Pith. sign in

REVIEW 4 major objections 6 minor 23 references

Autonomous Close-Proximity Photovoltaic Panel Coating Using a Quadcopter

T0 review · 4 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A quadcopter with only onboard sensing and model-based compensation can autonomously coat a photovoltaic panel, holding altitude to 3.0 cm RMSE outdoors.

desk verdict A solid systems paper that shows real compensation results; the per-rotor ground effect superposition is the one thin spot, but the empirical case is strong enough to send to review. read the letter →

arxiv 2509.10979 v3 pith:W5CFLYZH submitted 2025-09-13 cs.RO cs.SYeess.SY

Dimitri Jacquemont , Carlo Bosio , Teaya Yang , Ruiqi Zhang , Ozgur Orun , Shuai Li , Reza Alam , Thomas M. Schutzius
show 2 more authors
Simo A. Makiharju Mark W. Mueller
This is my paper · ORCID
classification cs.ROcs.SYeess.SY
keywords aerialroboticsphotovoltaicpanelcoatingapplicationgroundeffectcompensationmassvariationvisual-inertialodometryautonomousUAVclose-proximityflight
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that a small quadcopter can autonomously coat a photovoltaic panel by flying close above it, without GPS or external infrastructure. To make close proximity safe, the control loop must cancel two disturbances: the 'ground effect' that changes thrust near a surface, and the steady mass loss as liquid sprays out. The authors build a model-based compensation for both, apply it per rotor, and validate it indoors and outdoors. The result is that altitude tracking error drops from 8.9 cm to 3.0 cm RMSE outdoors, with roughly 70% panel coverage in a windy test. A sympathetic reader would take this as evidence that aerial re-coating of deployed panels is feasible and worth scaling.

What carries the argument

The load-bearing mechanism is the per-rotor ground-effect compensation formula F_comp_i / F_i = 1 - alpha_i * rho * (r / (4 * max(h_i, h_des_i)))^2, where alpha_i is the fraction of rotor i's disc overlapping the panel, r the propeller radius, and h_i the height above the panel. It turns a single-rotor hover correction into a spatially aware compensation that works on tilted panels and when entering from the side. The mass-loss compensation integrates the experimentally measured flow rate and subtracts the cumulative ejected mass from the hover thrust command.

What would settle it

Fly the compensated quadcopter on a trajectory that crosses the panel boundary at a fixed commanded altitude and record z-error; if the per-rotor ground-effect model is incorrect, the error will show a systematic bump or dip as each rotor disc enters or leaves the panel, something a correct model would not produce.

Watch

Extended reading notes

Core claim

The central claim is that a quadcopter equipped only with onboard stereo vision, a learned panel detector, and a pressurized spray system can fly a sweeping trajectory over a PV panel and apply a coating while holding a close, steady standoff. The key enabler is the disturbance compensation: a per-rotor ground-effect model that scales the classic hovering thrust-inflation formula by the fraction of each propeller disc that overlaps the panel, plus a mass estimator that integrates the known flow rate and subtracts the ejected mass from the thrust calculation. In controlled tests, ground-effect compensation reduces straight-line altitude RMSE from 7.2 cm to 1.5 cm, and mass compensation reduce

Load-bearing premise

The ground-effect model is calibrated for a rotor hovering above a flat infinite plane, and the paper assumes it can be applied independently to each rotor and scaled only by the disc-overlap fraction when hovering over tilted, partially covered panels; if that per-rotor superposition is wrong, the compensation will inject its own errors near panel edges.

