Droneulator: A Portable UAV Simulator for Agricultural Workflows with RotorPy and Godot 4

Jacob Swindell; Marija Popovic; Michael Lowen; Riccardo Polvara

arxiv: 2605.23386 · v1 · pith:ZNQG2J6Pnew · submitted 2026-05-22 · 💻 cs.RO

Droneulator: A Portable UAV Simulator for Agricultural Workflows with RotorPy and Godot 4

Jacob Swindell , Michael Lowen , Marija Popovic , Riccardo Polvara This is my paper

Pith reviewed 2026-05-25 04:19 UTC · model grok-4.3

classification 💻 cs.RO

keywords UAV simulatoragricultural workflowsRotorPyGodot 4ROS 2reinforcement learning3D reconstructionlocal planning

0 comments

The pith

Droneulator combines RotorPy dynamics with Godot 4 rendering to support three agricultural UAV workflows in one portable stack.

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

Agricultural UAV work needs simulators that deliver realistic scenes, accurate vehicle physics, and straightforward links to existing robotics software while running on varied machines. The paper introduces Droneulator to meet this need by pairing RotorPy multirotor dynamics with the Godot 4 engine for visuals and sensor output, then routing synchronized data through a Zenoh-based pipeline that matches ROS 2. Tests across image collection for 3D models, local obstacle planning, and reinforcement learning for navigation confirm the system maintains low-latency streams and produces usable results for each task. Researchers benefit because one deployable setup replaces the need for separate simulators when moving between data capture, planning, and learning experiments.

Core claim

Droneulator integrates RotorPy for multirotor dynamics with Godot 4 for rendering and sensor generation. It supplies PX4-based control and a lightweight WebSocket path while publishing synchronized visual and state streams through a Zenoh-based ROS 2-compatible pipeline. Validation on tree-scale image collection for COLMAP reconstruction, EGO-Planner collision-free local planning around canopy obstacles, and closed-loop reinforcement learning in a custom Gymnasium environment shows the simulator sustains low-latency sensing, supports reconstruction under varying capture densities, executes collision-free navigation, and enables stable depth-sensing policy training for obstacle-aware flight.

What carries the argument

RotorPy multirotor dynamics model paired with Godot 4 rendering and sensor generation, connected by a Zenoh-based ROS 2-compatible pipeline that synchronizes visual and state streams for both PX4 and custom command interfaces.

If this is right

The simulator sustains low-latency sensing for reconstruction-oriented data collection under varying capture density.
It executes collision-free local planning around canopy obstacles using EGO-Planner.
It supports stable depth-sensing-based policy training for obstacle-aware navigation in a Gymnasium environment.
A single deployable stack covers inspection data capture, ROS 2/PX4 planning, and reinforcement learning experiments without infrastructure changes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same architecture could support UAV tasks outside agriculture by swapping the Godot scene for different environments.
New sensor types could be added by updating only the Godot rendering layer rather than rebuilding control or dynamics code.
Direct side-by-side runs against physical drone flights would quantify how closely simulated outputs match real sensor noise and dynamics.

Load-bearing premise

The vehicle behavior and sensor data produced by combining RotorPy dynamics with Godot 4 rendering are realistic enough that successful workflow tests demonstrate the simulator's value for agricultural UAV research.

What would settle it

A quantitative comparison of reconstruction accuracy, collision rates, or policy performance between simulator runs and identical tasks flown on physical UAVs in matching agricultural scenes would show whether the reported results transfer to hardware.

Figures

Figures reproduced from arXiv: 2605.23386 by Jacob Swindell, Marija Popovic, Michael Lowen, Riccardo Polvara.

**Figure 1.** Figure 1: System architecture overview of Droneulator. RotorPy provides multirotor dynamics, Godot 4 provides rendering and sensors, Zenoh exposes ROS 2-compatible sensing, and commands arrive through either PX4 SITL or a lightweight WebSocket path, enabling a portable workflow that supports both PX4-based control and lightweight scripted experiments. Abstract—Agricultural UAV research requires simulators that integ… view at source ↗

**Figure 2.** Figure 2: Droneulator user interface and representative agricultural UAV scenes used for inspection-oriented experiments. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Orchard-style tree scene visualised in RViz from the simulator’s depth [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 5.** Figure 5: Executed UAV odometry from five EGO-Planner runs reaching distinct [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗

**Figure 4.** Figure 4: Dense COLMAP reconstructions from Droneulator-captured RGB [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

