Long-Duration Inspection of GNSS-Denied Environments with a Tethered UAV-UGV Marsupial System^†

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors The paper presents a tethered UAV-UGV marsupial system designed for long-duration inspection tasks in GNSS-denied environments. The authors develop a hardware architecture and propose a complete ROS-based software stack, integrating modules for localization, navigation, and trajectory planning. A key technical contribution is the adaptation of LiDAR Localization, which enables accurate pose estimation for both UAV and UGV, alongside a reformulated trajectory planner that explicitly incorporates tether constraints through line-of-sight feasibility checks. The developed software is made open accessed in the github and the system is validated via three representative field experiments . The manuscript has a well-structured framework, and the references cited are closely relevant to the research theme.  One recommendation is that the authors further discuss the system’s real-world applicability, addressing two key constraints:

• The method relies on pre-defined LiDAR maps, but uncharted GNSS-denied scenarios (e.g., post-disaster sites) lack such maps.
• The tether reduces UAV/UGV flexibility,vthe flexibility-performance trade-off is unaddressed.

 

Author Response

The paper presents a tethered UAV-UGV marsupial system designed for long-duration inspection tasks in GNSS-denied environments. The authors develop a hardware architecture and propose a complete ROS-based software stack, integrating modules for localization, navigation, and trajectory planning. A key technical contribution is the adaptation of LiDAR Localization, which enables accurate pose estimation for both UAV and UGV, alongside a reformulated trajectory planner that explicitly incorporates tether constraints through line-of-sight feasibility checks. The developed software is made open accessed in the github and the system is validated via three representative field experiments . The manuscript has a well-structured framework, and the references cited are closely relevant to the research theme.  One recommendation is that the authors further discuss the system’s real-world applicability, addressing two key constraints:

 

Thank you for your careful review. We are glad that the manuscript is interesting to you. We have considered your constraints. Please find the answers below in blue.

1. The method relies on pre-defined LiDAR maps, but uncharted GNSS-denied scenarios (e.g., post-disaster sites) lack such maps.

As the reviewer mentions, the current system requires a pre-built map for localization. The map is also used for trajectory planning to fulfill the inspection missions, which are then commanded to the marsupial system and executed autonomously. This assumption of a pre-built map is valid in many relevant applications, such as the periodic inspection of known structures.

But nothing precludes the system to employ a multi-robot SLAM method to explore and build a map in unknown scenarios. The planning and control systems can be used in that setup with no changes, provided both vehicles are localized in the same reference frame. Of course, we have not validated such an approach, which we leave for future work. This fact has been included in Section 7.1, final paragraph.

 

1. The tether reduces UAV/UGV flexibility, the flexibility-performance trade-off is unaddressed.

While it is true that tethering limits the UAV’s maneuverability and may constrain its flyable space, the extended flight duration enables our solution to carry out long-duration inspection missions, where steady flight is essential to ensure the validity of the measurements. In addition, the flight duration of our system is less sensitive to the payload onboard the UAV. This discussion has been added at the end of Section 6 Discussion.

Reviewer 2 Report

Comments and Suggestions for Authors

This article contains the results of preliminary experiments on the control of a system, which includes a ground robot and a tethered UAV. From indoor tests, the authors drew a number of important conclusions regarding the operation of the robots. The study utilized existing approaches, software, and hardware, which is likely why they are not described. The scientific novelty of the proposed solutions is not obvious. The style of the material is more reminiscent of a technical report or laboratory test protocol than that of a scientific article. Therefore, this material is not recommended for publication in a scientific journal.

 

Also, a number of specific comments should be made regarding the article:

- The assertion in line 45 that the use of a drone swarm is an alternative to increasing flight endurance is absolutely incorrect.

- Most of the survey part is devoted to general discussions of the problems solved by drones. In fact, only references 24-28 are devoted to similar approaches. The substantive review of analogs should be expanded, and the first twenty references to articles should be significantly reduced.

- The main contributions of this paper, formulated in terms 116-126, lack scientific novelty. Differences from similar approaches should be explicitly specified.

- Qualitative assessments such as "good precision", "very convenient alternative", "can significantly extend" and others are often encountered instead of quantitative ones.

