3D LiDAR Aided GNSS/INS Integration Fault Detection, Localization and Integrity Assessment in Urban Canyons

Round 1

Reviewer 1 Report

1.      ‘GNSS and IMU are loosely coupled in [9], and the FDE in [9]……... However, this scheme is only suitable for the tightly coupled architecture.’ This is wrong, the FDE is also suitable for tightly coupled scheme

 

2.      The definition of measurement vector Z in section 3.1.1 is suggested to be added.

 

3.      It is confusing that there are two red arrows pointing to the Integrity Assessment box.

 

4.      How is the protection level calculated in the proposed algorithm and the candidate algorithm?

 

5.      A priori 3D LiDAR global point cloud map is established based on IMU and LiDAR data produced by LOAM in this paper. But, the error of LOAM is gradually accumulated over time. Does it affect the performance of the proposed algorithm? Besides, it seems that the computational cost for construction of a priori global point map is high.

 

6.      How does the proposed algorithm deal with the scenario without a priori 3D point cloud map?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Overall the paper addresses an interesting problem. One concern I had is that real systems tend to have delays and hence feedback may not be immediate observed. The authors should take a look at the delay model in the following and discuss if their results can be generalized this way. 

Distributed stochastic optimization with large delays/Zhou et al, Math of Operations Research, 2021

Distributed Asynchronous Optimization with Unbounded Delays: How Slow Can You Go?/Zhou et al, ICML 2018

High-Dimensional Data Analysis with Low-Dimensional Models: Principles, Computation, and Applications/John Wright and Yi Ma, Cambridge University Press    Fully Convolutional Line Parsing, Xili Dai, Xiaojun Yuan, Haigang Gong, and Yi Ma, arXiv:2104.11207, Neurocomputing, 2022.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

First, congratulations on writing such a valuable article.

However, I propose to improve the article, in particular, the analysis of the literature taking into account the items from the MDPI publishing house.

1.        The structure of the paper is good. First, the introduction presents the theme briefly. Then, the paper has a literature review section and a problem description section.

2.        The objective and contribution could be written more clearly in the introduction section.

3.        The literature analysis presented in the paper can be improved. It was prepared briefly and does not take into account all the aspects relevant to the scope of research included in the article.

4.        I propose to include articles from the MDPI publishing house in the literature analysis.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

In the paper, the authors present an approach for fault detection in GNSS/INS using a 3D LiDAR. It is proposed to match objects detected by the LiDAR with a previously known map in order to detect issues with GNSS caused by NLOS and multi-path effects. If a fault is detected, the noise values are adapted according to the matching error. The approach is evaluated on real-world data and compared to an EKF without error detection and mitigation.

The paper is well written and structured. It addresses an issues that is highly relevant for autonomous driving and the approach is technically sound. Furthermore, the extensive evaluation of several aspects of the system using real-world data is commendable. There are however a few points that need to be improved:

- There is no critical discussion regarding the complexity of the proposed approach. In particular, it relies on an up-to-date 3D map of the environment, which is not easy to obtain, and computationally expensive point cloud processing needs to be performed. There is also no analysis of the runtime and memory complexity of the system.

- Usually, the quality of the RTK solution (e.g. fixed vs. float solution) provides a good indicator about the reliability of the GNSS system. This aspect should be analyzed. In particular, the quality of the RTK solution should be shown in the trajectory, deviation, and fault detection plots. In addition, the proposed localization algorithm could be compared to an EKF where the GNSS uncertainty is adapted according to the RTK quality (i.e. increased by a static factor when it drops from fixed to a degraded quality).

- It is not discussed why target matching is used instead of classical scan matching.

- Sect. 3.1.1 provides several standard equations of the EKF, but the actual models for the specific application (motion model and sensor models) are missing. The model equations should be given, while for the standard equations, a reference to existing literature is sufficient.