**Figure 6.** Figure 6: Training curves from a single 50,000-step SAC run show improving [PITH_FULL_IMAGE:figures/full_fig_p005_6.png] view at source ↗

**Figure 7.** Figure 7: Spatial and kinematic performance across 10 successful inference [PITH_FULL_IMAGE:figures/full_fig_p005_7.png] view at source ↗

read the original abstract

Agricultural UAV research requires simulators that integrate realistic 3D scenes, high-fidelity vehicle dynamics, and robotics middleware, while remaining practical to deploy across heterogeneous development machines. We present Droneulator, a portable UAV simulator architecture that combines RotorPy for multirotor dynamics with Godot 4 for rendering and sensor generation. Droneulator exposes both PX4-based control and a lightweight WebSocket command path, and publishes synchronised visual and state streams through a Zenoh-based ROS~2-compatible pipeline. This integration enables a single stack to support inspection-oriented data capture, ROS~2/PX4 local planning, and reinforcement learning experiments without modifying the simulator infrastructure. We present quantified validation of the current system across three agricultural UAV workflows: tree-scale image collection for 3D reconstruction with COLMAP, local planning around canopy obstacles using EGO-Planner, and closed-loop reinforcement learning through a custom Gymnasium environment. In the reported setup, the results show that the simulator can sustain low-latency sensing, support reconstruction-oriented data collection under varying capture density, execute collision-free local planning around canopy obstacles, and support stable depth-sensing-based policy training for obstacle-aware navigation. Together, these results show the potential of Droneulator for agricultural UAV inspection, planning, and learning within one deployable stack.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Droneulator integrates RotorPy and Godot into a portable ag UAV stack with Zenoh ROS2 but the claimed quantified validation lacks any metrics or fidelity checks.

read the letter

The paper presents Droneulator as a simulator that combines RotorPy dynamics with Godot 4 rendering and sensors, wrapped in a Zenoh-based ROS 2 pipeline that also supports PX4 and a WebSocket path. This single setup is meant to handle tree-scale image collection for COLMAP, collision-free local planning with EGO-Planner around canopy obstacles, and depth-based RL in a custom Gymnasium environment without changing the core infrastructure each time. The focus on portability across machines is a practical choice for agricultural UAV work where teams often move between different hardware setups. That kind of engineering attention to deployment can be genuinely useful for people running mixed data capture, planning, and learning experiments. The integration itself appears new in this specific agricultural context even if the pieces are established tools. The soft spot is the validation. The abstract says quantified validation occurred across the three workflows and that the system sustains low-latency sensing, supports reconstruction at varying densities, runs collision-free planning, and produces stable policy training. Yet no numbers, baselines, trajectory errors, sensor noise stats, or real-world comparisons are given. The stress-test concern about missing fidelity metrics for dynamics and sensor outputs stands, because without those details it is impossible to judge whether the simulator is realistic enough for the claimed agricultural uses. This work is for robotics researchers and developers who need a ready simulator for agricultural drone tasks rather than for readers seeking new theoretical results. A reader building or evaluating sim tools might extract value from the architecture and pipeline choices. It deserves a serious referee because the integration is coherent and the target application is clear, even though the evidence for the central realism claim needs to be supplied in the full version.

Referee Report

2 major / 0 minor

Summary. The paper introduces Droneulator, a portable UAV simulator that integrates RotorPy multirotor dynamics with Godot 4 rendering and sensor generation. It supports PX4 control, a WebSocket interface, and a Zenoh-based ROS 2 pipeline, enabling three agricultural workflows: COLMAP-based 3D reconstruction from image capture, EGO-Planner collision-free local planning around canopy obstacles, and Gymnasium-based reinforcement learning for depth-sensing obstacle avoidance. The abstract claims quantified validation demonstrating low-latency sensing, reconstruction support under varying densities, collision-free planning, and stable policy training.