- The experimental verification section presents the results of individual tests. A series of experiments should be conducted to substantiate the reliability of the proposed solutions.

- All experiments are conducted indoors, without exposure to wind and other disturbing factors. Therefore, the applicability of the proposed solutions outdoors is questionable.

- The authors correctly note the main problem with tethered UAVs, related to tether tension control. However, they do not take into account the energy expended on this control. In practice, there will be no ideal case; tether tension will always vary depending on the combined motion of the air and ground robots, as well as environmental conditions.

- Proposed solutions must be analytically described and justified.

Author Response

This article contains the results of preliminary experiments on the control of a system, which includes a ground robot and a tethered UAV. From indoor tests, the authors drew a number of important conclusions regarding the operation of the robots. The study utilized existing approaches, software, and hardware, which is likely why they are not described. The scientific novelty of the proposed solutions is not obvious. The style of the material is more reminiscent of a technical report or laboratory test protocol than that of a scientific article. Therefore, this material is not recommended for publication in a scientific journal.

Thank you for your careful review. The main goal of our paper is to present the design of a tethered UAV-UGV system from scratch so that the interested reader can learn from our experiences. The main novelty is the system as a whole. We present the whole solution, pointing out why each module was developed, and we test it in field experiments (now including experiments up to 2 hours) to validate its performance. As far as we know, this kind of experiment has not been reported before. We do think this information can be of interest for the scientific community,  and fits inside the scope of the Drones in the following categories (see https://www.mdpi.com/journal/drones/about): 

1. Design: power supply
2. Development: performance, planning
3. Applications: monitoring, architecture

In addition, we present two contributions over the state of the art in terms of localization and planning. First, regarding the planning algorithm, we have reformulated some of our prior developments to take into account the particularities of the proposed system, as detailed in Sections 4.1 and 4.2. Second, we have included an extended set of experiments as validation that demonstrate the reliability of our solution.

Finally, following the comment, we made some restructuring and rewriting of the whole manuscript so that it follows more closely the standards of publication in scientific journals. For example, we completely remade sections 2.5 Software and 5. Discussion in a more academic style. We hope that the new version meets the standard of the journal.

Also, a number of specific comments should be made regarding the article:

- The assertion in line 45 that the use of a drone swarm is an alternative to increasing flight endurance is absolutely incorrect.

You are right, our idea was to point out that we can achieve comparable results to a long endurance inspection task by using a team of drones flying simultaneously. This has been clarified in Section I. Introduction, page 1.

- Most of the survey part is devoted to general discussions of the problems solved by drones. In fact, only references 24-28 are devoted to similar approaches. The substantive review of analogs should be expanded, and the first twenty references to articles should be significantly reduced.

We have rewritten the introduction, reducing the first part that references general applications of the UAV and strengthening the part more related to our research. Please find the new state of the art in blue in the first section. We hope that the new version provides the readers with a more useful background. 

- The main contributions of this paper, formulated in terms 116-126, lack scientific novelty. Differences from similar approaches should be explicitly specified.

We have updated the contribution paragraph at the end of the Introduction section, explicitly comparing our system to existing approaches. We highlight the technological advances regarding flight duration, which have been validated in several field experiments. 

- Qualitative assessments such as "good precision", "very convenient alternative", "can significantly extend" and others are often encountered instead of quantitative ones.

We have rephrased the aforementioned sentences and made a thorough review of the whole manuscript to improve its quality. We expect that the new version meets the standards of the journal.

- The experimental verification section presents the results of individual tests. A series of experiments should be conducted to substantiate the reliability of the proposed solutions.

We have carried out three additional experiments and included the results in the revised version. In total, we present now seven different experiments in which our platform operated for more than four hours without any failures. We hope that the new set of experiments helps to demonstrate the reliability of our proposed solution. Please find the new experimental results in Section 5.

- All experiments are conducted indoors, without exposure to wind and other disturbing factors. Therefore, the applicability of the proposed solutions outdoors is questionable.

Thank you for your comment. Please note that the focus of the paper is the navigation of the marsupial system in GNSS denied areas. Therefore, its performance on outdoor scenarios is out of its scope. However we have included some remarks in this direction at the end of Section 7.1 Future work.

