Design of Unsignalized Roundabouts Driving Policy of Autonomous Vehicles Using Deep Reinforcement Learning

Round 1

Reviewer 1 Report

This paper proposed a driving policy by combining the Soft actor-critic algorithm with interval prediction and self-attention mechanism, which aims at achieving safe driving of ego-vehicle at unsignalized roundabouts. The topic is interesting and the proposed method is well developed. To further improve the paper, I have some specific comments as follows:

(1) The motivation of the paper is not clear. The introduction is too brief and it is not well discussed why the DRL approach is adopted for the considered task.

(2) DRL based methods can be well developed for simulation scenerios, but the sim2real problem is a great challenge. This issue needs to be well addressed.

(3) Ablation study is missing.    

Author Response

We feel great thanks for your professional review work on our article. Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors provided an interesting approach to improve ego-vehicle driving safety by training the ego-vehicles to safely navigate unsignalized roundabouts. Below, some comments are reported to enhance the paper's readability:

1) Section 2.2.: The authors introduced the roundabout obstacle vehicles. They stated that each obstacle vehicle is considered to be driven by a human driver and appears randomly in the roundabout scenario. In the real scenario, I wondering to know how the ego-vehicles can recognize the presence of obstacle vehicles in the roundabout. Should the obstacle vehicles be particularly equipped with advanced driving systems?

2) Does the roundabout scenario reflect any standard requirements?

3) Please add a discussion section to provide a comparison of your results with the existing literature, limitations, and, eventually, future directions of your study.

4) Recently, there is an increasing interest in proactive road safety assessments based on the examination of the characteristics of the road to identify the presence of risk factors (i.e., https://doi.org/10.1016/j.aap.2022.106858 was carried out for roundabouts). How may your study be of support for such safety approaches? Can be the proactive approach and the one you proposed be compared or potentially combined? Please add a comment in the discussion/conclusion section.

- Line 72, change "Carla" in "CARLA".

Author Response

We feel great thanks for your professional review work on our article. Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

All the comments have been well addressed and the paper is in a better shape now. The reviewer has no further comments and suggests the acceptance of this paper.