Real-Time Collision-Free Navigation of Multiple UAVs Based on Bounding Boxes

Round 1

Reviewer 1 Report

Review Report – Electronics – 936751

Multi-UAV collision-free navigation based on bounding boxes

Generally, this study presents an interesting issue in the field of aerial robotics and autonomous navigation systems. The core of study is about collision avoidance methods, which is well supported by a Bounding Box Collision Avoidance (BBCA) technique. The study has a good chance of publication in Electronics MDPI. However, the following errors should be corrected first:

Remarks:

• The main question is that how easy is to implement the BBCA collision avoidance technique in different environment. For example, the flight of aerial robotics, UAVs, is usually affected by uncertainties like strong winds.
• Another question is about real-time systems. The schedulability in real time is not well highlighted. It is only few times mentioned in the abstract and the body of the paper, and so it needs more clarification. Just as a suggestion, the title can be also changed to “Real-time collision-free navigation of multiple UAVs based on bounding boxes” to reflect this point.
• The multi-UAV systems are studied in Section 4.2. However, the section should further discuss duplicate detection/resolution where multiple UAVs simultaneously detect/resolve the same collision/deadlock.
• At the start of Section 2, it is mentioned that “Collision (or obstacle) avoidance is a widely explored problem, especially in the terrestrial mobile robot field [2][3][4][5].”. The paper clearly gains benefit from a general discussion and citation about robotics systems, instead of limitation to mobile robots, to show importance of collision avoidance. For instance, it can be achieved by saying “Collision (or obstacle) avoidance is a widely explored problem in the field of robotics (e.g., [a][b] for robotic arms and [2][3][4][5] for mobile robots).”

[a] Resolution of deadlocks in a robotic cell scheduling problem with post-process inspection system: Avoidance and recovery scenarios, 2015 IEEE International Conference on Industrial Engineering and Engineering Management, Singapore, pp. 1107-1111

[b] A cross-entropy method for optimising robotic automated storage and retrieval systems, International Journal of Production Research, vol. 56, no. 19, pp. 6450-6472

• The authors should use one of American (e.g., centralized,decentralized, trivialized and initialized) or British (e.g., centralised, decentralised) English, not both.
• Authors do not need "etc." if they have already said "such as". Either one says that the list is not complete. Both makes it redundant. I can see this error in many sentences (Page 1, Line 30; Page 1, Line 37; Page 3, Line 141).

Author Response

We would like to thank the reviewer for his exhaustive analysis of the manuscript, as well as his constructive comments, that we are sure will contribute to improving the quality of the original document. Next, we answer the reviewer issues:

The main question is that how easy is to implement the BBCA collision avoidance technique in different environment. For example, the flight of aerial robotics, UAVs, is usually affected by uncertainties like strong winds.

The original manuscript describes our navigation system assuming that the mobile nodes will act as requested, with no margin for error. The reviewer very aptly asks about its implementation in situations where possible external disturbances prevent nodes from accurately executing the provided trajectories. To address this issue, we have included section 3.4 (lines 345-361) in the new version of the manuscript:

3.4 Robustness and safety margins

In the previous sections it has been assumed that the UAVs will follow (without margin of error) the paths provided by the navigation system. In this section, we analyze the robustness of the algorithm to possible errors due to “external” factors, such as the presence of gusty winds. Other “internal” factors, such as calculation errors (or excessive latencies) in the geolocation process, or even in the path tracking process itself, should have to be addressed from a different perspective.

The airspace manager is responsible for estimating the magnitude of these external disturbances and adjust the algorithm parameters to establish a safety margin, which is determined by the expression d_safety = r - τ * v_max. The difference between the conflict detection radius and the distance an UAV can travel during an iteration determines the safety margin. In the next iteration, the navigation system will detect the error and will provide an updated trajectory.

Additionally, the BBCA algorithm is a memoryless process. In this way, it does not assume the correct follow-up of the instructions provided in previous iterations. On the contrary, each node recalculates its trajectory from its current location, the destination location, and the current location of the neighboring nodes. Consequently, the algorithm is highly robust. If the nodes suffer unexpected translations, BBCA continues navigating towards the destination, without collapsing or trying to recover the lost trajectory.

Another question is about real-time systems. The schedulability in real time is not well highlighted. It is only few times mentioned in the abstract and the body of the paper, and so it needs more clarification. Just as a suggestion, the title can be also changed to “Real-time collision-free navigation of multiple UAVs based on bounding boxes” to reflect this point.

Indeed. As the reviewer points out, our proposal fits into the category of dynamic or adaptive algorithms, in which routes are dynamically modified during the flight, if necessary. Precisely, it is this behavior that provides the robustness discussed in the previous comment. The alternative to this approach would be to apply static routes (defined in advance) so that they ensure conflict avoidance at the cost of drastically reducing performance (the UAV can be forced to take a detour, even if there are no UAVs in the neighborhood).

We have modified the title of the manuscript according to the reviewer suggestion. In addition, since the term “real time” has different interpretations depending on the scenario in which it applies, we have included the following comment in the first paragraph of section 3 (line 184): “These algorithms dynamically modify routes in real time, as needed”.