Editorial extensions

If this is right

  • PV panel re-coating can be automated with a small quadcopter using only onboard sensors, requiring no RTK GPS or ground infrastructure.
  • Enabling ground-effect compensation reduces straight-line altitude RMSE from 7.2 cm to 1.5 cm indoors; enabling mass compensation reduces final hover drift during discharge from 32.7 cm to 2.7 cm.
  • Outdoors, the combined compensations reduce z-axis RMSE during a full panel coverage flight from 8.9 cm to 3.0 cm, even under wind.
  • A single flight with 150 ml of liquid achieves about 70.2% panel coverage in a windy test, indicating the approach is usable but not yet complete.
  • The paper argues the same architecture scales to larger PV installations using multi-nozzle spray arrays and less tightly packed sweeps.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • I infer that the per-rotor ground-effect model, scaled by disc-overlap fraction, could be transferred to other close-proximity aerial tasks such as painting or non-destructive inspection, because those tasks also need steady hover over angled surfaces.
  • The single hover calibration at a 12.4 degree tilt does not test the overlap model near panel edges; a dedicated edge-crossing experiment would reveal whether the superposition assumption breaks down there.
  • Coverage was measured in windy conditions; I infer that adding wind estimation or active sweep-spacing adjustment could push coverage well above 70.2% in a single pass, a natural extension the authors do not develop.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This paper presents an autonomous quadcopter system for applying protective coatings to photovoltaic (PV) panels. The platform uses onboard visual-inertial odometry (VIO) for state estimation, a YOLO-based detector for PV panel corner localization, and a cascaded control architecture. Two model-based disturbance compensations are introduced: a ground-effect compensation that applies a known single-rotor ground-effect model per rotor and scales it by the rotor-surface overlap fraction, and a mass-loss compensation that subtracts ejected liquid mass using an experimentally measured, assumed-constant flow rate. The system is validated indoors with motion capture over flat and tilted surfaces and outdoors over a 1.1 x 2.3 m PV panel, with compensations both enabled and disabled. Reported z-axis RMSEs improve from 7.2 cm to 1.5 cm on a flat surface, from 11.9 cm to 1.8 cm on a 12.4-degree tilted surface, and from 8.9 cm to 3.0 cm outdoors. A separate outdoor test reports approximately 70.2% panel coverage.

Significance. The paper provides a complete, integrated demonstration of a UAV-based PV panel coating system, which is a practically relevant and relatively underexplored application. The experimental comparison of compensation on/off is clear and the improvements are large and consistent in direction. The compensation model is a pragmatic adaptation of established ground-effect models and is not circular: the only fitted parameter rho is calibrated on separate hover data, and the mass flow rate is independently measured. The main scientific risk is the unvalidated extrapolation of the per-rotor ground-effect superposition (Eq. 6) to partial rotor overlap and edge regions, which is precisely the regime where close-proximity coating must operate. If that model behavior is confirmed with additional experiments, the paper would provide a credible and useful system-level validation. The contribution is primarily experimental integration rather than new modeling or control theory.

major comments (4)
  1. [Section IV-A, Eq. (6)] The ground-effect compensation assumes each rotor independently follows the single-rotor, infinite-plane model of Eq. (5), with the effect scaled linearly by the rotor-disk overlap fraction alpha_i. Neither of these assumptions is derived or directly tested. The only tilted-surface validation is a single 1-minute hover at 12.4 degrees (Table III), and no experiment varies alpha_i or crosses a panel edge. Near edges, alpha_i changes rapidly and the compensation may inject vertical or pitch/roll errors exactly where the vehicle is closest to the surface. The h_des clipping in Eq. (6) is a heuristic that further masks model errors below the desired altitude. Since the central 'close-proximity' claim and the outdoor z-RMSE improvement are attributed to this compensation, the domain of validity of Eq. (6) is load-bearing. Please add experiments that systematically vary the rotor overlap fract
  2. [Section V-D, Table IV] The outdoor results, including the headline z-RMSE reduction from 8.9 cm to 3.0 cm, are based on a single run per condition (compensations enabled/disabled). No wind speed, number of runs, or run-to-run variability is reported. A single run cannot establish that the improvement is repeatable, especially outdoors where wind disturbances dominate. Please provide multiple outdoor runs with statistics (e.g., mean and standard deviation of per-run RMSEs), or clearly label the results as a single demonstrative trial rather than a repeatable measurement.
  3. [Section V-B, Tables II and III] The indoor RMSE values are quoted as point estimates without error bars or confidence intervals. Figures 8 and 9 show mean and standard deviation bands, but the RMSE tables do not include the spread across the three runs. With only three runs per condition, the reader cannot assess whether the observed differences (e.g., 1.5 vs 7.2 cm) are statistically meaningful or whether they are within the run-to-run scatter. Please report per-run RMSEs and their standard deviation or range, and state the total number of runs for each table.
  4. [Section V-D, coverage result] The manuscript reports 'approximately 70.2% coverage of the panel surface' but does not describe how coverage was measured or computed. There is no mention of image acquisition, thresholding, manual annotation, or any uncertainty associated with the estimate. Since this is the only quantitative result that directly supports the coating application claim (as opposed to altitude tracking), the methodology for this measurement must be described in sufficient detail to be reproducible.
minor comments (6)
  1. [Eq. (5) vs Eq. (6)] The notation shifts from T_in/T_out in Eq. (5) to F_comp_i/F_i in Eq. (6). Clarify the relationship between T_in and F_i and between T_out and the actual force produced by motor i, so the reader can verify the sign and scaling of the compensation.
  2. [Section IV-A] The computation of alpha_i (the fraction of the propeller rotational area overlapping with the surface) is not described. Please specify the numerical procedure (e.g., polygon clipping, pixel discretization, closed-form intersection) and how the surface boundary is represented.
  3. [Section IV-B] The mass compensation assumes a constant flow rate based on a measurement at 240 kPa. In practice, pressure in the reservoir decays as liquid is ejected, so the flow rate likely decreases over time. State this assumption explicitly and, if possible, quantify the resulting mass-estimation error after the full 150 ml discharge.
  4. [Section V-A] The VIO evaluation reports a 12-minute square trajectory, but no details are given about lighting, texture, or distance from the ground. Since VIO performance is panel-dependent, please provide the environment characteristics or at least mention that the test was conducted in the lab with motion capture as ground truth.
  5. [Figures 8 and 9] The standard deviation bands are described as 'based on three runs.' Clarify whether the bands represent the standard deviation across runs at each time step, or the standard deviation over time within a single run. If across runs, this should be stated in the figure captions.
  6. [Section V-D] The outdoor wind conditions are described qualitatively as 'windy' for the coverage test and as 'mainly wind disturbances' for the compensation comparison. Provide quantitative wind measurements (e.g., anemometer data at a nearby height) to support reproducibility and to allow the reader to judge the severity of the disturbance.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: calibration and validation are separate, and central claims rest on external models and direct experiments.

