Path Planning Based on Obstacle-Dependent Gaussian Model Predictive Control for Autonomous Driving

Round 1

Reviewer 1 Report

This paper proposed an optimal control path plan that enables safe driving, comfort, and low computational operation for autonomous driving by integrating the obstacle dependence Gaussian (ODG) and model prediction control (MPC). The ODG algorithm expresses the surrounding environment as a risk-based path using the information on lanes and the obstacle mobile robot. Optimal control is performed using QP to control the desired target direction under the constraints defined through the expressed information and vehicle modeling. An experiment was conducted on a mobile robot using the proposed method. The experiment confirmed that the ODG MPC algorithm operates in the direction of a low-risk path in the space measured by the sensor and path planning is more stable with less computational cost than the comparison algorithms. As a result, the proposed algorithm safely and quickly performs path planning to avoid surrounding mobile robots. The paper is interesting and well-written but has drawbacks:

1. The quality of Figures 19-36 should be improved.
2. The paper needs part Discussion. Authors should discuss the results and how they can be interpreted from the perspective of previous studies and of the working hypotheses. Future research directions may also be highlighted.
3. The scientific novelty of the paper should be stated clearly.
4. The results from the point of computational costs are not clearly presented.

Author Response

 

Please refer to Review report.

Author Response File: Author Response.pdf

Reviewer 2 Report

• The manuscript is very well written and contains interesting research.
• Please specify the content of the triple-citation [5-7] so that readers can understand the unique content to be sought in each, or alternatively reduce to the best reference to make the point being reinforced by the reference (here the positive effects on driving safety of ADAD).
• Please specify the content of the quadruple-citation [8-11] so that readers can understand the unique content to be sought in each, or alternatively reduce to the best reference to make the point being reinforced by the reference (here the efforts by Google and Tesla).
• Please specify the content of the quintuple-citation [12-16] so that readers can understand the unique content to be sought in each, or alternatively reduce to the best reference to make the point being reinforced by the reference (here research on obstacle avoidance).
• Please specify the content of the triple-citation [14,17,18] so that readers can understand the unique content to be sought in each, or alternatively reduce to the best reference to make the point being reinforced by the reference (here sampling based approaches in line 39 and disadvantageous inefficiency in line 50).
• Please specify the content of the quintuple-citation [20-24] so that readers can understand the unique content to be sought in each, or alternatively reduce to the best reference to make the point being reinforced by the reference (here curve-based methods).
• Please specify the content of the double-citation [23,24] so that readers can understand the unique content to be sought in each, or alternatively reduce to the best reference to make the point being reinforced by the reference (here mobile robotics).
• Please specify the content of the double-citation [26-27] so that readers can understand the unique content to be sought in each, or alternatively reduce to the best reference to make the point being reinforced by the reference (here obstacle collision).
• Please specify the content of the triple-citation [28-29] so that readers can understand the unique content to be sought in each, or alternatively reduce to the best reference to make the point being reinforced by the reference (numerous parameter values in line 77 and local path planning in line 95).
• Figures are of very good quality with some minor issues that must be corrected to make the manuscript legible and useful.
  • Figure 8 uses capitalized X and Y (not explicitly unit vectors connected to x and y measurements). Especially with the ubiquitous use of lower-case x in state-space formulation, increased articulation is needed to be sure readers understand which variable x is meant by the authors in each instance. Please strongly state the nature of capital X and Y, especially as they are expressed in the state x vector using subscripts X and Y.  Please consider reminding the reader of this occasionally by adding the strong statement in several places, so the reader doesn’t have to flip through the pages in instances of potential confusion.
  • With apologies for bluntness, Figures 9, 11, 12, 13, 14, 18-36 are all essentially useless, since the figures contain font too small for legibility. Thus, the figures relay no information to the readers Merely as a technique often used by the reviewer, don’t let any text in figures become smaller than the text size in the figure’s caption (the smallest permissible text size in the manuscript template). When the authors are increasing the font size, please take a moment to also consider normalizing the figures’ text font name to the template’s standard (Palatino Linotype) to increase the professional appearance of the authors’ work and increase the attractiveness of the manuscript to the readership.
• The Introduction section 1.0 is very well written in broad terminology, but sections 1.1. and 1.2 are replete with too many acronyms, and the trend is modestly continued. Please make efforts to improve the manuscript’s readability by paying attention to how many acronyms are used in each sentence and paragraph. Please make improvements with an assumption the broad readership may not be very fluent in ODG and MPC, so if the acronyms definition was originally seven or eight pages earlier, we’ve increased the reader’s labors attempting to follow the developments presented.
• Please specify the content of the triple-citation [35-37] so that readers can understand the unique content to be sought in each, or alternatively reduce to the best reference to make the point being reinforced by the reference (the 2DOF bicycle model in line 205).
• Use of tables is excellent. Please consider making extreme cases bold font (e.g. smallest and largest distance; comfort level scores, etc. ) to help the reader quickly highlight those results.
• The discussion and conclusion section is weak, and the reviewer offers two improvements:
  • 1) Please consider establishing a baseline case and then listing “percent improvement or degradation” from the baseline as figures of merit (in most general terminology) for inclusion in this section as well as the Abstract.
  • 2) Please consider adding a very brief few sentences at the end of the conclusions highlighting the authors’ opinions about future research directions following this manuscript’s publication. The reviewer offers a desire to see the proposed methods compared to the six methods (including optimal, predictive, classical feedback, and combined instantiations) presented in the following reference applied to motion mechanics, but did not include the scope of parameters included in this manuscript being reviewed: Sands, T. Comparison and Interpretation Methods for Predictive Control of Mechanics. Algorithms 2019, 12(11), 232.
• There are a sufficient number of variable and nomenclatures used to make a list of acronyms a very useful addition to the appendix.

