SLAM Meets NeRF: A Survey of Implicit SLAM Methods

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors This paper is interesting and covers up-to-date subject of NeRF-based SLAM. There are some shortcomings and inaccuracies in the paper that need to be corrected by the authors. 1) I am not a native speaker, however I see incorrect use of English wording in some parts of the paper, for example "The current SLAM methods are difficult to render real-time observation". - From the context, it seems that SLAM algorithms are difficult to use. Please revise the paper for such inaccuracies. 2) As authors mentioned, the type of SLAM algorithm depends on sensors that are used to acquire the data. Please make a systematic division in the case of methods using NeRF-based networks. I suggest comparison in a table. 3) "The current SLAM methods are difficult to render real-time observation". - It is not clear from the context which SLAM algorithms the authors are writing about. It is currently possible to implement SLAM on a Raspberry Pi with ROS OS, which will perform real-time SLAM and send the data for real-time visualization to, for example, a laptop. We have done such a project as well as there are dozens of them presented on youtube. Please be more precise.   4) Figure 1 - the figure at the top acts as a colored table in which the authors presented key challenges, mapping factors, tracking algorithms etc. The top right part of the figure (the one with graphs and colored points) and the entire bottom part of the figure (the one with grids, neural network) is not understandable at all. There is no description on it and it is not clear what it refers to. This should be corrected accordingly.

5) What is the loss function optimized by the NeRF in question? Please systematically describe them.

 

Comments on the Quality of English Language

I am not a native speaker, however I see incorrect use of English wording in some parts of the paper, for example "The current SLAM methods are difficult to render real-time observation". - From the context, it seems that SLAM algorithms are difficult to use.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

1.       The paper is rather a comprehensive analysis and synthesis of the current state and future outlook of NeRF-based SLAM systems. The paper is structured in a way that facilitates understanding of the evolution and advantages brought by integrating NeRF into SLAM systems.

2.       The authors are emphasizing the recent advancements in SLAM systems, highlighting the significant benefits brought by NeRF in terms of both performance and accuracy. It is important to note that the paper represents an effort to synthesize and evaluate the current stage of research, providing an overview of the directions for the development of this promising field.

3.       Also the paper examines in detail aspects such as environmental expression in a dynamic environment, considerations regarding SDF function and volume density, and improvement of loop detection and positioning performance. These aspects are treated critically and reflect the need to address practical issues that arise in the implementation of NeRF-based SLAM systems.

4.       The paper represents a valuable contribution to the research field in SLAM, providing a deep analysis and balanced perspective on the evolution and future challenges. It is evident that the efforts of the authors are directed towards strengthening the theoretical and applied foundation of this field, and the paper serves as a useful tool for researchers and practitioners interested in this subject.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors The authors have addressed my remarks. In my opinion, the paper can be accepted as it is.