- The authors correctly note the main problem with tethered UAVs, related to tether tension control. However, they do not take into account the energy expended on this control. In practice, there will be no ideal case; tether tension will always vary depending on the combined motion of the air and ground robots, as well as environmental conditions.

As stated in the paper, the tension mechanism imposes a tension to the cable of 1 N. If we assume the robot to be at hover operation, the majority of the control efforts are used to counteract the gravitational force acting on the quadrotor, which is approximately 4kg*9.8 m/s² =39,2 N. This force should be incremented to take into account the tension. The worst case scenario is produced when the tension is applied in the same direction as the weight (i.e. downwards). In this case, the extra effort is less than 2,5% of the total, which can be considered neglectable.

Please find this discussion in Section 6, second paragraph from last.

- Proposed solutions must be analytically described and justified.

We have included additional details regarding the proposed system and the experiments. We hope these details are sufficient.

Reviewer 3 Report

Comments and Suggestions for Authors

This is a good paper with strong background and results.

I have some minor comments.

Heading 6.1 “Lessons learned” seems inappropriate in this context. I would recommend changing it to something different. Actually, most information in this section is unnecessary. Mentioning battery preparation and charging conditions is not really needed in the context of this research paper.

Also remove line 541 to 543.

Line 509-510 the authors say “To emulate the defects, we disposed of an array of twelve Augmented Reality (AR) markers with a….” Disposed is incorrect word here. Probably use words like “Deployed”

  The topic is not original. Several similar works have been presented on tethered UAV-UGV teams.
This topic has been explored quite a lot. However, the authors present a slightly more efficient method. It demonstrates an efficient way for tethered UAV-UGV heterogeneous team to work together for inspection.    conclusions are consistent with evidence presented.   References are appropriate..  

No comments on tables and figures.
 

 

 

Comments on the Quality of English Language

See my comments above.

 

Author Response

This is a good paper with strong background and results.

I have some minor comments.

Heading 6.1 “Lessons learned” seems inappropriate in this context. I would recommend changing it to something different. Actually, most information in this section is unnecessary. Mentioning battery preparation and charging conditions is not really needed in the context of this research paper.

Also remove line 541 to 543.

Thank you for your positive review. We have reformulated Section 6 according to your suggestions. We expect that the new section is more adequate and academic in style.

Line 509-510 the authors say “To emulate the defects, we disposed of an array of twelve Augmented Reality (AR) markers with a….” Disposed is incorrect word here. Probably use words like “Deployed”

Thanks for pointing out the typo. We have changed the description of the experiment and provided more details on the construction of the mission. Please find the new version in Section 5.3.

  The topic is not original. Several similar works have been presented on tethered UAV-UGV teams.
This topic has been explored quite a lot. However, the authors present a slightly more efficient method. It demonstrates an efficient way for tethered UAV-UGV heterogeneous team to work together for inspection.    conclusions are consistent with evidence presented.   References are appropriate..  

You are right that there is a growing interest in tethered UAV-UGV systems, with different works presented in the last years regarding different aspects like planning, localization, etc. . However, we believe our system as a whole is a novel contribution. In the paper, we provide  a detailed description of the techniques involved and both the software and hardware components. The paper includes a thorough validation of the system (with 7 experiments in the revised versions), including a new 2-hour long experiment with the system in continuous operation. We have revised the manuscript to better highlight this contribution and have explicitly included this point in the contribution statement at the end of Section I.

Reviewer 4 Report

Comments and Suggestions for Authors

- The paper presents a tether powered marsupial system that pairs UAV and UGV to enable long duration inspection in GNSS denied environments. The authors detail off the shelf hardware and an open source ROS stack that combines A LOAM mapping with Direct LiDAR Localization over a Euclidean distance field. A two step planner first uses RRT*, then a non linear optimizer that enforces a taut tether and line of sight. Field trials report more than one hour endurance, real time localization, and centimeter scale inspection. Novelty includes a reproducible system, multi robot DLL, tether aware planning, and released datasets and code.

- By replacing catenary collision checks with a  straight segment  between the UGV and the UAV, the planner accepts only LoS configurations, which implicitly ignores sag, incidental contacts with obstacles, and lateral forces induced by the cable. The control stack does not model cable dynamics, and the authors report that lateral tension perturbs attitude and creates snagging risks that required manual monitoring  during experiments, thus the safety margin and the ability to generalize to cluttered environments appear constrained, especially when potentially optimal paths are excluded by the procrustean LoS requirement. 