Significance. If the claimed fidelity and integration hold with supporting metrics, the work could provide a practical, single-stack simulator for agricultural UAV research that avoids separate tools for dynamics, rendering, and middleware. The emphasis on portability and ROS 2/PX4 compatibility addresses a real deployment barrier in heterogeneous lab environments.

major comments (2)

[Abstract] Abstract: The central claim of 'quantified validation' across the three workflows is unsupported because no numerical results (e.g., trajectory RMSE, depth-image error statistics, latency distributions, success rates, or sim-to-real gaps) are supplied, nor are baselines, error bars, or exclusion criteria referenced. This directly undermines the assertion that the RotorPy+Godot integration is sufficiently realistic for the reported workflows.
[Abstract] Abstract (and implied Results section): The weakest assumption—that the combined dynamics and sensor streams are faithful enough to validate COLMAP reconstruction, EGO-Planner planning, and Gymnasium RL—is invoked without any fidelity metrics or hardware calibration details, leaving the 'demonstrated potential' claim untestable from the given text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for highlighting the mismatch between the abstract's claims and the supporting evidence. We agree that stronger quantitative support is needed and will revise the manuscript accordingly.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim of 'quantified validation' across the three workflows is unsupported because no numerical results (e.g., trajectory RMSE, depth-image error statistics, latency distributions, success rates, or sim-to-real gaps) are supplied, nor are baselines, error bars, or exclusion criteria referenced. This directly undermines the assertion that the RotorPy+Godot integration is sufficiently realistic for the reported workflows.

Authors: We accept the point. The abstract and results describe workflow outcomes qualitatively (low-latency sensing, collision-free planning, stable training) without the requested statistical metrics or comparisons. In revision we will add explicit numerical results drawn from the experiments, including latency distributions, planning success rates, reconstruction density effects, and any available error statistics, along with baselines and error bars where the data support it. revision: yes
Referee: [Abstract] Abstract (and implied Results section): The weakest assumption—that the combined dynamics and sensor streams are faithful enough to validate COLMAP reconstruction, EGO-Planner planning, and Gymnasium RL—is invoked without any fidelity metrics or hardware calibration details, leaving the 'demonstrated potential' claim untestable from the given text.

Authors: The work emphasizes integration and workflow execution rather than exhaustive fidelity benchmarking. No dedicated hardware calibration campaign or sim-to-real gap quantification was performed. We will revise the abstract and discussion to clarify the scope, report the sensor and dynamics parameters used, and moderate phrasing from 'quantified validation' and 'demonstrated potential' to 'demonstrated functionality for the three workflows within the simulator'. Any additional fidelity indicators computable from existing runs will be included. revision: partial

Circularity Check

0 steps flagged

No circularity: system architecture and empirical validation contain no derivations or self-referential reductions

full rationale

The paper describes an integration of RotorPy dynamics with Godot 4 rendering, a Zenoh/ROS 2 pipeline, and validation runs on three workflows (COLMAP reconstruction, EGO-Planner, Gymnasium RL). No equations, fitted parameters, predictions, or uniqueness theorems appear. Claims rest on reported execution of the described stack rather than any quantity defined in terms of itself. Self-citations are absent from the provided text. This matches the default expectation of a non-circular systems paper whose central results are independent of any internal definitional loop.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the domain assumptions that RotorPy supplies accurate dynamics and Godot 4 supplies usable sensor data for the listed agricultural tasks; no free parameters or invented entities are introduced in the abstract.

axioms (2)

domain assumption RotorPy multirotor dynamics are sufficiently accurate for the tested agricultural workflows
Invoked when the abstract treats RotorPy output as the basis for planning and RL validation.
domain assumption Godot 4 rendering and sensor generation produce data usable for COLMAP reconstruction and depth-based policies
Invoked when the abstract equates simulator outputs with successful 3D reconstruction and stable policy training.

pith-pipeline@v0.9.0 · 5771 in / 1484 out tokens · 29534 ms · 2026-05-25T04:19:29.061733+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages

[1]

RotorPy: A python-based multirotor simulator with aerodynamics for education and research,

S. Folk, J. Paulos, and V . Kumar, “RotorPy: A python-based multirotor simulator with aerodynamics for education and research,”arXiv preprint arXiv:2306.04485, 2023

work page arXiv 2023
[2]

Godot engine

Godot Engine contributors, “Godot engine.” https://godotengine.org/

work page
[3]

Px4 autopilot user guide

PX4 Development Team, “Px4 autopilot user guide.” https://docs.px4. io/v1.15/en/. Accessed: 2026-04-10

work page 2026
[4]

Eclipse zenoh

Eclipse Foundation, “Eclipse zenoh.” https://zenoh.io/. Accessed: 2026- 04-10

work page 2026
[5]

zenoh bridge dds documentation

ROS 2 Documentation Project, “zenoh bridge dds documentation.” https://docs.ros.org/en/jazzy/p/zenoh bridge dds/index.html. Accessed: 2026-04-10

work page 2026
[6]

