Reinforced NEAT Algorithms for Autonomous Rover Navigation in Multi-Room Dynamic Scenario

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this paper authors evaluate the performance of the developed  AI model and autonomous rover in multi-room environments.  simulation  performed on a 2D map created using Pygame,  and rover  image as an agent  will move throughout the environment to learn things, for which NEAT applied. Working of the NEAT algorithm also properly explained with example. Overall its good work, but I have some suggestion to improve the paper.

 

1.      Figure 1 and 2 need to explain properly like what is white, black and green space represent. As per my understand its room and entry points.

2.      It will be better if these algorithm check on a particular robot for better understating as simulation results only valid for the 60-70 percent.

3.       Sensors required for path planning/ navigation also need to explain in detail.

Also need to explain the  significance of AI model for this problem.

Author Response

Comments 1: Figures 1 and 2 need to explain properly what white, black, and green space represents. As per my understanding, it's room and entry points.
Response 1: Addressed in lines 144-149.

Comments 2: It will be better if these algorithm check on a particular robot for better understating as simulation results only valid for the 60-70 percent.
Response 2: Addressed in lines 798-832, 843-848.

Comments 3: Sensors required for path planning/ navigation also need to explain in detail. Also need to explain the significance of AI model for this problem.
Response 3: Addressed in lines 790-795 and 762-769.

Reviewer 2 Report

Comments and Suggestions for Authors

Review of Reinforced NEAT Algorithms for Autonomous Rover Navigation on Multiroom Dynamic Scenario

The study comprehensively examines the effectiveness of NEAT (NeuroEvolution of Augmenting Topologies) algorithms in autonomous rover navigation within multi-room dynamic scenarios. In particular, the clear demonstration of the positive effects of the transfer learning approach on performance supports the study's provision of significant contributions to the literature.

Below are some suggestions to improve the study:

• The reasons for selecting the parameters used in the NEAT algorithm and the effects of these parameters on performance should be explained in more detail.
• More discussion should be added on how simulation results can be transferred to the real world.
• Comparisons with similar studies in the literature should be expanded. Direct comparisons with previous research can be made to more clearly demonstrate the originality of the study.
• The selection and optimization process of the NEAT algorithm's parameters should be detailed.
• A literature review of methods that can be used for optimization should be presented. Examples of studies^1,2 related to optimization in the literature can be provided.
• The limitations of the algorithm should be discussed more explicitly.
• The differences between standard learning and transfer learning should be compared more comprehensively, and it should be clearly stated in which situations transfer learning outperforms standard learning.
• There are studies³ in the literature that explain the concept of Transfer Learning in more detail. Based on these studies, the concept of Transfer Learning can be explained more clearly.
• Heatmaps and other visuals related to transfer learning should be analyzed in more detail. In particular, more explanations are expected to be added on how transfer learning works and its specific effects on performance.
• Further analysis should be conducted on the reasons for fluctuations in fitness scores, and how these fluctuations are related to the NEAT algorithm should be discussed.
• The manuscript should be reviewed for grammar and language correctness.

Aydemir, F., & Arslan, S. (2023). A system design with deep learning and IoT to ensure education continuity for post-COVID. IEEE Transactions on Consumer Electronics, 69(2), 217-225.

 

 

Author Response

Comments 1: The reasons for selecting the parameters used in the NEAT algorithm and the effects of these parameters on performance should be explained in more detail.
Response 1: Addressed in lines 225-227 and 235-239.

Comments 2: More discussion should be added on how simulation results can be transferred to the real world.
Response 2: Address in lines 803-813.

Comments 3: Comparisons with similar studies in the literature should be expanded. Direct comparisons with previous research can be made to more clearly demonstrate the originality of the study.
Response 3: Agreed, and it was our original approach. Direct comparisons are challenging since the optimization of the specific environment cannot be found in the literature review. However, we did aim to highlight similar goals in other applications and environments to demonstrate the feasibility of the methodology. However, we have expanded our discussion in the literature in lines 69-87.

Comments 4: The selection and optimization process of the NEAT algorithm's parameters should be detailed.
Response 4: Addressed in lines 227-234.

Comments 5: A literature review of methods that can be used for optimization should be presented. Examples of studies related to optimization in the literature can be provided.
Response 5: Addressed in lines 88-110

Comments 6: The limitations of the algorithm should be discussed more explicitly.
Response 6: Address in lines 819-823.

Comments 7: The differences between standard learning and transfer learning should be compared more comprehensively, and it should be clearly stated in which situations transfer learning outperforms standard learning.
Response 7: Addressed in lines 770-784.