- Localization evaluation lacks an independent ground truth outside the system. Instead of aligning against external fiducials, the study uses the mean  distance from each LiDAR point to the nearest obstacle in the prior map as an error proxy, while that very map also serves as the DLL reference, which raises an endogeneity concern. Computation times are reported on a Ryzen 7 desktop rather than on onboard avionics, so direct extrapolation to aerial execution remains uncertain, and centimeter scale accuracy claims should be regarded as preliminary.

- Evidence on endurance and energy efficiency shows lacunae. Although the design targets roughly two hours, the endurance experiment is a manual flight slightly over one hour, and Table 3 on page 15 shows both packs near half charge with a total draw of 1635 Wh, without repetitions or component level power apportionment and without measured efficiencies across battery, inverter, and tether station conversions. The discussion acknowledges higher than expected consumption and attributes it tentatively to auxiliary loads or heat losses, yet no corroborating measurements are provided, fault cases such as tether power loss or winch malfunction are not exercised, and the TB55 backup offers only about ten minutes, so operational safety under failures is opaque.

- Autonomy and coordination appear limited, which constrains scalability. The mapping stage requires a short manual flight to gather A LOAM data, so missions at new sites are not fully automated, and coordination relies on stop and go  time synchronization between waypoints with both vehicles capped at 0.25 m per second, which elongates missions and invites drift when the environment changes. In Scenario 2, the UGV could not navigate autonomously because the floor was damaged, hence the cooperative evaluation in challenging terrain occurs only in Scenario 3 under more benign conditions, which leaves performance in larger and more anfractuous spaces unresolved.

- Inspection validation remains simplified and does not reflect field reality. Scenario 3 uses twelve ArUco markers as defect surrogates, the main outputs are centimeter scale position estimates and successful detection of all markers over three runs  totaling thirty five minutes, which provides function but limited verisimilitude. The design does not quantify core inspection indicators such as surface coverage, spatial resolution as a function of working distance, and reliability for true defects under occlusion or changing illumination, hence practical utility requires further assessment using real defects and explicit image quality metrics.

Author Response

- The paper presents a tether powered marsupial system that pairs UAV and UGV to enable long duration inspection in GNSS denied environments. The authors detail off the shelf hardware and an open source ROS stack that combines A LOAM mapping with Direct LiDAR Localization over a Euclidean distance field. A two step planner first uses RRT*, then a non linear optimizer that enforces a taut tether and line of sight. Field trials report more than one hour endurance, real time localization, and centimeter scale inspection. Novelty includes a reproducible system, multi robot DLL, tether aware planning, and released datasets and code.

We would like to thank you for your detailed review and we are glad that you appreciate our efforts in providing the reader with a detailed description of our developments. Please find the answer to each of your questions below them in blue. We hope that we can bring light to each one of the concerns expressed in your questions.

- By replacing catenary collision checks with a  straight segment  between the UGV and the UAV, the planner accepts only LoS configurations, which implicitly ignores sag, incidental contacts with obstacles, and lateral forces induced by the cable. The control stack does not model cable dynamics, and the authors report that lateral tension perturbs attitude and creates snagging risks that required manual monitoring  during experiments, thus the safety margin and the ability to generalize to cluttered environments appear constrained, especially when potentially optimal paths are excluded by the procrustean LoS requirement. 

Thank you for your valuable input. Our proposed system has a mechanism to maintain the tension of the cable so that it is kept straight. That’s why we can consider that the cable does not collide with any obstacle as long as there is LoS between the platforms and we ignore configurations of the cable with sag. Regarding the lateral forces, they are handled as perturbations by the control system. The extra control effort has been estimated in Section 6 second paragraph from last, where it is considered to be neglectable.

We understand your concern about the LoS requirement imposed by the cable, which might be seen as artificial a priori. This requirement is needed because we want to avoid collisions of the cable with obstacles, in order to reduce snagging risks which might happen when the tether makes contact with some surfaces; and because the LTS4 system ensures that the cable is kept straight during the whole experiment, in contrast to our previous approach (citation [28]), which did not assume a straight tether. 