Structure-from-motion revisited,

J. L. Sch ¨onberger and J.-M. Frahm, “Structure-from-motion revisited,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4104–4113, 2016

work page 2016
[7]

Ego-planner: An ESDF-free gradient-based local planner for quadrotors,

X. Zhou, Z. Wang, N. Pan, F. Gao, and S. Shen, “Ego-planner: An ESDF-free gradient-based local planner for quadrotors,”IEEE Robotics and Automation Letters, 2021

work page 2021
[8]

Survey of simulators for aerial robots: An overview and in-depth systematic comparisons,

C. A. Dimmig, G. Silano, K. McGuire, C. Gabellieri, W. H ¨onig, J. Moore, and M. Kobilarov, “Survey of simulators for aerial robots: An overview and in-depth systematic comparisons,”arXiv preprint arXiv:2311.02296, 2025

work page arXiv 2025
[9]

Px4 simulation integration survey

K. McGuire, “Px4 simulation integration survey.” https://www. mcguirerobotics.com/px4 sim research report/, 2025. Accessed: 2026- 04-10

work page 2025
[10]

Design and use paradigms for gazebo, an open-source multi-robot simulator,

N. Koenig and A. Howard, “Design and use paradigms for gazebo, an open-source multi-robot simulator,” in2004 IEEE/RSJ international conference on intelligent robots and systems (IROS)(IEEE Cat. No. 04CH37566), vol. 3, pp. 2149–2154, IEEE, 2004

work page 2004
[11]

Cyberbotics ltd. webotstm: professional mobile robot simu- lation,

O. Michel, “Cyberbotics ltd. webotstm: professional mobile robot simu- lation,”International Journal of Advanced Robotic Systems, vol. 1, p. 5, 2004

work page 2004
[12]

Isaac Sim

NVIDIA, “Isaac Sim.” https://github.com/isaac-sim/IsaacSim

work page
[13]

Airsim: High-fidelity visual and physical simulation for autonomous vehicles,

S. Shah, D. Dey, C. Lovett, and A. Kapoor, “Airsim: High-fidelity visual and physical simulation for autonomous vehicles,” inField and service robotics: Results of the 11th international conference, pp. 621–635, Springer, 2017

work page 2017
[14]

Flightmare: A flexible quadrotor simulator,

Y . Song, S. Naji, E. Kaufmann, A. Loquercio, and D. Scaramuzza, “Flightmare: A flexible quadrotor simulator,” inProc. of the Conf. on Robot Learning (CoRL), pp. 1147–1157, PMLR, 2021

work page 2021
[15]

Rotors—a modular gazebo mav simulator framework,

F. Furrer, M. Burri, M. Achtelik, and R. Siegwart, “Rotors—a modular gazebo mav simulator framework,” inRobot Operating System (ROS): The Complete Reference (V olume 1), pp. 595–625, Springer, 2016

work page 2016
[16]

Learning to fly—a gym environment with pybullet physics for reinforcement learning of multi-agent quadcopter control,

J. Panerati, H. Zheng, S. Zhou, J. Xu, A. Prorok, and A. P. Schoel- lig, “Learning to fly—a gym environment with pybullet physics for reinforcement learning of multi-agent quadcopter control,” in2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 7512–7519, 2021

work page 2021
[17]

Minimum snap trajectory generation and control for quadrotors,

D. Mellinger and V . Kumar, “Minimum snap trajectory generation and control for quadrotors,” in2011 IEEE international conference on robotics and automation, pp. 2520–2525, IEEE, 2011

work page 2011
[18]

Geometric tracking control of a quadrotor uav on se (3),

T. Lee, M. Leok, and N. H. McClamroch, “Geometric tracking control of a quadrotor uav on se (3),” in49th IEEE conference on decision and control (CDC), pp. 5420–5425, IEEE, 2010

work page 2010

[1] [1]

RotorPy: A python-based multirotor simulator with aerodynamics for education and research,

S. Folk, J. Paulos, and V . Kumar, “RotorPy: A python-based multirotor simulator with aerodynamics for education and research,”arXiv preprint arXiv:2306.04485, 2023

work page arXiv 2023

[2] [2]

Godot engine

Godot Engine contributors, “Godot engine.” https://godotengine.org/

work page

[3] [3]

Px4 autopilot user guide

PX4 Development Team, “Px4 autopilot user guide.” https://docs.px4. io/v1.15/en/. Accessed: 2026-04-10

work page 2026

[4] [4]

