RNGC-VIWO: Robust Neural Gyroscope Calibration Aided Visual-Inertial-Wheel Odometry for Autonomous Vehicle

Round 1

Reviewer 1 Report (Previous Reviewer 1)

The revised version is of high quality for publication.

Author Response

Dear Reviewer

On behalf of my co-authors, thank you once again for your positive feedback. Thank you for your comments which are valuable and helpful for revising and improving our paper, as well as the important guiding significance to our research.

Thank you and best regards

Reviewer 2 Report (New Reviewer)

The paper introduces neural network approach to noise estimation in visual inertial (and wheel) odometry in order to reduce the long-tail error accumulation. The paper is interesting and technically sound and conveys its approach in an understandable manner. It is also supported by experiments and comparisons with baselines. There are quite a bit of silly typos and formatting issues that the authors should correct. Ideally, authors would use full functionality of LaTeX and automate the process of subsection style/numbering as well as referencing tables and figures in the text.

Some larger issues:

I could not find any information about dealing with synchronization issues between the many sensors used in the odometry approach. This should at least be touched upon in some way. Especially for neural network inputs, where recorded sequence may or may not be equally distributed in time.

It is unclear what is the input to the neural network. The paper mentions the sequence of sensor data. Is this a one-dimensional vector of fixed lenght? At some point number of 505 is mentioned as value for N, which would be the sequence length. How is this number chosen? And why is the input to the neural network a 6-channel input? These details should be clarified.

 

Minor issues:

Line 129: Great -> great

There are a lot of abbreviations used in the paper, not all of which are explained. MDPI journals have an abbreviation section where authors should show their meaning. This is not present in this paper but would be very helpful.

poly-nomial -> polynomial

Words in subsections (i.e. 2.1 2.2) capitalized and then not capitalized. Some subsections are in bold, some just italic. Please use latex functions (i.e. \subsection{Subsection Title}) to define subsections in order to avoid such issues.

How is related work different from approach? Relevant literature is given but no direct comparisons made. This should be updated.

Line126: related works -> related work

preprocessing or pre-processing. Please choose one way of writing

Figure 1 is unclear. There is a boolean "initialized" but no option for answer no. What happens if it is not initialized? Why provide such check if there is only single answer? Is initialization required on every pass?

Line 243: comma before respectively

Explained meaning of N in line 280 and line 287. It is unclear if it is the same variable and used in the same way.

Subsections 3.2.1 Gyroscope correction Model and 3.2.1 Data preprocessing have the same subsection number. Please use latex functionality.

i repeats as time-step and level of depth in network in line 317. Some variables seem to be re-used many times with different meanings. This is confusing and should be corrected.

Some variables are italic and some are not. Please stick to one style (italic preferrably)

k is image frame and integral increment length in line 346

3.3.4 Implementation Details should be 3.2.4

Line 435: donates -> denotes

3.3.4 Nonlinear Optimization -> why is it bold?

Line 484: Formula -> equation

Line 546: we use the same training sequence and test sequence - what is meant by this? Used for what exactly?

Line 731: hard to understand that - is a symbol and not part of text.

References to tables and figures are sometimes present and sometimes missing. In one part of the text authors with write "table" and in others "Table". These things should be corrected. Please use latex references function to refer to the tables and figures. This is present in some parts of the text but not others. Please correct this.

Quality of English language is fine, though there are some typos here and there.

Author Response

Dear Reviewer:

Thanks for your comments concern our manuscript entitled”RNGC-VIWO: Robust Neural Gyroscope Calibration Aided Visual-Inertial-Wheel Odometry for Autonomous Vehicle” (Manuscript ID: remotesensing- 2553481). Those comments are valuable and helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied all comments carefully and have made correction which we hope meet with approval. In the revised paper, all the changes are highlighted in red. The main corrections in the manuscript and the responds to the comments are are in the uploaded attachments.