full rationale

The derivation chain is self-contained and non-circular. The only fitted parameter is the ground-effect coefficient rho=5.71, obtained by least-squares from hover tests (Section IV-A) and then inserted into Eq. (6); the compensation is subsequently validated on separate trajectories (straight-line at 25 cm, 12.4-degree tilted hover, outdoor coverage), so the evaluation data are not the same as the fitting data. The mass-loss compensation is calibrated by an independent volume-over-time nozzle measurement and validated in hover tests. Panel detection uses a YOLO11n model fine-tuned on an external dataset [18], and VIO accuracy is measured against motion capture. The quadcopter dynamics reference [20], [21], including the first author's textbook, but these are standard equations and are not used as a uniqueness argument or to forbid alternatives. The per-rotor alpha_i superposition in Eq. (6) is a modeling assumption whose edge-regime validity is only partially tested, but that is a correctness/robustness concern, not circularity: no prediction in the paper reduces by construction to its input, and no result is forced by a self-citation chain. The stated limitation about 3D point-cloud quality/reflections is likewise an implementation caveat, not a circular step.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities. It relies on two empirical parameters (rho, flow rate) and several domain assumptions about the validity of ground effect superposition and vision-based scene estimation.

free parameters (2)
  • Ground effect coefficient rho = 5.71
    Fitted by least squares from hover tests at various heights above the ground, then used in Eq. (6) for compensation.
  • Nozzle flow rate = Not stated numerically
    Measured by timing ejection of a known liquid volume at 240 kPa; integrated over time to estimate released mass for compensation.
assumptions (4)
  • standard math Quadrotor dynamics follow Newton-Euler equations (1)-(4) as given in [20], [21].
    The control and compensation are built on this standard model.
  • domain assumption The ground effect model T_in/T_out = 1 - rho*(r/(4h))^2 (Eq. 5) is valid for quadcopter rotors near panels, including per-rotor superposition with alpha_i overlap scaling (Eq. 6).
    This is the core of the ground effect compensation; validated only on a single tilted-surface test (Table III) and a flat-surface trajectory.
  • domain assumption Liquid mass flow rate is approximately constant over the dispensing interval despite pressure decay.
    The mass compensation integrates a single measured flow rate; pressure decay in the 240 kPa container would reduce flow, but this is not modeled.
  • domain assumption The PV panel is planar, its corners remain stationary during flight, and the YOLO/RANSAC pipeline correctly estimates the panel plane.
    Trajectory generation depends on these corner estimates; accuracy is acknowledged as limited by point cloud quality and no quantitative detection evaluation is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Autonomous Close-Proximity Photovoltaic Panel Coating Using a Quadcopter." pith.science (2026). https://pith.science/paper/W5CFLYZH

@misc{pith2026250910979,
  author       = {Pith},
  title        = {Pith review of: Autonomous Close-Proximity Photovoltaic Panel Coating Using a Quadcopter},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W5CFLYZH}},
  note         = {Machine review of arXiv:2509.10979}
}
read the original abstract

Photovoltaic (PV) panels are becoming increasingly widespread in the domain of renewable energy, and thus, small efficiency gains can have massive effects. Anti-reflective and self-cleaning coatings enhance panel performance but degrade over time, requiring periodic reapplication. Uncrewed Aerial Vehicles (UAVs) offer a flexible and autonomous way to apply protective coatings more often and at lower cost compared to traditional manual coating methods. We present a quadcopter-based system, equipped with a liquid dispersion mechanism, designed to automate such tasks. The localization stack only uses onboard sensors, relying on visual-inertial odometry and the relative position of the PV panel detected with respect to the quadcopter. The control relies on a model-based controller that accounts for the ground effect and the mass decrease of the quadcopter during liquid dispersion. We validate the autonomy capabilities of our system through extensive indoor and outdoor experiments.