Eclipse zenoh

Eclipse Foundation, “Eclipse zenoh.” https://zenoh.io/. Accessed: 2026- 04-10

work page 2026

[5] [5]

zenoh bridge dds documentation

ROS 2 Documentation Project, “zenoh bridge dds documentation.” https://docs.ros.org/en/jazzy/p/zenoh bridge dds/index.html. Accessed: 2026-04-10

work page 2026

[6] [6]

Structure-from-motion revisited,

J. L. Sch ¨onberger and J.-M. Frahm, “Structure-from-motion revisited,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4104–4113, 2016

work page 2016

[7] [7]

Ego-planner: An ESDF-free gradient-based local planner for quadrotors,

X. Zhou, Z. Wang, N. Pan, F. Gao, and S. Shen, “Ego-planner: An ESDF-free gradient-based local planner for quadrotors,”IEEE Robotics and Automation Letters, 2021

work page 2021

[8] [8]

Survey of simulators for aerial robots: An overview and in-depth systematic comparisons,

C. A. Dimmig, G. Silano, K. McGuire, C. Gabellieri, W. H ¨onig, J. Moore, and M. Kobilarov, “Survey of simulators for aerial robots: An overview and in-depth systematic comparisons,”arXiv preprint arXiv:2311.02296, 2025

work page arXiv 2025

[9] [9]

Px4 simulation integration survey

K. McGuire, “Px4 simulation integration survey.” https://www. mcguirerobotics.com/px4 sim research report/, 2025. Accessed: 2026- 04-10

work page 2025

[10] [10]

Design and use paradigms for gazebo, an open-source multi-robot simulator,

N. Koenig and A. Howard, “Design and use paradigms for gazebo, an open-source multi-robot simulator,” in2004 IEEE/RSJ international conference on intelligent robots and systems (IROS)(IEEE Cat. No. 04CH37566), vol. 3, pp. 2149–2154, IEEE, 2004

work page 2004

[11] [11]

Cyberbotics ltd. webotstm: professional mobile robot simu- lation,

O. Michel, “Cyberbotics ltd. webotstm: professional mobile robot simu- lation,”International Journal of Advanced Robotic Systems, vol. 1, p. 5, 2004

work page 2004

[12] [12]

Isaac Sim

NVIDIA, “Isaac Sim.” https://github.com/isaac-sim/IsaacSim

work page

[13] [13]

Airsim: High-fidelity visual and physical simulation for autonomous vehicles,

S. Shah, D. Dey, C. Lovett, and A. Kapoor, “Airsim: High-fidelity visual and physical simulation for autonomous vehicles,” inField and service robotics: Results of the 11th international conference, pp. 621–635, Springer, 2017

work page 2017

[14] [14]

Flightmare: A flexible quadrotor simulator,

Y . Song, S. Naji, E. Kaufmann, A. Loquercio, and D. Scaramuzza, “Flightmare: A flexible quadrotor simulator,” inProc. of the Conf. on Robot Learning (CoRL), pp. 1147–1157, PMLR, 2021

work page 2021

[15] [15]

Rotors—a modular gazebo mav simulator framework,

F. Furrer, M. Burri, M. Achtelik, and R. Siegwart, “Rotors—a modular gazebo mav simulator framework,” inRobot Operating System (ROS): The Complete Reference (V olume 1), pp. 595–625, Springer, 2016

work page 2016

[16] [16]

Learning to fly—a gym environment with pybullet physics for reinforcement learning of multi-agent quadcopter control,

J. Panerati, H. Zheng, S. Zhou, J. Xu, A. Prorok, and A. P. Schoel- lig, “Learning to fly—a gym environment with pybullet physics for reinforcement learning of multi-agent quadcopter control,” in2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 7512–7519, 2021

work page 2021

[17] [17]

Minimum snap trajectory generation and control for quadrotors,

D. Mellinger and V . Kumar, “Minimum snap trajectory generation and control for quadrotors,” in2011 IEEE international conference on robotics and automation, pp. 2520–2525, IEEE, 2011

work page 2011

[18] [18]

Geometric tracking control of a quadrotor uav on se (3),

T. Lee, M. Leok, and N. H. McClamroch, “Geometric tracking control of a quadrotor uav on se (3),” in49th IEEE conference on decision and control (CDC), pp. 5420–5425, IEEE, 2010

work page 2010