Author Response File: Author Response.docx

Reviewer 3 Report (New Reviewer)

Line 30 – VIWO not VWIO

Line 33 – What’s RNGC? It has not been defined before.

Line 40 – Change from High accurate to High accuracy

Line 122 – It should be multi-modal sensors, not multi-model

Line 206 -  This phrase need to the paraphrase: "…outputs is simple directly input…"

Line 213-213 – Please specify what parameters are calculated?

In Figure1, change word lastest to last.

Simplify Figure 1 into a block diagram.

A proofreading is required to enhance the readability of the manuscript.

Author Response

Dear Reviewer:

Thanks for your comments concern our manuscript entitled”RNGC-VIWO: Robust Neural Gyroscope Calibration Aided Visual-Inertial-Wheel Odometry for Autonomous Vehicle” (Manuscript ID: remotesensing- 2553481). Those comments are valuable and helpful for revising and improving our paper, as well as the important guiding significance to our researches. We have studied all comments carefully and have made correction which we hope meet with approval. In the revised paper, all the changes are highlighted in red. The main corrections in the paper and the responds to the comments are in the uploaded attachments.

Author Response File: Author Response.docx

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

This paper presented a novel method for IMU calibration. Overall, the structure of this paper is well organized, and the presentation is relatively clear. The work is meaningful. However, there are still a few problems that need to be carefully addressed. More specifically,

1. For autonomous vehicle localization with multi-sensor fusion, it is usually complemented by sensor fusion such as IMU, GNSS, LiDAR, and camera. Thus, some related work should be included and discussed: an automated driving systems data acquisition and analytics platform, yolov5-tassel: detecting tassels in rgb uav imagery with improved yolov5 based on transfer learning, improved vehicle localization using on-board sensors and vehicle lateral velocity. In the above fusion process, the alignment of time and space is fully considered.

2. As mentioned in this paper, the attitude estimation of the vehicle will fully rely on the correction of the IMU. In the process of IMU correction through acceleration and angular velocity, the reasonable introduction of the vehicle's own characteristic information such as wheel speed and steering wheel angle will also help the error accumulation of IMU. Thus, some related work should be included: autonomous vehicle kinematics and dynamics synthesis for sideslip angle estimation based on consensus kalman filter, vision-aided intelligent vehicle sideslip angle estimation based on a dynamic model, automated vehicle sideslip angle estimation considering signal measurement characteristic, estimation on imu yaw misalignment by fusing information of automotive onboard sensors. The above strategies could help IMU calibration with its applications to autonomous driving.

 

3. Please double-check where the reference is inserted into the error. To optimize the curve in Figure 6, in addition to the color, it is also necessary to use different line types to distinguish.

Reviewer 2 Report

First of all, I would like the authors for the effort done and put in the work. However, I have few questions and concerns.

Reviews:

1.      For Figure 1, after the Initialization decision making, if the answer is No, it starts an initialization stage which from the figure is considered as a separate stage that is not needed by any means for any of the work. The outcome of the figure is not being used anywhere.

2.      For Figure 1, for the decision making “Odometer Input” if the answer is Yes, all what is done is the IMU-Odometer Pre-integration Alignment, what is next?

3.      For Figure 1, for the decision making “Odometer Input” if the answer is No, all what is done is the Vision Inertial Alignment, what is next?

4.      I am not sure how is the Initialization decision takes velocity and gives acceleration afterwards to me merged or compared with the Denoised IMU Pre-integration acceleration?

5.      English needs revision, for example: nothing is called “Lastest”

6.      For Equation (1), usually the Gyroscope output is a position, how is that being directly added as a bias for an angular velocity?

7.      In Equation (2), please use a different term for the scaling factor “S” as that term usually represents the Skew matrix.

8.      The graphs do not make any sense, talking about an experiment for 8 km in horizontal motion between (318 km to 326 km) and 5 km altitude between (4152 km to 4157 km). I do not know how you were able to reach these altitudes.

The paper required a lot of effort to follow the explanation of the figures and tables. It needs to be clearer and having a logic flow.