Author Response

 

Please refer to Review Report.

Author Response File: Author Response.pdf

Reviewer 3 Report

 

Dear Authors, thank you very much for submitting a paper. After reviewing heavily reviewing your paper i came to the following conclusions:
The paper should be accepted after some major revisions.

The paper is overall of good merit but lacks some information, state of the art literature, structural refinements and some additional explanations. In the following i will display what needs to be added to the paper:

1. General Objective and Abstract: Your paper displays a very interesting ideas in combination with the risk-based field and the MPC. Therefore the novelty of the paper is very high and the approach interesting. The objective of your paper is clear, you are describing a good introduction with a good overview of autonomous driving.  
What i am currently missing in the abstract is better display of the outcomes of this paper. Please enhance the abstract

 

2. State of the Art: The state of the art is good but is missing essential paper in different fields of plannig, MPC and risk based approaches. i recommend to add them to make the state of the art clear. In addition you need to dervice why MPC works best here:

 

Overall Path Planning Papers

T. Stahl, A. Wischnewski, J. Betz, and M. Lienkamp, “Multilayer Graph-Based Trajectory Planning for Race Vehicles in Dynamic Scenarios,” presented at the 2019 IEEE Intelligent Transportation Systems Conference - ITSC, Oct. 2019, doi: 10.1109/itsc.2019.8917032.

 

M. Werling J. Ziegler S. Kammel and S. Thrun "Optimal trajectory generation for dynamic street scenarios in a frenét frame" IEEE International Conference on Robotics and Automation pp. 987-993 2010.

 

C. Katrakazas M. Quddus W.-H. Chen and L. Deka "Real-time motion planning methods for autonomous on-road driving: State-of-the-art and future research directions" Transportation Research: Emerging Technologies vol. 60 pp. 416-442 2015.

 

MPC Path Planning (with and Important here: Learning MPC)

J. Kabzan, L. Hewing, A. Liniger, and M. N. Zeilinger, “Learning-Based Model Predictive Control for Autonomous Racing,” IEEE Robot. Autom. Lett., vol. 4, no. 4, pp. 3363–3370, Oct. 2019, doi: 10.1109/lra.2019.2926677.

 

U. Rosolia A. Carvalho and F. Borrelli "Autonomous racing using learning model predictive control" American Control Conf pp. 5115-5120 2017.

 

F. Borrelli P. Falcone T. Keviczky J. Asgari and D. Hrovat "MPC-based approach to active steering for autonomous vehicle systems" Int. J. Vehicle Auto. Syst. vol. 3 no. 2 pp. 265-291 2005.

 

Gaussian Path planning and Learning based MPC

 

L. Hewing J. Kabzan and M. N. Zeilinger "Cautious model predictive control using gaussian process regression" IEEE Transactions on Control Systems Technology 2019.

