Double Deep Q-Network with Dynamic Bootstrapping for Real-Time Isolated Signal Control: A Traffic Engineering Perspective

Round 1

Reviewer 1 Report

This paper proposed a methodology that uses a double deep Q-network as a reinforcement learning method to find a solution to isolated intersection signal control. Although the authors referred to many previous studies, they failed to demonstrate their academic and practical contributions differentiating from other studies. I cannot determine whether this study is good, even though the developed method shows better results than the FACT method. My recommendations are as follows.

-      The literature reviews are a little bit wide and general. It is better to focus on the previous studies on isolated intersection signal control to demonstrate why this study should be conducted.

-      The literature reviews and theoretical parts would be better to be concise and the technical procedures to implement the proposed methodology to VISSIM or field would be better to be expanded in detail. For example, how to implement the grid-based state representation to VISSIM and fields?

-      Please answer the following questions. 1) The proposed methodology utilizes 150m of the state observation zone that consists of a series of 4m long grids for each approach but the FACT method utilizes only a detector for each lane. Is this comparison reasonable? I think it is natural the proposed method shows better results. 2) Can the proposed method be implemented in practice?

-      Page 16, Line 547, 3s is used as a condition for phase end. It would be nice to present how the results change as you vary the 3s because 3s might not be the optimal value. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors, 
thank you for the interesting research, and paper about junctions, and traffic control. 

Chapter "Conclusion" should be changed because that form is unclear. Example: "Discussion" chapter is missing. Table 3 and Figure 7 are placed in the middle of the sentence which causes bad visual reception for the reader.

Algorithm 1 is extreamly unclear to read maybe better will be a flowchart? 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

As I addressed earlier, how the proposed methodology is implemented in VISSIM should be explained in detail. Especially, the state observation zone should be described in detail. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf