Mobile Robot in Large-Scale and Sparse Scenes
Round 1
Reviewer 1 Report
The paper improved. However, the authors need to address the following comments:
1. insert the novelty and contribution of the paper in the abstract.
2. the problems in lines 57 to 64 should be discussed and supported by new papers (REFERENCES from 2017 or 2018 preferably).
3. insert a discussion section including the novelty of the paper, contributions, comparing the results with other papers (references are required), limitations and future directions.
Author Response
Thank you for your comments concerning our manuscript.
Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval.
We tried our best to improve the manuscript and made some changes in the manuscript. These changes will not influence the content and framework of the paper. We appreciate for Editors/Reviewers’ warm work earnestly, and hope that the correction will meet with approval. The main corrections in the paper and the responds to the reviewer’s comments are attached.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The quality of the revised paper has improved. While all issues have been addressed, several of the main concerns not to a sufficient extent.
1. Eq. 2 is still not clear and I suspect wrong. According to Bayes, the following should be the case: P(X|Z,U) = P(Z|X,U)P(X|U) / P(Z|U) However, the prior term, i.e. P(X|U) is still conditioned on Z in Eq.2 ( P(x_m|z_1:M-1,u_1:M) ). Furthermore, I assume that P(Y|C) = P(y1:M|C) = P(yM|y0:M-1,C) P(y1:M-1|C) is used for various variables y (i.e. X and Z) and conditions (mostly U). However, somehow only the first term P(yM|y0:M-1,C) remained while it is not clear how the second term P(y1:M-1|C) vanished. Thirdly, the second line only contains x_M. The third line contains x_M and x_M-1 (and x_m-1 where I guess m should be M). The last line suddenly contains all x from x_0 to x_M. Thus, this equation is either quite wrong or a lot more explanations are necessary.
2. Eq10: I asked in my previous review why there is an additional Q^T. The response of the authors is that it was simply added and does not change the results. Please note that this is not an acceptable answer nor explanation. A (presumably) more correct answer would have been: Plugging in Eq.9 into Eq.8 leads to the first line in Eq.10. As only the length of the difference vector Q[R 0]theta - b matters, this difference vector can be rotated arbitrarily as a rotation does not change the length of the vector. Thus, Q^T can be multiplied to the difference vector. Since Q is an orthonormal matrix, Q^TQ=E (E is identity) which allows to simplify the equation to the third line in Eq.10. Please note: If the authors decide that a derivation as the one given in Eq.8-10 is necessary, those kind of details must not be left for the reader to figure out but need to be provided by the authors.
3. P9L224: Please explain how the reference pose has been defined, i.e. how were the numerical values of the reference pose derived?
4. While I appreciate that the authors included additional scenes to evaluate the proposed approach. However, those scenes are again own data. No experiments have been carried out on publicly available benchmark data. But even more importantly, the new scenes have only been used in a qualitative analysis. No quantitative results are provided for those scenes. Consequently, the proposed method is still only quantitatively evaluated by two scenes and the performance is thus not comparable to other works.
5. The authors included ICP as a baseline method. However, the authors did not include any other state-of-the-art reference methods besides the single one of the previous version.
6. Tab.2: Please note that the time difference between LOAM and MIM_SLAM is rather small (i.e. roughly 15%). Thus, if MIM_SLAM fulfills real-time requirements, it is hard to believe that LOAM does not (as P9L230 seems to indicate).
Without a proper evaluation, i.e. quantitative results on multiple data sets (preferable including public benchmarks) and comparisons to several state-of-the-art methods, I cannot recommend this paper for publication.
Author Response
Thank you for your comments concerning our manuscript.
Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied comments carefully and have made correction which we hope meet with approval.
We tried our best to improve the manuscript and made some changes in the manuscript. These changes will not influence the content and framework of the paper. We appreciate for Editors/Reviewers’ warm work earnestly, and hope that the correction will meet with approval. The main corrections in the paper and the responds to the reviewer’s comments are attached.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The paper improved. However, the conclusion section is still short. It contains broad information and doesn't include specific information, findings etc. It should be definitely extended.
Author Response
According to the reviewer's comment, we have modified the manuscript. It is shown from lines 302 to 317 in new version.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors addressed all of my comments and showed the competitiveness of their approach on public benchmark data. There are only two issues remaining that should be further addressed:
1. Please revise your manuscript regarding proper scientific English writing. There are many language issues. Please also note that "Wolfgang" in Ref [28] is the first name.
2. Eq.2 remains unclear... Please include a step-by-step derivation and do not leave important steps out. If X=x_1:M, Z=z_1:M, and U=u_1:M, then
P(X|Z,U) = P(Z|X,U)P(X|U)/P(Z|U)
<==> P(x_1:M|z_1:M,u_1:M) = P(z_1:M|x_1:M,u_1:M)P(x_1:M|u_1:M)/P(z_1:M|u_1:M)
However, the paper states as first line after applying Bayes Theorem:
P(x_M|z_1:M,u_1:M) = P(z_M|x_M)P(x_M|z_1:M-1,u_1:M)/P(z_M|z_1:M-1,u_1:M)
This is not Bayes Theorem. Either this part is wrong, or there are multiple non-trivial steps missing. The same goes for the following terms. It is not clear how the RHS of an equal sign is derived from the LHS.
Author Response
1. According to the 1st issue, we have modified the manuscript. It is shown from lines 63 to 64 in new version.
2. According to the 2nd issue, we have modified the manuscript. It is shown from lines 93 to 98 in new version.
Author Response File: Author Response.pdf