We have clarified this requirement in Section 4.

- Localization evaluation lacks an independent ground truth outside the system. Instead of aligning against external fiducials, the study uses the mean  distance from each LiDAR point to the nearest obstacle in the prior map as an error proxy, while that very map also serves as the DLL reference, which raises an endogeneity concern. Computation times are reported on a Ryzen 7 desktop rather than on onboard avionics, so direct extrapolation to aerial execution remains uncertain, and centimeter scale accuracy claims should be regarded as preliminary.

It is true that the validation results on the localization method cannot guarantee the centimeter scale accuracy claim, so we have relaxed the claim of Section 5.2. However, in our eyes it is a common baseline to compare the different localization systems, and demonstrates that our approach performs as well as other methods in the State of the Art, but runs much faster. 

Regarding the endogeneity concern, please note that as we are comparing the relative performance of the methods against the same map, map that is generated with a different method (LOAM). In addition, we use the localization method to estimate the poses of the AR markers disposed in Scenario 3 (Section 5.3). These detections had a precision of the order of centimeters against their real positions (also computed offline with LOAM) of the emulated defects. This is clarified in Section 5.3.

Finally, we have repeated the localization experiments on the PC onboard  the UAV, to clarify that our method can be executed online internally. You can find the new results also in Section 5.2.1.

- Evidence on endurance and energy efficiency shows lacunae. Although the design targets roughly two hours, the endurance experiment is a manual flight slightly over one hour, and Table 3 on page 15 shows both packs near half charge with a total draw of 1635 Wh, without repetitions or component level power apportionment and without measured efficiencies across battery, inverter, and tether station conversions. The discussion acknowledges higher than expected consumption and attributes it tentatively to auxiliary loads or heat losses, yet no corroborating measurements are provided, fault cases such as tether power loss or winch malfunction are not exercised, and the TB55 backup offers only about ten minutes, so operational safety under failures is opaque.

We have included two new experiments regarding flight endurance. One in which we demonstrate the claimed flight autonomy of two hours, and another extra experiment of one hour. In total, we present three different flight endurance experiments, with detailed logs of the energy consumption of the system. We hope that it will bring evidence of the performance of our system. Please find the new experiments in Section 5.1. 

Regarding the operational safety, in the current state the flight should be monitored by a safety operator, which has real-time information of the level of the UAV backup battery. In case of failure of the tether system, the operator will detect an abnormal decrease of the level and take control to cancel the mission and land the vehicle. We think ten minutes are more than enough to perform this safety procedure. This procedure has been clarified in Section 2.4 in the paper and confirmed in Section 5.1. 

- Autonomy and coordination appear limited, which constrains scalability. The mapping stage requires a short manual flight to gather A LOAM data, so missions at new sites are not fully automated, and coordination relies on stop and go  time synchronization between waypoints with both vehicles capped at 0.25 m per second, which elongates missions and invites drift when the environment changes. In Scenario 2, the UGV could not navigate autonomously because the floor was damaged, hence the cooperative evaluation in challenging terrain occurs only in Scenario 3 under more benign conditions, which leaves performance in larger and more anfractuous spaces unresolved.

Please note that the focus of the paper is the generation of detailed inspection missions, which have to be carefully planned to ensure coverage and to obtain precise enough measurements. Also, they should be flown at low speeds to ensure the quality of the collected data. For this reason, we opted for the two step approach that enables the operator to carefully design the mission with the 3D model of the environment. Once the mission is designed, our software stack generates feasible trajectories and guarantees that they can be autonomously executed without requiring manual intervention, just supervision. The procedure of the mission is clarified in Section 5.3. 

Regarding the problems with the UGV. Again, the main focus is the UGV-UAV cooperation and the design of the autonomous system. This problem can easily be solved by changing the traction type, moving away from the current mecanum wheels currently installed in the UGV to off-ground wheels or tracked wheels would allow it to autonomously navigate over such terrains. This fact is included in the Future Work Section 7.1. 

- Inspection validation remains simplified and does not reflect field reality. Scenario 3 uses twelve ArUco markers as defect surrogates, the main outputs are centimeter scale position estimates and successful detection of all markers over three runs  totaling thirty five minutes, which provides function but limited verisimilitude. The design does not quantify core inspection indicators such as surface coverage, spatial resolution as a function of working distance, and reliability for true defects under occlusion or changing illumination, hence practical utility requires further assessment using real defects and explicit image quality metrics.