 

A. Wischnewski, J. Betz, and B. Lohmann, “Real-Time Learning of Non-Gaussian Uncertainty Models for Autonomous Racing,” presented at the 2020 59th IEEE Conference on Decision and Control (CDC), Dec. 2020, doi: 10.1109/cdc42340.2020.9304230.

 

Risk based approaches:

S. Kolekar, J. de Winter, and D. Abbink, “Human-like driving behaviour emerges from a risk-based driver model,” Nat Commun, vol. 11, no. 1, Sep. 2020, doi: 10.1038/s41467-020-18353-4.

 

Dunning, A. Ghoreyshi, M. Bertucco, and T. D. Sanger, “The Tuning of Human Motor Response to Risk in a Dynamic Environment Task,” PLoS ONE, vol. 10, no. 4, p. e0125461, Apr. 2015, doi: 10.1371/journal.pone.0125461.

 

3. Overall structure of the paper: The overall structure of the paper is good but it is missing a few parts. In addition, a few things need to be redone
- Please redo all images because they are from low quality. Almost all plots are pixelated or too small so nothing can be recognized. Especially images like figure 17 do not fit to the rest of the paper. The oval is too big.
- You are mentioning that you are using a Bicycle model + you are quoting H. Pacjeka in your references. It is unclear if you are calculating the tyre forces or not. Please add a list of the paramters you are using in the bycycle model. If you are using tyre forces please add the particular Pacjeka magic formula + the corresponding Paramters for your tyres
- Why is the maxium of the risk always directly in front of the vehicle? if  vehicles are around the opponent vehicle it will lead to a different distriubtion or? Please explain that kind of approach more in detail. The same is for figure 10 and 11. Driving outside of the lane is basically from high risk - yes. But not driving directly inside the lane.  Does the maximum in the middle make sense here?

 

4. Results:

• Please redo all  the plots because they are from low quality. Almost all plots are pixelated or too small so nothing can be recognized. In addition the numers and letters on both the axes are too small. All the text in images needs to be the same size as the text in the normal text.
  - Please create for all the experiments the same kind of setup for the results: 
  1* One plot shows the controller error
  * One plot shows the acceleration of the vehicle -> Explanation for stability
• One plot shows the path driven.
  The current plots e.g. distance and velocity do not explain your algorithms very well and dont provide a good insight

5. Discussion: The discussion part is completly missing here. You need to provide a detailed insight in the results. Why is your approach so good? What needs to be enhanced? What happens when you drive faster? Your current results show high accelerations which would lead to even higher peaks when the cars are driving faster. You need to explain how you can smoothen this. Please provide a completly new discussion about your findings.

 

6. Outlook: What is the outlook and what are additional experiments and extensions?

In addition i think to make the paper more complete it would be helpful for everyone if you open-source your code on Github or similar.

Author Response

 

Please refer to Review Report.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors addressed all my concerns. The paper can be accepted in the present form.

Author Response

 

Please refer to Review Report. 

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear authors, thanks for the revised version of your paper. 
I went over all your comments and i am happy to see that some of them where acknowledge and answered. Unfortunately there are a few things that are not solved yet.
1. The additional papers for the state of the art listed was not integrated completely so a few papers are still missing and i can recommend to add them so the the state of the art - especially from a path planning point of view - is more complete
2. The Additional list of parameters you are using is corrupted/misdisplayed and the PDF does only display the paramter descritpiton but not the real values. Please fill out the list with the parameters you are using in your calculation and setup so other research can test it.
3. I asked the question Q3b) where you are answering that the generated path does make a difference: This is correct, because you are driving slow and without any high lateral acceleration, therefore you are of course in the lininerized region of the tyre forces. Again here, you need to explain how your system can work when you drive faster, too. therefore a more holistic explanation of this model and its parameters is needed.
4. Why does the vehicle stop when the sum of ODG risk is too high? This looks for me like a freezing robot problem and needs to adressed in both the paper and in your answer too
5. I am happy to see that you added all of these discussion points in your paper so from that side i think we are already fine.

From my side its just some minor revisions that need to be done.

Author Response

 

Please refer to Review Report.

Author Response File: Author Response.pdf