Figures

Figures reproduced from arXiv: 2509.10979 by the authors.

Figure 1
Figure 1. The quadcopter flying over a 1.1×2.3 m photovoltaic panel, showing the reservoir mounted on top and the liquid being dispersed through a jet nozzle mounted at the bottom. lifetimes are often limited to around 8 years or less, a fraction of the typical 25-year lifespan of PV modules, necessitating periodic reapplication to maintain performance. Given the vast scale of modern solar farms, re-coating is logistically ch… view at source ↗
Figure 2
Figure 2. Quadcopter hardware configuration. The platform integrates a Qual [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Autonomy stack block diagram showing the system architecture across five main components: the RB5 companion computer handles high-level autonomy including flight stage, trajectory generation, localization, PV panel detection, and compensation algorithms; the PixRacer flight controller executes low-level control logic; the ESP32 spray controller controls the valve, and sensors and actuators represent the physical int… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Photovoltaic panel detection pipeline: from image processing to world coordinate mapping. YOLO11n model [17] fine-tuned with a PV panel dataset [18]. YOLO11n is selected due to its efficiency in resource￾constrained environments. The model runs inference at 2 Hz on the…
Figure 6
Figure 6. Figure 6: Representation of the propeller rotational area [PITH_FULL_IMAGE:figures/full_fig_p004_6.png]
Figure 7
Figure 7. Figure 7: Plot of the ratio F comp i /Fi, with αi = 1 and h des i = 0.27 m. Let Fi be the setpoint force for motor i, i ∈ {0, 1, 2, 3}, the compensated motor force in presence of ground effect F comp i can then be deduced and applied on each individual motor ( [PITH_FULL_IMAGE:…
Figure 9
Figure 9. Figure 9: Hovering performance during water release at [PITH_FULL_IMAGE:figures/full_fig_p005_9.png]
Figure 10
Figure 10. Figure 10: Coverage flight over a 1.1 × 2.3 m photovoltaic panel with compensations on/off. The x and y directions are aligned with the short and long sides of the PV panel, respectively. The sweeps are done in the x direction. The y coordinate is monotonically increased to avoi…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references

  1. [1]

    Renewables 2024,

    IEA, “Renewables 2024,”IEA, Paris https:// www.iea.org/ reports/ renewables-2024, Licence: CC BY 4.0, 2024

    2024

  2. [2]

    Review of degradation and failure phenomena in photovoltaic modules,

    M. Aghaei, A. Fairbrother, A. Gok, S. Ahmad, S. Kazim, K. Lobato, G. Oreski, A. Reinders, J. Schmitz, M. Theelen,et al., “Review of degradation and failure phenomena in photovoltaic modules,”Renewable and Sustainable Energy Reviews, vol. 159, p. 112160, 2022. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 7

    2022

  3. [3]

    A comprehensive review of automatic cleaning systems of solar panels,

    J. Farrokhi Derakhshandeh, R. AlLuqman, S. Mohammad, H. AlHus- sain, G. AlHendi, D. AlEid, and Z. Ahmad, “A comprehensive review of automatic cleaning systems of solar panels,”Sustainable Energy Technologies and Assessments, vol. 47, p. 101518, 2021

    2021

  4. [4]

    A review on transparent superhydrophobic coatings for self-cleaning solar cell panels: Its fabrication, robustness and industrial implementation,

    S. S. Ingole, R. S. Sutar, P. P. Gaikwad, A. R. Jundle, R. A. Ekunde, S. Liu, and S. S. Latthe, “A review on transparent superhydrophobic coatings for self-cleaning solar cell panels: Its fabrication, robustness and industrial implementation,”Surfaces and Interfaces, vol. 70, p. 106794, 2025

    2025

  5. [5]

    The performance and durability of anti-reflection coatings for solar module cover glass–a review,

    A. M. Law, L. O. Jones, and J. M. Walls, “The performance and durability of anti-reflection coatings for solar module cover glass–a review,”Solar Energy, vol. 261, pp. 85–95, 2023

    2023

  6. [6]

    Available at: https://www.antlab.in/hydrasol-solar-nano-coating.php

    ADV ANCED NanoTech Lab. Available at: https://www.antlab.in/hydrasol-solar-nano-coating.php. Accessed: July 2025