As stated in our previous answer, we have included a description of the inspection mission generation in Section 5.3, which details how the inspection mission is planned to ensure surface coverage and to obtain precise enough measurements. In this regard, the detection of the ArUco markers was thought as a method to validate the spatial resolution claims when designing the inspection mission and also the performance of the localization system. However, the inspection mission has been thought as a validation of the marsupial system and occlusion and changes of illumination are out of the scope of our paper.

Reviewer 5 Report

Comments and Suggestions for Authors

The submitted manuscript presents a tethered UAV–UGV system for long-duration inspection in GNSS-denied sites. The problem is timely, and the architecture (ROS, DLL localisation, RRT*+optimizer, coordinator) is coherent and practical. The pros of this manuscript are that it very well explains hardware-software integration, with sensible tether-aware constraints and field trials. The main claim, extended endurance via mobile power, is compelling but lacks sufficient support. Only partial endurance traces are shown (no full ≥120-min run with synchronised power telemetry).

Evaluation of the results lacks baselines or ablations against fixed-tether or untethered alternatives; additionally, planner/coordination metrics are limited, with no statistics available for solve time, success rate, tether clearance, or waiting time. Likewise, the safety treatment is procedural mainly, and fault detection and recovery behaviours aren't experimentally validated. The localisation performance itself is more than promising, yet it is lacking ground truth calibration and stress tests.

Unfortunately, autonomy claims are based on limited AR-marker demonstrations, and the scalability and generality of perception remain unclear. Proprietary components constrain reproducibility, and it is needed to open substitutes and sim scaffolding. I recommend providing a latency/resource breakdown across the sensing, localisation, planning, and control loops. And reconcile the theoretical power budget with measured consumption, including inverter and tether losses. Additionally, discuss the effects of EMI and metal structures on tether, sensors, and communications in confined interiors. Report mission-level KPIs, including coverage, missed waypoints, detection precision/recall, and operator interventions.

Despite the mentioned gaps, the system is thoughtfully engineered and near-publishable, after adding more substantial evidence. For this reason, I recommend a major revision before reconsidering the publication.

My recommendations for the authors are:

1. Please, deliver a complete (≥120minute) endurance validation with synchronised UAV/UAS/UGV power logs, reconciled budget, and thermal discussion.
2. Add quantitative planner/coordination results and at least one ablation or baseline, covering solve times, success rates, tether clearances, and induced waiting.
3. Strengthen reproducibility with public repos (Dockerised launchers, ROSBag datasets, or documented hardware) substitutes or high-fidelity simulation paths.

Author Response

The submitted manuscript presents a tethered UAV–UGV system for long-duration inspection in GNSS-denied sites. The problem is timely, and the architecture (ROS, DLL localisation, RRT*+optimizer, coordinator) is coherent and practical. The pros of this manuscript are that it very well explains hardware-software integration, with sensible tether-aware constraints and field trials. The main claim, extended endurance via mobile power, is compelling but lacks sufficient support. Only partial endurance traces are shown (no full ≥120-min run with synchronised power telemetry).

We are glad that your paper is of your interest. Following the comment, we have included new experiments to fully demonstrate the claimed 2-hour flight time. Please find the new results on Section 5.1.

Evaluation of the results lacks baselines or ablations against fixed-tether or untethered alternatives; additionally, planner/coordination metrics are limited, with no statistics available for solve time, success rate, tether clearance, or waiting time. Likewise, the safety treatment is procedural mainly, and fault detection and recovery behaviours aren't experimentally validated. The localisation performance itself is more than promising, yet it is lacking ground truth calibration and stress tests.

We have included metrics and the example of plan generation in Section 5.3. In the experiment the mission was executed three times, all of them successful. 

We have also included more details about  the safety procedure on fault detection, however our safety approach is not automated and is the responsibility of the monitoring operator. Please find the new details in Section 5.3. 

Finally, even though our localization experiments lack a ground truth, they can serve as a common baseline to compare the different localization systems, and demonstrates that our approach performs as well as other methods in the State of the Art but runs much faster. Please find the new localization results in Section 5.2.1.

