A Double-Layer Model Predictive Control Approach for Collision-Free Lane Tracking of On-Road Autonomous Vehicles
Round 1
Reviewer 1 Report
The article proposes a double-layer MPC algorithm for integrated path planning and trajectory tracking of autonomous vehicles on roads. The Introduction is sufficiently detailed, it reviews enough sources to reveal the essence of the problem under consideration. Most of the reviewed sources are fairly recent.
Model Predictive Control and and the method for selecting lane waypoints are presented in Section 2. Section 3 details the formulation of the double-layer structure MPC proposed in this paper. Vehicle model design, collision avoidance constraint design and double-layer MPC structure design are presented in this Section in sufficiently detailed.
Section 4 describes on-road autonomous vehicle trajectory tracking experiments with static and dynamic obstacles. Experiments were conducted using MATLAB. The main question is how accurately the simulations correspond to the real movement of vehicles. Because the Figures show the movement and positions of the cars schematically. To what extent does this movement correspond to the technical capabilities and dynamic characteristics of real cars, how are road surface and traction parameters evaluated? In this case, there is a lack of links to the real traffic and tests under real conditions, or at least a discussion of what assumptions are made, what is evaluated and what is not evaluated. This would significantly strengthen the validity of the conclusions of the article and the paper itself.
All abbreviations should be explained where they are used for the first time.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper proposed a double-layer model predictive controller for lane tracking in autonomous vehicles. The system modelling, controller design and simulation results with static and dynamic obstacle scenarios have been performed.
In equation (2), k=1,…,no_obs has not been used. Also, safe_dist, pred_dist and F_load are needed to be mathematically defined. This requires to include the dynamic system modelling before covering the controller. Otherwise, you can just provide a general formula for the MPC as the additional constraints are covered in equation (26).
It is mentioned in line 173 that “the reference vectors for the input vector are zero vectors.” However, you did not include the reference vector in equation (1). Although this is in line with equation 1.
Define m, I, l_r and l_f in equation (3).
Define p and q in equation (8).
In line 320, it should be to convert equation (8) into the following.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
This paper proposes a hierarchical planning and control algorithm for automated vehicle collision avoidance, also considering rollover prevention. Matlab simulations demonstrate the effectiveness and performance enhancement of the proposed method w.r.t. Two benchmark methods. The authors can consider the following comments, which might further improve the draft in its current draft.
Major issues:
It is interesting to have predicted vehicle position considering the perception limit. However, the perception range can change dramatically per weather conditions, etc. Therefore, the prediction length in (2) will need to be adaptable, which implicitly requires the MPC formulation needs to be re-compiled online. How could this be possible considering the current automotive electronic/electric architecture?
In Fig. 1, ‘p_{ref_y} is selected per Y(t)’. What if the reference path is part of a circle, and there are more than one P_{ref_y} according to the same Y(t)?
In eq(12), ‘K represents the total number of the surrounding vehicles’ How to define ‘surroudning vehicles’? Any vehicle within the communication/perception range of the ego-car? If so, a varying number of constraints need to be compiled online, which yields the same issue as the first major issue.
Per table II, the required solving time is above 0.3s, however, the sampling period pf MPC is less than 0.1s, which means the proposed method cannot be implemented in real time. Please have hardware in the loop simulation, instead of Matlab simulation, added.
Minor issues:
Why there are ‘2’s in eq(4) and eq(7)?
In eq(5)-(6), why not consider longitudinal and lateral force coupling? \dot_\omega_z is clearly wrong.
Before eq(9), it should be ‘..convert Equation (8) into’
After eq(25), the sentence is incomplete.
What is \delta T in eq(27)?
In eq(28), why F_load <= F_\{z threshold}? Not >?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The draft in its current form is acceptable.
