Learning to Optimise a Swarm of UAVs
Masahiro Shibata
Round 1
Reviewer 1 Report
The authors propose an area coverage method by Unmanned Aerial Vehicles (UAVs). The authors devise the expression of objective functions and get good performances.
I think that the paper has good motivation. However, the paper has relatively many errors, in particular in English and mathematical expressions.
I think the authors get proofreading and should then read the content of the paper more carefully.
For example, I have the following comments.
page 2, line 57, “In section 5 are”: This expression is grammatically incorrect.
page 4, line 154:
proposes to improved
->
proposes to improve
page 6, section 3.1.3, line 9: The meaning of “cardinality” is unclear for me.
page 4, section 3.2.1:
- lines 2 to 3, “In the CCUS3O model, UAVs are considered to fly at a constant speed. The time is therefore equivalent to the distance travelled”: This explanation is unclear because there is no assumption that a UAV moves one unit distance in one unit time.
- It might be good to change the order of “Coverage time” and “Coverage rate” because, if so, readers may be able to understand what “coverage” means well.
page 7, section 3.2.2: In the current definition of O^(rate)(S), the value is always a non-positive number. Is it correct?
page 8, caption of Figure 3, line 2:
length. (3b)
->
length (3b).
page 8, lines 245 to 246:
based on one heuristic template (see section 1)
->
based on one heuristic template (see section 4.1.1)?
page 10, line 295:
Most of the literature considers
->
Most of the literature consider
page 10, section 4.2.2: The definition of Reward is O(S_{t+1}) – O(S_t), instead of O(S_{t}) – O(S_{t+1})?
page 12, section 4,2.3,
- line 4, “at each tine t is thus added”, This expression is grammatically incorrect.
- v_{t-\tou} is not explained in the definition of the 5-tuple.
page 13, line 374:
Lines 21-24 simply corresponds to
->
Lines 21-24 simply correspond to
page 14, caption of Figure 7: Swarm -> swarm
page 14, line 400, r \in {0,1}: In this expression, r takes zero or one, not a value between 0 and 1.
page 16, lines 440 to 441:
They however do not belongs to
->
They however do not belong to
page 16, paragraph 2, line 6, “So at each position is assigned”: This expression is grammatically incorrect.
page 17, Table 1: Normally, a caption of a table should be put above of the table.
page 17, line 482: asses -> assess
Throughout the paper: The authors sometimes use the phrase “the latter”, but this expression makes the paper unclear.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Dear authors,
the article "Learning to Optimize a Swarm of UAVs" presents the approach to optimizing the area coverage with a UAV swarm, i.e. automating the swarm's behavior in the context of a specific area reconnaissance mission (the so-called coverage of the area with swarm's flight routes). This is another article by the authors from the series on the optimization of drone swarm routes using "Multi-objective reinforcement learning".
The work is undoubtedly interesting and noteworthy, but requires significant revisions to make it understandable to a reader who is not a specialist in this narrow field. I have put my comments below. In particular, please add additional descriptions to the places I have indicated in the article.
I have placed the comments in the order that corresponds to the text of the article.
1. lines 55-58: "Section", not "section"
2. line 61-62: "no single definition of a swarm yet exists" - I'm not sure you can say something like that; please check the definition in one of the latest articles "Siemiatkowska, B .; Stecz, W. A Framework for Planning and Execution of Drone Swarm Missions in a Hostile Environment. Sensors 2021, 21, 4150. https://doi.org/10.3390/" s21124150 " The lack of an officially accepted definition does not limit us from defining the swarm in the way shown in the article you refer to.
3. In the "Related Works" section, there is no reference to other methods of planning swarm missions, e.g. based on integer programming methods. This is the well-known "Vehicle Route Planning with Time Windows" task class. The reference to these methods is all the more important as you use G connection graphs in your work. Another interesting work in this area is "Fatih Semiz, Faruk Polat, Solving the area coverage problem with UAVs: A vehicle routing with time windows variation, Robotics and Autonomous Systems, 2020, https://doi.org/10.1016/j.robot.2020.103435. " It is up to the authors to decide whether they want to mention other methods described widely in various works available in literature. However, I believe that the reference to VRP methods will enrich the article (see for example line 186).
line 188: Why do you use index "e" for a graph Ge and for a set of edges? If this is first letter of "environment", use this as superscript Ge.
line 191: If you use E, you may define function as dist:E->R
line 194: What is the interpretation of the communication graph? What does it mean when two vertices are connected?
line 197: Please change Dcom to Dcom . I think it is not clear when 3 letters are used where the indexes are usually placed.
line 205: C(G, k) - Aditional comments describing "the class of instances" are needed.
line 206: Is the last definition correct?
line 207: In my opinion, it is very important to precisely define what an area coverage is. There is no such definition before line 207 and the reader may misunderstand the definitions and methods provided in this article. Is this definition given in line 220?
lines 228-232: These two sentences are too complicated while the problem is not. Simplify it, please.
line 246: "see section 1" or "see section 4.1.1"?
line 256: f the UAV can fly without taking into account the edge of the Ge, what is this graph for?
line 323: "relu function" - Where can the reader find the definition?
line 336: relu - "rectified linear unit"? There is no definition.
line 358-359: "w" variable should be described here.
As I wrote at the beginning, introducing the suggested corrections will make the work easier for other people involved in optimization.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The paper was improved. So, it deserves acceptance.
Reviewer 2 Report
Dear Authors,
all my comments have been taken into account. I have no further comments. However, I do not undertake an assessment of the linguistic correctness of the work.