Unfortunately, autonomy claims are based on limited AR-marker demonstrations, and the scalability and generality of perception remain unclear. Proprietary components constrain reproducibility, and it is needed to open substitutes and sim scaffolding. I recommend providing a latency/resource breakdown across the sensing, localisation, planning, and control loops. And reconcile the theoretical power budget with measured consumption, including inverter and tether losses. Additionally, discuss the effects of EMI and metal structures on tether, sensors, and communications in confined interiors. Report mission-level KPIs, including coverage, missed waypoints, detection precision/recall, and operator interventions.

In the approach presented in our paper, we rely on reliable and widespread off-the-shelf solutions regarding the quadcopter and the powering system, and we focus on developing the software modules to coordinate the marsupial system, which are released as open-source code. 

With regard to drone control, we only need standard high-level velocity commands from the drone controller. Hence, it can be replaced with an open-sourced alternative such as PX4, ArduPilot and many others. 

During the experiments, we did not perform systematic electromagnetic interference (EMI) studies to assess the effect of the tether on the sensors or communications. However, we did not observe any noticeable increase in noise or interference in either the sensors or the communication systems. All components operated as they did in the absence of the tether; therefore, we assume that the tether does not significantly influence the system.

Regarding latency breakdown, please note that the planning is carried out offline and that the control loop is executed in the DJI internal controller at a high frequency. This is why we have not included such information in the paper. However, we would like to point out that the DLL localization algorithm can be executed online as the mean execution time (0,07 and 0,11 seconds) in the same order as the time between LiDAR measurements (0,1 seconds). 

Finally, we have included additional details on the mission accomplishment in Section 5.3.

Despite the mentioned gaps, the system is thoughtfully engineered and near-publishable, after adding more substantial evidence. For this reason, I recommend a major revision before reconsidering the publication.

My recommendations for the authors are:

1. Please, deliver a complete (≥120minute) endurance validation with synchronised UAV/UAS/UGV power logs, reconciled budget, and thermal discussion.

We have carried out three additional experiments and included the results in the revised version, including a new experiment to fully demonstrate the claimed flight time. Please find the new results on Section 5.1.

1. Add quantitative planner/coordination results and at least one ablation or baseline, covering solve times, success rates, tether clearances, and induced waiting.

We have included more details in the planner and coordination results in Section 5.3.

1. Strengthen reproducibility with public repos (Dockerised launchers, ROSBag datasets, or documented hardware) substitutes or high-fidelity simulation paths.

We would like to note that in the original manuscript we already referenced our published open dataset of the experiments of the paper (https://robotics.upo.es/datasets/marsupial/) as well as our repositories where the open-sourced code of all our systems can be found (https://github.com/robotics-upo/marsupial_launchers). Following your suggestion, we are making efforts to include dockerized launch files to easily reproduce the results published in the paper, which are being developed in a separate branch of the aforementioned repository. Finally, please note that in the Introduction section, we provide the reader with one method to simulate marsupial systems developed by our group, which is presented in a separate paper (reference [25]).

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have made formal revisions, but the material has not added any scientific innovation. I still recommend rejecting the article.

Author Response

The authors have made formal revisions, but the material has not added any scientific innovation. I still recommend rejecting the article.

We regret that the efforts made in the article were not sufficient for you. We would like to emphasize that, in our opinion, our paper presents some relevant scientific contributions. First and foremost, we present a detailed description of a novel tethered UGV-UAV marsupial system as a whole, including a detailed hardware description of the off-the-shelf components and the description of the software stack of our prototype. This software stack includes novel techniques with regards to trajectory planning and system coordination. First, we include the modification of the two-step planning procedure for generating safe trajectories for the whole system. Finally, we describe the trajectory tracking and the marsupial coordinator systems. 

From a technological point of view, this paper is the first to ever demonstrate a UGV-UAV tethered marsupial system continuously operating for over two hours, to the best of our knowledge.

 

We sincerely appreciate the time and effort you dedicated to reviewing our manuscript. Your insightful comments and suggestions have been invaluable in enhancing the overall quality of our work.