The multi-UAV systems are studied in Section 4.2. However, the section should further discuss duplicate detection/resolution where multiple UAVs simultaneously detect/resolve the same collision/deadlock.

Our algorithm decomposes a conflict among multiple nodes into the combination of multiple conflicts between pairs of nodes. In an independent way, each node modifies its trajectory to solve 50% of the conflict, assuming that the other node will do the same. Several algorithms in the literature act in the same way.

This allows, among other things, that two nodes fly side by side maintaining the minimum safety distance, instead of repelling each other.

With the permission of the reviewer, we describe this situation not in section 4.2, but in the paragraph in section 3.2 (line 268) where this operation is performed. We have reinforced the original comment by including this issue.

At the start of Section 2, it is mentioned that “Collision (or obstacle) avoidance is a widely explored problem, especially in the terrestrial mobile robot field [2][3][4][5].”. The paper clearly gains benefit from a general discussion and citation about robotics systems, instead of limitation to mobile robots, to show importance of collision avoidance. For instance, it can be achieved by saying “Collision (or obstacle) avoidance is a widely explored problem in the field of robotics (e.g., [a][b] for robotic arms and [2][3][4][5] for mobile robots).”

[a] Resolution of deadlocks in a robotic cell scheduling problem with post-process inspection system: Avoidance and recovery scenarios, 2015 IEEE International Conference on Industrial Engineering and Engineering Management, Singapore, pp. 1107-1111

[b] A cross-entropy method for optimising robotic automated storage and retrieval systems, International Journal of Production Research, vol. 56, no. 19, pp. 6450-6472

We have added the suggested sentence at the beginning of Section 2 (lines 69-70), with the two references proposed [2][3], as well as a recent review on collisions in the robotic field [4]: S. Haddadin, A. De Luca, and A. Albu-Schäffer, “Robot collisions: A survey on detection, isolation, and identification,” IEEE Trans. Robot., vol. 33, no. 6, pp. 1292–1312, 2017.

The authors should use one of American (e.g., centralized,decentralized, trivialized and initialized) or British (e.g., centralised, decentralised) English, not both.

Authors do not need "etc." if they have already said "such as". Either one says that the list is not complete. Both makes it redundant. I can see this error in many sentences (Page 1, Line 30; Page 1, Line 37; Page 3, Line 141).

We thank the reviewer for these comments. We have revised the wording of the complete manuscript, assuming the use of American English everywhere for consistency (9 items). We have also revised the writing of all these open lists (14 items).

Reviewer 2 Report

This paper presents Bounding Box Collision Avoidance (BBCA), a simplified velocity obstacle-based technique that achieves a balance between efficiency and cost. The performance of the proposal is analysed in detail in different airspace configurations. Simulation results show that the method is able to avoid all the conflicts in two-UAV scenarios, and most of them in multi-UAV ones.

The paper is interesting and important to the research community. UAV is a recently hot topic that attract many researchers’ attentions. The paper can be further improved in following items:

1. The technical depth and width can be further extended. More cutting-edge literatures, especially in top-tier conferences and journals are expected in the reference. Also, try to analyze the major advantage and drawbacks of the cutting-edge.
2. The experimental results can be compared with similar approaches, instead of intuitive implementations.
3. Figures and tables can be optimized for better format.

Author Response

We would like to thank the reviewers for their exhaustive analysis of the manuscript, as well as their constructive comments that we are sure will contribute to improving the quality of the original document. Next, we answer the reviewer issues.

The technical depth and width can be further extended. More cutting-edge literatures, especially in top-tier conferences and journals are expected in the reference. Also, try to analyze the major advantage and drawbacks of the cutting-edge.

In our honest opinion, we have intended to provide a wide and depth background to support our proposal. Currently, the paper contains 37 references (12 of them from 2019/20), some of which may be considered cutting-edge from our modest point of view. The purpose of this work is not to provide an exhaustive review, but to describe some of the most representative techniques before detailing and analyzing the benefits of a new proposal. For the interested reader, we mention a couple of complete and recently published surveys.

Of course, we agree with the reviewer that we could extend the literature. In this way, we would be more than happy to include specific references proposed by the reviewers. In fact, we have already included two references related to collision management in robotics in its broadest sense suggested by another reviewer.

The experimental results can be compared with similar approaches, instead of intuitive implementations.

According to this comment, we have incorporated to our comparative a generic APF (Artificial Potential Field) algorithm. After implementing the APF behavior, we have simulated exactly the same configurations that had already been studied for the BBCA and direct routing algorithms. The results obtained by using APF have been incorporated into the corresponding plots (figures 15, 16, and 17), and they have been considered in the respective discussions (lines 369-70, 440-444, 456).

Figures and tables can be optimized for better format.

We apologize for these errors. Inadvertently, images in Figures 2 to 6 were converted to bitmaps, and their resolution considerably decreased when they were exported to PDF format. In order to solve this issue, we have redone them as vector images. We have also prevented all the tables containing algorithms from being cut into adjacent pages.

Round 2

Reviewer 2 Report

The revision is overall good and acceptable.