A Hierarchical Framework of Decision Making and Trajectory Tracking Control for Autonomous Vehicles
Round 1
Reviewer 1 Report
1. While the framework is validated through co-simulation, it's unclear if it has been tested in real-world driving scenarios. It's possible that the framework could perform differently or encounter unexpected challenges in the real world.
2. The paper mentions using quintic polynomial equations to generate paths, which may not be sufficient for more complex driving scenarios. Additionally, the paper doesn't discuss how the framework would handle non-ideal road conditions or unexpected obstacles.
3. It's possible that reinforcement learning could be a useful approach to address some of the challenges in autonomous vehicle decision-making, especially in dynamic and uncertain driving scenarios. In the context of the paper "Deep dispatching: A deep reinforcement learning approach for vehicle dispatching on online ride-hailing platform", the authors are specifically addressing the problem of dispatching vehicles on a ride-hailing platform. The authors could consider discussing the potential benefits and challenges of RL, as well as how it could be applied to their specific problem.
Author Response
Dear Reviewer,
Thanks for your valuable comments, we have provided a point-by-point response to the your valuable comments, Please see the attachment. Thank you again for your time and effort in reviewing the manuscript.
Best regards
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Dear authors,
From my perspective, that is really good paper with sufficient methodology. The novelty could be more articulated, but overall the paper is conducted in proffesional and academic way. I will recommend to accept the paper in present form.
Regards,
Author Response
Dear Reviewer,
Thank you for taking the time to review our paper and provide valuable suggestions. Based on your feedback on the novelty of the paper, we would like to clarify that most research treats decision-making, planning, and trajectory tracking control of autonomous vehicles as separate components. Our study integrates these separate parts into a systematic, hierarchical framework. This approach allows for a clearer understanding of the hierarchy and interaction of each module, as well as for the reasonable modeling of each part. Additionally, it provides an effective method for analyzing the influence of each module on the system performance and coordinating each part to ensure optimal performance of the entire system.
Reviewer 3 Report
Dear Authors,
My comments related to your study are as follows:
From my point of view, the proposition in the abstract written as “which affects its performance under complex driving environments” is a bit bold. It requires supportive arguments or direct references from the literature.
In the decision-making phase rule-based approaches are easier to apply. However, such approaches may experience difficulties in real-world scale complex environments (not well-defined environments).
Line 79 rapid exploring tree method à rapidly exploring random trees
Line 137 “Then a cluster of the trajectories is generated by the quintic polynomials which 137 connecting the starting point and endingpoints, from which the optimal one is selected 138 based on the cost function of each trajectory” -> how did you determine the final time to reach the end point? It is a bit unclear from the sentence.
Line 144 à It should be detailed why LQR is preferred for lateral and double PID is preferred for longitudinal.
Figure 3 à I think it should be revised slightly. Currently it gives the intention that only lane info is considered. However in the text above it is mentioned that state of the ego and the surrounding vehicles also considered.
You have 4 driving modes(Car Following, Free Driving, Left Lane Change and Right Lane Change). à How do you handle the cases that requires emergency brake?
Table 1 à These transition conditions are OK. It could work in your framework problemless however scaling up to real world scenarios could be challenging in such rule based approaches.
Line 190 à The structure of the “reference line” is a bit unclear to me. Is it a polynomial, Bezier curve or really a line? More details should be included here. (I also couldnot find the mathematical structure of your reference line in section 4.1)
Lines 195-202 à Do you consider the changes in the maximum drivable speed? For example what happens when maximum speed limit decreases from 120km/h to 90km/h in a certain section?
Section 3 à How do you actually execute the state change? Immediately or do you need a buffer or transition time between driving modes? From my point of view this is also critical information.
Line 256 à Using “l” as a variable for lateral displacement made it a bit hard for the readability of the text. Of course it is the authors` decision at the end, but I saw frequently “d” for it.
Section 4.3 à Structure of the longitudinal trajectory was not given or mentioned detailed enough.
General question à How do you handle infeasible trajectories? The trajectories that could not be realized by EV. Do you have any validation checks in the phase of trajectory planning or path planning?
Line 329 à Typo in s2.
Figure 11 à As it is, two PIDs seem to be open-loop (from the figure). Please also consider demonstrating the feedback loops explicitly.
Figure 11 à It is not directly clear from the figure how you manage the model mismatch between the inverse model and the actual inverse behavior of the system.
Best regards,
Author Response
Dear Reviewer,
Thanks for your valuable comments, we have provided a point-by-point response to the your valuable comments, Please see the attachment. Thank you again for your time and effort in reviewing the manuscript.
Best regards
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Accept in present form