- In Sects. 3.2.2 and 3.2.3, several details and references are missing. The process should be described more in detail and the references to all algorithms and software should be given.

- Giving values of the results should be avoided in the abstract, because they cannot be interpreted without knowing the experiment setup. A qualitative assessment should be given instead.

- Since the reference source (ground truth) is based on GNSS/INS as well, it should be discussed to which degree this system is affected by NLOS and multi-path effects in the critical situations.

- It should be discussed why a growing threshold is advantageous compared to a fixed one. Fig. 13 suggests that the latter is sufficient or even better.

- All parameters used in the evaluation should be given.

- The exact meaning of "NARROW INT" should be explained, since this is not a standard RTK term.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 5 Report

The paper introduces the original 3D LiDAR aided real-time

fault detection algorithm of the GNSS - INS navigation.

The paper presents, besides the theoretical analyses also, results of the testing of two scenarios. 

The detection algorithm is based on the Kalman filter.

The measurement equation (2) is not sufficiently discussed. It is not clear whether the filter processes the GNSS position or row code and phase pseudoranges. Similarly, the model of the LiDAR measurement is missing. 

The description of the GNSS INS algorithm is missing. There is information about the sampling rate of the GNSS RTK position and INS sensor only. 

The usage of the NovAtel SPAN-CPT7 as a source of absolute position is questionable. Why have they developed such a complicated system if there is a solution?

In the introduction, the authors write that multipath propagation and NLOS propagation have the main effect on the precision of the GNSS - INS position. The logical approach is to solve the problem at such a level, i.e., isolate the affected satellites and calculate the propagation effect-free position. Instead of this, the authors developed complicated algorithms for the isolation of the positions with higher errors. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 6 Report

Dear authors,

the paper is very interesting, but I would suggest adding some information:

1. could you provide, even in the text, what are the differences between your method and reference 29?

2. in section 2, I suggest to add a detailed description of the proposed algorithm, not only an overview.

3. in section 5, the sensors setup is described. could you provide information about the acquisition system and how all the sensors are synchronized?

4. in both tests your algorithm and a ''traditional EKF'' have been compared. In order to highlight the performance achieved, the reference system (i.e. traditional EKF) must be described in terms of accuracy and uncertainty. Moreover, is this a certified system? in this case please provide some information about the manufacturer. Otherwise, if this reference system is realized by the authors, please provide a detailed description of the realization and implementation.

5. In the conclusion, I suggest to add a results comparison with other scientific research, where the main and similar results are highlighted.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

It seems that there is a redundant half bracket in the ‘Error bounds (Protection Level)’ box in Figure 1.

 

In the introduction, some GNSS fault detection methods have been reviewed. Besides FDE, however, there are still other methods proposed for dealing with NLOS/multipath effects in urban environments, such as antenna design-based methods (Jiang et al., 2014), pseudorange error modeling (Sun et al., 2022), and 3D city model aided NLOS detection (Groves and Adjrad, 2017). It could be better if those methods are also reviewed in the paper.

 

Groves, P. D., Adjrad, M. (2017). Likelihood-based GNSS positioning using LOS/NLOS predictions from 3D mapping and pseudoranges. GPS Solutions, 21(4): 1805-1816.

 

Jiang, Z., & Groves, P. D. (2014). NLOS GPS signal detection using a dual-polarization antenna. GPS Solutions, 18(1).

 

Sun, R., Zhang, Z., Cheng, Q., Ochieng, W. Y. (2022). Pseudorange error prediction for adaptive tightly coupled GNSS/IMU navigation in urban areas. GPS Solutions, 2022, 26(1): 1-13.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors have addressed all my comments for during this revision. Recommend accept.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The authors addressed all my comment properly.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

In the revision, the authors carefully addressed the reviewers' comments. Additional discussion have been added, the results have been extended, missing references are now provided, and the description of the algorithm has been improved. I've no further comments.

Author Response

Please see the attachment.

Author Response File: Author Response.docx