Time Minimization of Rescue Action Realized by an Autonomous Vehicle
Round 1
Reviewer 1 Report
The Introduction of the paper should describe what problems rescue action includes, and analyze the research status of these problems. Which parts of these problems will affect time minimization, how about the existing methods, and what improvements are proposed for the existing problems in this paper. The introduction should provide sufficient background and include more relevant references. In the experimental verification, it is suggested to analyze the rescue action of a specific scene to show that the proposed method is indeed better than the existing methods.Author Response
Dear Reviewer,
Thank you for your analyse of our work. Your responses help us to improve quality of the paper.
“The Introduction of the paper should describe what problems rescue action includes, and analyze the research status of these problems. Which parts of these problems will affect time minimization, how about the existing methods, and what improvements are proposed for the existing problems in this paper. The introduction should provide sufficient background and include more relevant references. In the experimental verification, it is suggested to analyze the rescue action of a specific scene to show that the proposed method is indeed better than the existing methods.”
In the introduction we focused on fast travel problem itself and not on the general problem of rescue action. We think you have a point here; the general description of the problem will improve the paper. We extended the introduction with a part that describes rescue action problems. The parts of a rescue action that affect time-minimization are shown. The references to the papers that are focused on mobile robot design for rescue action are added (lines 47-56). The work on the experiment in real environment is ongoing and will be the subject of the further paper.
Reviewer 2 Report
An interesting paper, with a clear aim, good methodology, and a well-justified choice of algorithms. However, from my point of view, some points could be improved:
1.- Section 6.2. Lines 225-233. The authors explain the difficult balance among path definitions, quality, and performance of the algorithms (computation time). Have you considered a sensitivity study? For example, what level of paths would be necessary for a determined quality in the outputs or in relation to the computation time. A general comment (lines 232-233) is indicated, but I would like to know if some specific ratios or figures could be obtained.
2.- It is suggested to rewrite the following sentence in relation to optimization methods (lines 313-314): “They need just to compare solutions and choose the best ones.”
This is true for heuristic and meta-heuristic algorithms, like the genetic ones. However, there are other optimization methods, as all those algorithms with a convex objective function (simplex algorithm), which are able to obtain a global optimum (minimum or maximum).
3.- Table 6 and 7 should be Tables 1 and 2. They should be moved forward (where are cited) and add more comments about them to understand their importance and highlight the similarities and the differences among the considered parameters.
Author Response
Dear Reviewer,
Thank you for your analyse of our work. Your responses help us to improve quality of the paper.
Response to Review Report 2
“An interesting paper, with a clear aim, good methodology, and a well-justified choice of algorithms.”
We are glad to hear that. Thank you.
“1.- Section 6.2. Lines 225-233. The authors explain the difficult balance among path definitions, quality, and performance of the algorithms (computation time). Have you considered a sensitivity study? For example, what level of paths would be necessary for a determined quality in the outputs or in relation to the computation time. A general comment (lines 232-233) is indicated, but I would like to know if some specific ratios or figures could be obtained.”
Yes, we did a simple, empirical sensitivity tests, although the exact studies are planned to be done in the future. We prepare an experiment in the real environment now and the study of the PPGA will be included in the further paper. For now, the simulations are done on the PC-class computer and, as the PPGA is designed for mobile devices, which often provide much lower level of computational power, the study should be done on the target hardware to obtain the more accurate results. We added a note (line 253) that the sensitivity study will be included in further paper.
“2.- It is suggested to rewrite the following sentence in relation to optimization methods (lines 313-314): “They need just to compare solutions and choose the best ones.
This is true for heuristic and meta-heuristic algorithms, like the genetic ones. However, there are other optimization methods, as all those algorithms with a convex objective function (simplex algorithm), which are able to obtain a global optimum (minimum or maximum).”
You have a point here. It was an oversight. In this sentence we took into consideration only heuristic and meta-heuristic methods. We clarified it (line 337).
“3.- Table 6 and 7 should be Tables 1 and 2. They should be moved forward (where are cited) and add more comments about them to understand their importance and highlight the similarities and the differences among the considered parameters.”
The role of the parameters can be deduced from the method descriptions. Although the additional explanation, which you suggest, can improve the intelligibility of their role, so we moved the tables and added a short description of them (lines 429-435).
Reviewer 3 Report
I think that the authors do not present a complete method of tracking. In section 8 (Computer simulations) it is not evident how the set of control points is obtained. The method presented is not clearly explained above.
It is recommended to perform simulations with an autonomous vehicle, the biography mentions a rover-like wheeled vehicle.
Optimization and the concepts explained are quite generic, so they seem out of place. Perhaps they should be placed first.