Comments 8: There are studies in the literature that explain the concept of Transfer Learning in more detail. Based on these studies, the concept of Transfer Learning can be explained more clearly.
Response 8: Addressed in lines 770-785 (with comment #7)

Comments 9: Heatmaps and other visuals related to transfer learning should be analyzed in more detail. In particular, more explanations are expected to be added on how transfer learning works and its specific effects on performance.
Response 9: Addressed in lines 770-785  (with comment #7)

Comments 10: Further analysis should be conducted on the reasons for fluctuations in fitness scores, and how these fluctuations are related to the NEAT algorithm should be discussed.
Response 10: Address in lines 791-795.

Reviewer 3 Report

Comments and Suggestions for Authors

This study is a study on autonomous driving in a limited space using NEAT Algorithms.

 

In an emergency situation, an autonomous rover is deployed in a closed space such as an apartment or building to explore the entire area and return to the initial position, thereby configuring and learning an autonomous driving algorithm for efficient information acquisition of indoor spaces for fire suppression.

The introduction focuses on theoretical descriptions rather than the appropriateness of the NEAT Algorithms used in this study, and there are many parts where the information on the learning model and the results of learning are too insufficient and the interpretation of the results is not appropriate. In addition, the review of the results is too poor, so it is thought that overall revision is necessary.

Author Response

Comments 1: In an emergency situation, an autonomous rover is deployed in a closed space such as an apartment or building to explore the entire area and return to the initial position, thereby configuring and learning an autonomous driving algorithm for efficient information acquisition of indoor spaces for fire suppression.

The introduction focuses on theoretical descriptions rather than the appropriateness of the NEAT Algorithms used in this study, and there are many parts where the information on the learning model and the results of learning are too insufficient, and the interpretation of the results is not appropriate. In addition, the review of the results is too poor, so it is thought that overall revision is necessary.

Response 1: Introduction concerns: The introduction provides the necessary theoretical foundation to establish the significance of autonomous navigation in firefighting scenarios. It highlights the complexities of multi-room exploration and the relevance of adaptive algorithms. NEAT is explicitly introduced as a robust solution due to its evolutionary design, which is particularly well-suited for dynamic and unpredictable environments like those encountered in firefighting missions. We will further refine the introduction to emphasize the unique strengths of NEAT for this application.

Why NEAT: NEAT was selected for its ability to evolve both neural network topology and weights, making it ideal for environments where the optimal network structure is unknown or highly variable. As shown in the results section, NEAT facilitated significant improvements in rover navigation strategies, with fitness scores and task completions improving markedly across generations. This aligns with the algorithm’s intended adaptability to dynamic indoor environments during emergency scenarios. Additional emphasis will be added in the discussion to outline why NEAT suits this study.

Learning model: We describe the learning process in detail, outlining the NEAT algorithm’s configuration, parameters, and reward system. The results section comprehensively discusses fitness score trends, heatmap analyses, and task completion metrics across multiple scenarios and generations. For example, the comparative performance of standard learning and transfer learning is thoroughly analyzed, with clear evidence of transfer learning’s benefits in early adaptation and task completion efficiency, per other reviewers' comments.

Interpretation of discussion of results: We present results using quantitative (e.g., fitness scores, task completion rates) and qualitative (e.g., heatmaps) analyses, demonstrating how NEAT and transfer learning improve rover performance in multi-room scenarios. For instance, the heatmaps highlight key navigation patterns, and the fitness trends reveal progressive optimization over generations. Hence, we have expanded the Discussion section in the updated draft.

Limitation of results: We appreciate the reviewer’s feedback on needing a more comprehensive discussion. We have addressed this by expanding the discussion based on other reviewers’ comments. The updated discussion now includes an in-depth exploration of the NEAT algorithm for the proposed environment and application, detailing the testing of neural networks and the tuning process for each network configuration. While we acknowledge the value of exploring additional techniques and network designs, we argue that transfer learning offers a pragmatic approach, particularly in constrained computational resources. Transfer learning enables quicker implementation by leveraging pre-trained models, reducing the time required for network design, configuration, and testing. This efficiency is crucial in applications such as emergency response, where rapid deployment is paramount and computational resources may be limited. This balance between efficiency and performance makes transfer learning a practical and effective choice for the proposed application.

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

All the comments are well addressed and now paper may be accept in current form.  

Author Response

Comment: All the comments are well addressed and now paper may be accept in current form.  

Response: Thank you

Reviewer 3 Report

Comments and Suggestions for Authors

.

Comments on the Quality of English Language

.

Author Response

Comments: none

Response: No comments were provided for us to address. We conclude that the reviewer accepts paper revisions as they are.