Multiagent Manuvering with the Use of Reinforcement Learning
Round 1
Reviewer 1 Report
The authors present a multi-agent based approach to define and solve the dynamic cooperative maneuvers problems for autonomous driving. They implemented their method for different scenarios and the results show that their algorithm/technique is succesfull to find the solution for these scenarios and its adaptability is sufficient. This paper can be accepted after some minor revisions, which are listed below.
1- Abstract was well written but the authors should add the key contribution(s) of this paper and one or two of the key findings of this study in the abstract section.
2- The authors gave related literature in the text but they should add the scientific contributions and difference(s) this paper according to the previous studies they stated in the text.
3- There are some spelling mistakes in the text. Please recheck all the text for correction of this type mistakes.
4- The caption of the Figure 6 is too long. The architecture of the used neural network was given in this figure. But, the algorithm was explained in the figure caption. The authors should add this explanation into the text to improve the readability of the paper.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments
This paper presents an approach, combined with the PPO algorithm and multi-head attention mechanism, to solve the dynamic cooperative maneuvers problem in autonomous driving applications. The proposed method is tested by simulations on different scenarios. Multi-agent maneuvering problem is attractive. Unfortunately, the innovative is weak and the contributions are not sufficient for publication in the journal. Some specific comments are listed as follows.
Major concern:
1. Dynamic maneuvers problem has been deeply studied in several fields, such problem can be solved by various algorithms, not just the algorithm used in this article. Besides, although the effectiveness of the presented algorithm is verified through several experiments, the comparison with other methods is not enough.
2. Multi-agent reinforcement learning algorithm via multi-head attention mechanism has been utilized for a long time. Besides, the bicycle model is simple and common in autonomous driving. The article only combines the above two aspects, resulting in a lack of innovation.
Other concern:
1. There are some mistakes in the system description.
2. Specific details about multi-agent RL algorithms have not been fully described. In addition, the meaning of “All agents share the same policy parameters.” is not clear.
3. In Section 5.2, it is suggested to explain the impact of time reading, the ratio of actual distance and shortest path distance on the reward.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
This paper presents an approach for defining, solving, and implementation of dynamic 1 cooperative maneuvers problems in autonomous driving applications. The paper is well presented and the following comments can be considered to improve its quality.
1- What are the specs of the machine used to run the experiments?
2- If the configuration parameters of the conducted experiment are summarized in a table, it will be much clearer to readers.
3- A comparison with other methods is lacking.
4- Clearly specify the merits of the proposed method in comparison to the existing methods.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
No further comment. The revision can be accepted now.
Reviewer 3 Report
Authors addressed the suggested comments. Thank you
