An Improved Soft Actor–Critic Task Offloading and Edge Computing Resource Allocation Algorithm for Image Segmentation Tasks in the Internet of Vehicles

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This is a high quality article with good preparation, formatting and attractive presentation. There are some errors that required rectification as below:

1. The similarity index is 20% which is at the boundary of acceptance. Suggest author to rephase or modify the sentences with similarity that more than 1%.
2. The abstract should give summary and highlight the contribution of results obtained in the current studies. 
3. Avoid using personal pronounce term like we, out, etc in the technical paper writing. Suggest to relook into the entire article and correct it. 
4. The current proposed solutions should be benchmarked with other available methods and highlight the current design uniqueness. 
5. Offloading required more complex distribution network and infrastructure that eventually increase the overall CAPEX and OPEX. 
6. How to prevent overload or peak demand during bad traffic scenario? What is the worst case scenario. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors 1. This research is an interesting topic, but it needs to describe latest related works in more detail.   2. What specific improvements should the authors consider regarding the methodology? 3. Is it a conclusion consistent with the evidence and arguments presented, and does it address the main questions? 4. Is the reference appropriate? and also, references must have applied to related MDPI papers.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper addresses the challenge of offloading image segmentation tasks and allocating edge computing resources for the Internet of Vehicles (IoV) by proposing an edge computing solution that uses an improved Soft Actor-Critic (iSAC) algorithm with maximum entropy, enhanced by Prioritized Experience Replay (PER) to boost learning speed and stability. An integrated computing and networking scheduling framework is also introduced to minimize task completion time. Simulation results show that iSAC outperforms baseline algorithms in both accuracy and efficiency, supporting its use in latency-sensitive IoV applications. Please consider the comments below to further enhance the quality of this paper.

There are still several editorial issues in this version of the paper. Please review and correct all possible errors.

Most of the cited papers in the related work section are outdated. This section must be revised by incorporating more recent literature, preferably up to 2025. Additionally, a summarized list of contributions from these updated references is recommended. This list should include an analysis of their system models, solutions, performance metrics, and limitations.

There are many system entities shown in Figure 1 (e.g., satellites, rockets, VR, robots, industrial systems) that must be clearly described. The authors should revise both Figure 1 and Section 3.1, which is currently vague. The figure should be slightly redesigned to better reflect the hierarchy of the proposed system model which, notably, is not entirely novel. Are you primarily focusing on IoV?

Based on the model's hierarchical structure, why are multi-agent approaches not considered to prevent potential bottlenecks at processing servers while maintaining system resilience?

Today’s connected vehicles are capable of more than just handling their own image segmentation tasks. They can also receive and process tasks from other vehicles or from servers in a distributed manner. This could reduce the processing burden on servers while leveraging the computation capabilities of modern vehicles. Additionally, the "agent" must determine which servers will handle the tasks, which can increase system overhead. Should the system model be adjusted or redesigned to reflect this?

In Section 3.2, the role of cloud server(s) is omitted. In which scenarios or use cases would tasks be offloaded to the cloud?

What is the queuing model used in the system?

Are task diversity and priority handling addressed in this paper?

How are system overheads handled in your proposed model?

When considering IoV, critical factors include vehicular mobility and handoffs. If these are not addressed, the work becomes a special case of the broader edge computing paradigm. In that case, both the title and focus of the paper should be adjusted accordingly.

The names “Methods” in Section 6.2 should be renamed to reflect the actual algorithm names or a clear distinction between baselines and the proposed solution.

Furthermore, additional performance metrics, such as energy consumption, resource utilization, and acceptance ratio, should be included in the evaluation.

In Figure 6, all algorithms perform similarly when task sizes are below 80 MB. Therefore, it is suggested to expand the task size in the evaluation to better demonstrate the efficiency of the proposed solution.

How do the authors ensure that the simulation results are statistically reliable and that PER-iSAC performs optimally?

Moreover, the evaluation only compares PER-iSAC with PPO and standard SAC. It omits comparisons with other recent approaches (e.g., hybrid offloading methods, federated learning, multi-agent DRL, LLM-based approaches) that could offer competitive or superior performance. Please address this limitation in the revised version.

Lastly, please include a rebuttal that clearly indicates where the authors made changes (e.g., page numbers, line numbers, and specific modifications), and ensure these changes are reflected both in the rebuttal and the main text. Thank you

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

This ppaper titled " An improved Soft Actor-Critic Task Offloading and Edge Computing Resource Allocation Algorithm for Image Segmentation Tasks in the Internet of Vehicles" proposed an efficient edge computing resource allocation system based on an improved model-free Soft Actor-Critic (iSAC) algorithm  with maximum entropy to investigates the offloading and allocation of image segmentation tasks for connected cars in IoVs supported by computing power networks. This is an interesting study and the manuscript can be accepted for publicaton in WEVJ journal after the authors have responded to the following comments.

1. The abstract contains relevant information but needs improvement. The background information is overly detailed. The authors should reduce the background information and summarize what they have done in their study. The authors should also add the significance or practical engineering application of their study at the end of the abstract.

2. The introduction covers essential background but lacks depth and balance. The related work section is comprehensive but lacks a focused comparison with the most recent works on DRL-based task offloading in vehicular edge computing, especially those published in the last two years.

3. There are inconsistencies in the use of terms (e.g., "edge computing," "edge intelligence," "roadside servers") and abbreviations. All acronyms and abbreviations should be defined upon first use, and terminology should be used consistently throughout.

4. In-text citations: In-text citatons starting with a reference should start with the authors name and then the number in a bracket. for example Li et al. [18]. The authors should do same for reference number [12], [13], [14], [15] etc.

5. Page 3, Line 105: "Reference [14]". The name of the referenced authors should replace Reference.

6. The equations should be referenced in the text preceeding them in the manuscript.

7. Figure and figure for referencing the figures in the manuscript. The authors should maintain consistency. Figure 1 , figure 4.

8. The experimental validation discussion lacks detailed descriptions of important things. The experimental setup, including hardware, simulation parameters, and dataset characteristics are not adequately articulated in the manuscript.

9. There is no validation on real vehicular data or deployment scenarios, which limits the practical impact of the findings.

10. The authors failed to discuss the scalability of the proposed iSAC algorithm to large-scale IoV scenarios with high vehicle density and task heterogeneity. Also the generalizability of the approach to other types of edge computing tasks beyond image segmentation is not addressed.

10. Authors should discuss the limitations of their study.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you to the authors for their efforts in addressing my comments. Although many of the important concerns have been deferred to future work, which is somewhat disappointing, I am satisfied with the way the authors have responded overall.

Reviewer 4 Report

Comments and Suggestions for Authors

I thank the authors for taking their time to address all the comments I raised in the last review. The manuscript in its current form is suitable for publication in WEVJ, and I recommend its acceptance.