Reviewer 3 Report

Comments and Suggestions for Authors

The authors have addressed my comments and also improved the paper overall. It can now be accepted for publication

Reviewer 4 Report

Comments and Suggestions for Authors

Thanks for the responses. the modification helps the manuscript reach a level of an acceptance.

Reviewer 5 Report

Comments and Suggestions for Authors

Thanks for the thoughtful revision of the manuscript. I really appreciate your work. The manuscript is close to acceptance with a few focused improvements:

1. Please, add synchronised power logs for a mission (UAV/UGV/inverter, and V/I versus time).
2. Report solve times (mean +/- std), success beyond three runs.
3. Tether-clearance stats, and a breakdown of waiting (UGV, safety, coordination...). 
4. And, if feasible, a light baseline/ablation study could help to confirm your claims.
5. Spell out fault triggers and recovery steps, count operator interventions.
7. Also "tidy" your manuscript - decimals, first-use acronyms, and units on plots/graphs.

With these additions, I'll be comfortable recommending acceptance - for now, your manuscript needs minor revisions.

Author Response

Thanks for the thoughtful revision of the manuscript. I really appreciate your work. The manuscript is close to acceptance with a few focused improvements:

We are glad that you have found the review adequate. We tried our best to answer your question and to improve the quality of our manuscript. Please find the answers to each question below:

1. Please, add synchronised power logs for a mission (UAV/UGV/inverter, and V/I versus time).

Regarding voltage and intensity, we are afraid that we don’t have V/I information available in the current setup, and with the time to provide the new version of the manuscript (three days) we didn’t have the necessary time to include multimeters in our set up, perform the experiments and analyze the results. However, please note that we added three additional experiments including one where the marsupial system was continuously operated for over two hours. Please find the new plot of battery level (in percentage) against the time, which indicates that the discharge rate is almost steady during all the experiment. We have clarified this circumstance in the manuscript in Section 5.1. We have also included in this version the battery levels in Experiment 7, in which we executed the autonomous inspection mission. Please find the new plot in Section 5.3.2. 

2. Report solve times (mean +/- std), success beyond three runs.

We have included new results of our planner including planning times and success rate in Section 5.3.1.

3. Tether-clearance stats, and a breakdown of waiting (UGV, safety, coordination...). 

We have included more details in the waiting times of the execution of the three repetitions, in order to study the variability in each execution. Please find the new details in Section 5.3.2.  Regarding tether-clearance, we have included this information in the planned trajectories. Please find the information in Section 5.3.1.  

4. And, if feasible, a light baseline/ablation study could help to confirm your claims.

We have included a new section (Section 5.3.1)  in which we detail the performance of the proposed planning system and a comparison with our previously published planner [28]. The results indicate that the new planner is more efficient computationally, while generating trajectories of the comparable quality.

5. Spell out fault triggers and recovery steps, count operator interventions.

The presented experiments were carried out without any issues, and therefore, no operator interventions were required. Please find this information at the beginning of Section 5.3.2. Regarding the fault triggers, there are two main aspects that must be monitored during mission execution: the charge level of the UAV’s backup battery and the potential entanglement of the cable. If one anomaly is detected, the operator stops the mission and takes control of the UAV platform. This procedure is clarified also in Section 5.3.2.

6.  Also "tidy" your manuscript - decimals, first-use acronyms, and units on plots/graphs.

We have made a review of the whole manuscript searching for the aforementioned typos and making improvements for the sake of conciseness and clarity. Please find the changes done in blue. Regarding the units in plots, we didn’t find any omission, however we reviewed the caption of all figures also to clarify their contents. 

With these additions, I'll be comfortable recommending acceptance - for now, your manuscript needs minor revisions.

We sincerely appreciate the time and effort you dedicated to reviewing our manuscript. Your insightful comments and suggestions have been invaluable in enhancing the overall quality of our work. We hope that the revised version of the manuscript meets your expectations.

Round 3

Reviewer 2 Report

Comments and Suggestions for Authors

In the latest version, the authors made cosmetic changes to the text and illustrations. The reviewer comments is still being ignored. The scientific novelty of the presented system description is not obvious.

Reviewer 5 Report

Comments and Suggestions for Authors

Thank you for your careful reworking and clear answers. The manuscript has improved significantly - it now meets the quality criteria and is suitable for publication.