A GNSS/LiDAR/IMU Pose Estimation System Based on Collaborative Fusion of Factor Map and Filtering

Round 1

Reviewer 1 Report

The paper undertakes an interesting and importantproblem of trajectory finding in different areas with GNSS/IMU/SLAM fusion.

It contains a solid theoreticla backgroud strongly supported by high-quality experiment.

 

my remarks are mostly proposed to improve the readability of the paper.

 

- fig 1 - edges on the figuer are between variable nodes, while in description it is between factor nodes aand variale nodes - please clarify and unify

- sec. 2.1 - is it possible to give simple example for this factor graph optimization? for some simple variables? - the techniques is not widely known and this would improve the reception of further part

- sec. 3.1 - please explain SO(3)

- eq. 1 - please clarifiy what does theoretical means here? what are these values in fact?

- eq. 6 - shouldn't we use brackets in this equation?

- please explain somewhere the abbreviations like Lo, Lio, MSFO etc. it will ease the perception to reader and not all of them are clarified

- fig. 9 - tractory? probably trajectory

- sec. 4.2 - please add statistics of related accuracy jointly for all trajectory points (if possible)

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper present a GNSS-IMU-LiDAR Constraint Kalman Filter (abbreviated as GIL-CKF), which has the characteristics of full coverage, effectively improving absolute accuracy and high output frequency. Some points need to be improved before the publication.

 

1.     The full name of GNSS/IMU must be firstly given.

2.     Fig 1 is so fuzzy and quality needs be improved.

3.     Section 3.2.2, the full name of PDOP, HDOP and so on are lossing.

4.     Figures 6-8 and 10, with the black background of the figures, it is difficult for readers to know the content of the figure. The contents must be improved.

5.     More information should be introduced for the Table such as Table 2 and 3.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

 

This paper proposes an interesting positioning method based on GNSS/LiDAR/IMU data fusion and the use of EKF. I think the work is relevant but lacks information that is relevant to evaluate the method, such as the performance specifications of the sensors. Additionally, I believe that readers would not be able to reproduce the proposed method based on the information provided in the manuscript, improvements on this are mandatory. I think these aspects can be clarified by the authors but the theme of the paper does not seem to be a clear contribution to a journal of Remote Sensing area. Because it is out of scope my recommendation is to reconsider after major modifications of the manuscript or forward to another journal more in tune with the subject.

Some other considerations:

The abstract does not adequately define the problem and the results achieved are not clear

Mathematical formalism is missing in section 2 on PGM and EKF. Classical references on the subject are not cited

When figure 1 is mentioned in the text, it cannot be understood, as information is missing. Improve the description of figure 1

 It is recommended that you devote more time to describing Fig3. It is the key point of your work and I cannot follow everything that is processed in the blocks without a proper description. I honestly had trouble understanding Fig3 with the diagram in Fig5. Further on in the experiments section (5) it gets even more confusing for me because relate those figs, there it seems to be a combination of methods and not a navigation based on sensor fusion.

 Section 3 lacks information about the the sensors. For example, what is the accuracy of the GNSS used (the manufacturer alone is not enough information)? Which observables, code, carrier, two frequencies? nothing is informed, Fig4 suggests that NMEA messages are used. The explanation in Fig4 is superficial. I don't know if this could strictly speaking be called a filter. It looks more like a quality control strategy of the positioning data. At some point you mentioned centimeter level for GNSS, with SPP and NMEA messages? In section 4.1.1 you mentioned about differential GNSS, then. Please clarify.

 The same question goes for the IMU, is it MEMS? what are its accuracy of the sensors? For example, you cite that IMU bias drift is usually very accurate, what does this mean in factual values?

 Extending this information to the Lidar is also recommended (C16 only). The accuracy of your instrumentation will surely affect the performance of the system as a whole.

 Sample rates are not addressed, but how do you deal with different sample rates? In fig3 we have 200 Hz for the IMU, what about the other sensors?

 How is the parameter K in equation 7 chosen?

Include the coordinates of the tests, even if it is approximate to a reference point. Which campus? Please identify better this location.

Define what is a point cloud map (from lidar I guess?).

Section 4 needs to be better explained in its first subsection about field data details. For example, you mention in tab 1, 400m, was that the length of single lap or the whole test? Why not include a panoramic photo of the test environments?   There are papers in the literature that show that dense vegetation is one of the most critical environments considering GPS multipath (also for gaps), the more information the better.

I recommend that the authors to consider include histograms of the errors of their tests for different sensor combinations and environments. This may give a better idea of the results of the proposed method and its tolerance to different conditions.

 LOAM appears for the first time in pag9, needs to be explained properly.

Figure 7 has 6 panels and you spent 2 lines in the first paragraph and more 5 next to explain it all. I don't have the capacity to interpret all the information that is there alone. Please try to improve the description and content of the figures. What do the colors mean? Talk about the differences. Consider improving the description of the legend, for example providing description of the panels.

When you say that LIO has an error in fig7 case, please include values, there is no in-depth discussion of this error in the way the text is presented.

 You wrote in sec 3.2.2 that GNSS and LOI are highly complementary. Why don’t you show the performance of GNSS independent in this evaluation?

When the authors say “When LIO returns to the original point, it produces error accumulation and leads to point cloud stratification, such as Figure 7b and Figure 7c.” I honestly cannot extract this information from the figure, maybe I need to show it in another way.

Please check the sentence: “Based on the campus dataset, we compared and tested the LO, LIO, MSFO schemes”. Is that according to fig10?  

On the penultimate paragraph on page 12, the discussions are too brief and I have trouble keeping up your findings, consider improving this part of the text.

Fig 12, 13 15 and 16, is there a unit for the axis?

Were you able to map what is the maximum limit of the LIO that you can safely navigate, with an acceptable error limit? Clearly not 400m, but 100 or 50m for example?

The discussion section is too short, it is not justified and can be included in the previous section. Finally, the conclusion needs to better summarize the contributions of the paper.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

The letter to the reviewers was satisfactory. The insertions in the text of the manuscript could be more detailed but are sufficient. The citation of a reference by a professor's first name seems misplaced. The units of the axes in some figures are still missing even after requesting their inclusion. The authors have inserted a text of the unit in the figure legend, which sounds careless (including panel b of fig16).

Author Response

Please see the attachment.

Author Response File: Author Response.docx