Task Allocation of Multiple Unmanned Aerial Vehicles Based on Deep Transfer Reinforcement Learning

Round 1

Reviewer 1 Report

The task allocation of multiple UAVs is a key problem when they perform complex tasks considering the efficiency and quality of completing tasks. Most of existing research works apply heuristic methods to solve the problem of UAV task assignment, which have the limitation of local optimization. Moreover, the reinforcement learning methods applied in the problem face the challenges of efficiency, such as slow algorithm convergence and long running time. 

 

This paper proposed a deep transfer reinforcement learning algorithm based on QMIX to solve the multi-UAV task assignment problem. The algorithm combines the experience of the source task in the new task and accelerates the convergence speed by migrating the network parameters learned by QMIX in the source task. The experimental results show that the algorithm is significantly better than the traditional heuristic method with the same running time. 

This paper is addressing an important challenge. The paper is well organized and easy to follow. The proposed approach and its experimental results are very convinced. 

My major concern is that the experiment is relatively easy, only the 5 UAVs, 10 target points and 3 kinds of resources. I recommend the authors to consider more complicated experiment setting. The other issue is the grammar of this paper. The authors have to polish the paper significantly, for example "6. Algorithm Verification and Analysis" should be "6. Conclusion". 

I recommend to accept this paper after the authors addressing the above problems. 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript works with the problem of task assignment for multiple UAVs through of the deep reinforcement  learning algorithm based on QMIX.

In general, manuscript is easy to understand. However, there are parts which should be revised:

1) Some abbreviations and acronyms are not defined;

2) The manuscript contains some errors in grammar which need to be corrected;

3) The variables of the equations must all be defined throughout the text;

4) Improve the quality of Fig 3;

5) Better explain the reason for the test in Fig. 8 be as oscillatory when compared to the results presented in Fig. 10. Why this occurs when the number of UAVs is less than the number of target points;

6) There were moments in the test of Fig 8 that the algorithm stayed in local minima;

7) For how many different scenarios was the algorithm tested? Just the ones presented in the article? More experiments should be made and the achieved results should be described more clearly.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Linha 243 = Figure ??

You cite that in the manuscript: In this paper, the algorithm is tested on 5, 10, 20 UAVs and 10, 20, 30 target points, and the experimental results are in line with expectations.

Why are results with 20 UAVS and more point target not displayed? Either by graphs or by tables. This information is provided in the text, but is vague, without conclusion. You just mention that the results are in line with expectations, but does not display numerical comparison data.

Author Response

Please see the attachment

Author Response File: Author Response.docx