Most of the work takes information from ref [20]
The coding of the individuals (possible solutions to the problem), the set of parameters (genes, chromosomes), etc., are not explained. It does not necessarily have to be constituted by the {0, 1}.
Anothers minor corrections
Line 25, Mobile robotics ref??
Line 25, rescue patrol ref??
Line 25, ref??
Line 34, rover-like robots ref??
Line 153, polylines of straight segments ref, b-splines ref and NURBS curves ref
Figure 3, 5, 8, 10, 11. Taken from another paper, it should been referenced.
Eq 7, subscripts (Pl , Pl+1), should not be i
Eq 12, bad aligned.
Line 198 - 200, the explanation must be clarified, rewritten.
Line 202, angle β, not defined in the Figures.
Line 214 - 217, The path generation algorithm consists of two parts: primary path generation algorithm (PPGA) and whole path generation algorithm (WPGA). It is recommended to synthesize pseudo codes from them using the Latex libraries (Algorithm2e, Algorithmic package) and then write the respctive explanations.
Section 6.3. Whole path generation algorithm - global path. The list of described steps could be detailed within one algorithm (Algorithm2e, Algorithmic package). A better description could be included in a section detailing the "Methodology" of the problem resolution.
The velocity profiles in Figures 15, 16 and 17 should be compared with the motion tests of an autonomous wheeled vehicle, in order to clarify the results.
Author Response
Dear Reviewer,
Thank you for your analyse of our work. Your responses help us to improve quality of the paper.
Response to Review Report 3
“I think that the authors do not present a complete method of tracking. In section 8 (Computer simulations) it is not evident how the set of control points is obtained. The method presented is not clearly explained above.”
We presented a complete description of the method of tracking. To improve clarity, we updated the Figure 7 to emphasize the control point list creation moment. The paths control points are not presented on the maps (in the example paths in Figure 14), because the result path is constructed by joining following short path segments. Marking all the control points, often numerous, will make the map unreadable. Our software does not allow to prepare maps with paths in better resolution for now.
“It is recommended to perform simulations with an autonomous vehicle, the biography mentions a rover-like wheeled vehicle.”
A rover for the field experiments is under preparation. The results of the experiments in the real environment will be presented in the next paper. This paper presents the current stage of the research.
“Optimization and the concepts explained are quite generic, so they seem out of place. Perhaps they should be placed first.”
Could you explain of which exactly part you are talking about?
“Most of the work takes information from ref [20]”
Our random search method for path finding is based on the same primary path generation algorithm as the GA-based method presented in [22], so obviously it should be presented here too. And as long as the paper presents the comparison of variants of these methods, a short description of the GA-based method has to be included. The paper adds new observations and conclusions of properties of the mentioned methods, thus it extends topics of the previous article. What is more, in this paper a new version of the GA is presented too, in which the fitness function is changed from approximating fitness function to travel time value, obtained from velocity profile.
“The coding of the individuals (possible solutions to the problem), the set of parameters (genes, chromosomes), etc., are not explained. It does not necessarily have to be constituted by the {0, 1}.”
The GA-based method was presented in [22]. Here only a necessary summary is presented. For more details please look in [22]. Although, for clarity, we have added the types of parameters that creates a chromosome (lines 142-143).
“Anothers minor corrections”
“Line 25, Mobile robotics ref??
Line 25, rescue patrol ref??
Line 25, ref??
Line 34, rover-like robots ref??”
Could you explain exactly to which parts you think that references are needed and why? It is not clear.
“Line 153, polylines of straight segments ref, b-splines ref and NURBS curves ref”
You are right. There should be a reference. We added one to the book [23].
“Figure 3, 5, 8, 10, 11. Taken from another paper, it should been referenced.”
It was an oversight. Proper references are added.
“Eq 7, subscripts (Pl , Pl+1), should not be I”
The naming in the paper is consistent. We do not understand, why you suggested that in subscripts should not be 'i'. Maybe you have in mind that ‘i’ looks there different. Unfortunately, it is the issue of the font and its size.
‘’Eq 12, bad aligned.’’
All the equations in the paper are center-aligned and we follow this consistently.
“Line 198 - 200, the explanation must be clarified, rewritten.”
Can you show the exact fragment that you suggest to be rewritten?
‘’Line 202, angle β, not defined in the Figures.’’
An oversight. We updated the Figure 6.
“Line 214 - 217, The path generation algorithm consists of two parts: primary path generation algorithm (PPGA) and whole path generation algorithm (WPGA). It is recommended to synthesize pseudo codes from them using the Latex libraries (Algorithm2e, Algorithmic package) and then write the respective explanations.”