    2025

  7. [7]

    Available at: https://tri-nano.co/about-us/

    TriNANO. Available at: https://tri-nano.co/about-us/. Accessed: July 2025

    2025

  8. [8]

    Photovoltaic module antireflection coating degradation survey using color microscopy and spectral reflectance,

    T. Karin, M. Reed, J. Rand, R. Flottemesch, and A. Jain, “Photovoltaic module antireflection coating degradation survey using color microscopy and spectral reflectance,”Progress in Photovoltaics: Research and Applications, vol. 30, no. 11, pp. 1270–1288, 2022

    2022

Show all 23 references
  1. [9]

    Available at: https://bladeranger.com/.Accessed: July 2025

    BladeRanger. Available at: https://bladeranger.com/.Accessed: July 2025

    2025

  2. [10]

    Available at: https://solarcleano.com/en/

    SolarCleano. Available at: https://solarcleano.com/en/. Accessed: July 2025

    2025

  3. [11]

    Review of agricultural spraying technologies for plant protection using unmanned aerial vehicle (uav),

    H. Chen, Y . Lan, B. K. Fritz, W. C. Hoffmann, and S. Liu, “Review of agricultural spraying technologies for plant protection using unmanned aerial vehicle (uav),”International Journal of Agricultural and Biolog- ical Engineering, vol. 14, no. 1, pp. 38–49, 2021

    2021

  4. [12]

    Paintcopter: An autonomous uav for spray painting on three-dimensional surfaces,

    A. S. Vempati, M. Kamel, N. Stilinovic, Q. Zhang, D. Reusser, I. Sa, J. Nieto, R. Siegwart, and P. Beardsley, “Paintcopter: An autonomous uav for spray painting on three-dimensional surfaces,”IEEE Robotics and Automation Letters, vol. 3, no. 4, pp. 2862–2869, 2018

    2018

  5. [13]

    PX4 Autopilot Firmware

    PX4 Development Team, “PX4 Autopilot Firmware.” https://px4.io, 2025

    2025

  6. [14]

    Unified temporal and spatial calibration for multi-sensor systems,

    P. Furgale, J. Rehder, and R. Siegwart, “Unified temporal and spatial calibration for multi-sensor systems,” in2013 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 1280–1286, 2013

    2013

  7. [15]

    Openvins: A research platform for visual-inertial estimation,

    P. Geneva, K. Eckenhoff, W. Lee, Y . Yang, and G. Huang, “Openvins: A research platform for visual-inertial estimation,” in2020 IEEE Interna- tional Conference on Robotics and Automation (ICRA), pp. 4666–4672, IEEE, 2020

    2020

  8. [16]

    A multi-state constraint kalman filter for vision-aided inertial navigation,

    A. I. Mourikis and S. I. Roumeliotis, “A multi-state constraint kalman filter for vision-aided inertial navigation,” inProceedings 2007 IEEE International Conference on Robotics and Automation, pp. 3565–3572, 2007

    2007

  9. [17]

    Yolov11: An overview of the key architectural enhancements,

    R. Khanam and M. Hussain, “Yolov11: An overview of the key architectural enhancements,” 2024

    2024

  10. [18]

    real solarpanelsoil-shading dataset

    S. panel, “real solarpanelsoil-shading dataset.” https://universe.roboflow. com/soler-panel/real solarpanelsoil-shading, mar 2025. visited on 2025- 07-15

    2025

  11. [19]

    Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,

    M. A. Fischler and R. C. Bolles, “Random sample consensus: a paradigm for model fitting with applications to image analysis and automated cartography,”Commun. ACM, vol. 24, p. 381–395, June 1981

    1981

  12. [20]

    Quadcopter dynamics,

    M. W. Mueller, “Quadcopter dynamics,” inDynamics and Control of Autonomous Flight, pp. 93–103, Springer, 2025

    2025

  13. [21]

    Mueller,Dynamics and Control of Autonomous Flight

    M. Mueller,Dynamics and Control of Autonomous Flight. Springer, 2025

    2025

  14. [22]

    The effect of ground on a helicopter rotor in forward flight,

    I. C. Cheeseman and W. E. Bennett, “The effect of ground on a helicopter rotor in forward flight,” tech. rep., Aeronautical Research Council, 1957

    1957

  15. [23]

    Autonomous landing of quadrotor based on ground effect modelling,

    L. Danjun, Z. Yan, S. Zongying, and L. Geng, “Autonomous landing of quadrotor based on ground effect modelling,” in2015 34th Chinese Control Conference (CCC), pp. 5647–5652, 2015

    2015

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.