We prepare the paper in MS Word (due to some reasons). It not provides a good environment for writing algorithms. Inserting listings prepared in Latex as pictures would not be a good practice.
“Section 6.3. Whole path generation algorithm - global path. The list of described steps could be detailed within one algorithm (Algorithm2e, Algorithmic package). A better description could be included in a section detailing the "Methodology" of the problem resolution.”
The resolution of the problem is mentioned in the method description and we think that is no need to duplicate it in the “Methodology” section. It is better, when this section is about only the methodology of simulation experiments.
To both above comments: The mentioned algorithms (lines 298-306) and (lines 356-375) are described in details in [22]. Do you think a more accurate description is needed in this paper too?
“The velocity profiles in Figures 15, 16 and 17 should be compared with the motion tests of an autonomous wheeled vehicle, in order to clarify the results.”
As it is mentioned in the one of above responses, the real-world experiment is under preparation and the paper presents results of the current state of the research.
Round 2
Reviewer 1 Report
sounds good.
Reviewer 3 Report
The authors have improved the manuscript by adding more details about the Figures and another explanations. It is good to compare the performance of the proposed method with other [22], were the fitness function is changed from approximating fitness function to travel time value. The proposed method is only tested using simulation only, it is recommended that in the current stage of the research the set of computer simulations be added, under an environment simulator with the kinematic/dynamic model of the vehicle that will be used for the field experiments. The proposed method was applied to static environments, the authors should test the proposed method in dynamic and uncertain environments. The contribution of the proposed work may not be sufficient enough for journal paper in path planning nowadays. The submitted manuscript may be out of scope of the journal. The authors are suggested submitting the manuscript to journals related to Robotics (special issues: path planning, motion planning, avoidance, etc).
Author Response
Dear Reviewer,
The comparison of proposed method is already presented in the results. In the Table 3, there are collected results of the tests of the GA -based method from [22] (that uses the approximating fitness function), the GA that uses velocity profiler, the random search-based method that uses approximating fitness function and the RS-based that uses velocity profiler. Moreover, in the Tables 6 and 7, the results of the GA-based method from [22] are used as reference level for comparison of the performance of the all presented methods, both in expected robot travel time and computation time.
The dynamical simulation depends not only on path-planning methods, but also on navigation, obstacle detection and path-following algorithms. To make the results useful and comprehensible, it would be needed to include description of the used methods too, which would make the paper much longer and complicated. Since it is quite complex now, adding more content can make the paper difficult to understand. We decided to divide our work and present the current stage of the research in this article to maintain readability. The real-world experiment is under preparation now, i.e. we are constructing an autonomous rover based on ATV with combustion engine. The results with used algorithms and description of the rover will be included in further paper.
The optimized path can be obtained quite fast, so it can be used without modifications in the case of detection of major differences between the map in memory and the state of the real world. If a new obstacle is detected far enough, the rover can slow down, calculate a new path from its current position and continue the ride. Although, as it was mentioned, it is the part of the path-following algorithm, not the path-finding one, so it is out of scope of this paper.
The global, time-optimal path-finding in the highly variable, dynamic environment is impossible due to inconstant conditions. Our scope of research is focused on the travel time optimization. In this case, the method has to be designed for pretty constant environment in the time frame of the operation, where dynamical obstacles appear rarely enough to treat them as exceptions or disruptions. For example, let consider a rescue action in the forest: the transport of an injured person. The forest is a quiet area, where animals and people are rarely spotted (with exceptions, of course). The environment not changes fast, so it can be assumed as constant in the time period of considered rescue action. Thus, the appearance of an unexpected changes can be considered as an exception and be neglected in the path-finding algorithm, and be handled by path-following method, which is higher level algorithm in comparison to the discussed one.
The paper describes actually 3 new methods built on the same base (PPGA, WPGA), which differ in optimization algorithm and fitness function. The presented methods are designed especially for mobile robots with low-end components. The optimization criterion is travel time (which is taken directly from velocity profile generator or the time-focused quality is approximated using path parameters), not on a path length, which is the approach not spotted in the literature. What is more, the methods with velocity profile fitness function are based on heuristic knowledge. Thousands of years of humankind development gave us knowledge of walking, running, riding animals and driving vehicles, which we have used to improve the algorithm, i.e. the velocity profiler slows the robot down near obstacles to improve accuracy of following the path, reducing the probability of a collision. Another advantage of the paper is usage of a velocity profile, which generation is often omitted in other papers.
Round 3
Reviewer 3 Report
The manuscript is well written and organized and has reasonable merit. It can be considered for acceptance in Electronics. However, it should be noted that , the computer simulations were indicated as experiments. Lab and field experiments are much more desired, so the authors are suggested to carry out field experiments for a next work.
