pith. sign in

arxiv: 2602.07264 · v2 · submitted 2026-02-06 · 💻 cs.RO · cs.AI· cs.SE

aerial-autonomy-stack -- a Faster-than-real-time, Autopilot-agnostic, ROS2 Framework to Simulate and Deploy Perception-based Drones

Jacopo Panerati , Sina Sajjadi , Sina Soleymanpour , Varunkumar Mehta , Iraj Mantegh This is my paper

Pith reviewed 2026-05-16 06:17 UTC · model grok-4.3

classification 💻 cs.RO cs.AIcs.SE
keywords aerial autonomyROS2drone simulationPX4ArduPilotperceptionfaster-than-real-timeautopilot interface
0
0 comments X

The pith

A ROS2 framework enables over 20 times faster-than-real-time simulation of complete drone perception and control stacks.

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

The paper introduces aerial-autonomy-stack, an open-source framework built on ROS2 that connects GPU-accelerated perception modules to flight controllers. It supplies a shared interface for two widely used autopilots, PX4 and ArduPilot, and runs the full pipeline from sensing through networking and edge compute. The central demonstration is that this end-to-end simulation executes more than twenty times faster than real time. Such speed allows repeated testing of perception-based autonomy without waiting for physical hardware. If the simulation matches reality closely, the time from code changes to flight tests shrinks substantially.

Core claim

aerial-autonomy-stack supplies a ROS2-based common interface for PX4 and ArduPilot autopilots and supports complete end-to-end simulation of perception-to-action autonomy stacks, including edge compute and networking, at speeds exceeding twenty times real time.

What carries the argument

aerial-autonomy-stack, a ROS2 framework providing autopilot-agnostic interfaces and accelerated simulation of the full perception, compute, and control pipeline.

If this is right

  • Perception algorithms can be iterated and validated many times faster than hardware tests allow.
  • Full stacks that include communication links and edge processing can be exercised entirely in simulation.
  • The overall development cycle from code edit to physical flight shortens because simulation replaces many real-world trials.
  • Teams can test the same software binary on simulated and real platforms without major rewrites.

Where Pith is reading between the lines

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

  • The same ROS2 interface pattern could be extended to other vehicle types such as ground robots or underwater systems.
  • Accelerated simulation at this scale opens the possibility of running reinforcement-learning loops inside the framework for policy training.
  • If networking fidelity is high, the framework could support testing of multi-drone coordination without physical fleets.

Load-bearing premise

The simulated perception, networking, and control dynamics match real-world behavior closely enough that faster-than-real-time results transfer to physical drones with little extra tuning.

What would settle it

Run identical perception and control code in the simulator and on a physical drone under matched conditions and compare success rates or trajectory error metrics.

Figures

Figures reproduced from arXiv: 2602.07264 by Iraj Mantegh, Jacopo Panerati, Sina Sajjadi, Sina Soleymanpour, Varunkumar Mehta.

Figure 1
Figure 1. Figure 1: Gazebo Sim scene rendering (right), FPV camera view with YOLO [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Block diagram of the software-in-the-loop simulation architecture comprising of the [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Vehicle models: Holybro X500v2 (top left); 3DR Iris (top right); Standard VTOL (bottom left); and ALTI Transition (bottom right) [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: 3D world models: Plain (top left); Empty (top right); City (bottom left); and Mountain (bottom right). a) Flight Review: A web-based tool for visualizing uLog data, essential for probing into PX4 estimator states and, for real flights, vibration analysis16 . b) MAVExplorer, part of MAVProxy: Provides powerful command-line log analysis for ArduPilot binary logs, sup￾porting custom graphing and data introspe… view at source ↗
Figure 5
Figure 5. Figure 5: Block diagram of the hardware-in-the-loop simulation architecture for two vehicles, with the [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Command-line interface for the four high-level ROS2 Actions ( [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

Unmanned aerial vehicles are rapidly transforming multiple applications, from agricultural and infrastructure monitoring to logistics and defense. Introducing greater autonomy to these systems can simultaneously make them more effective as well as reliable. Thus, the ability to rapidly engineer and deploy autonomous aerial systems has become of strategic importance. In the 2010s, a combination of high-performance compute, data, and open-source software led to the current deep learning and AI boom, unlocking decades of prior theoretical work. Robotics is on the cusp of a similar transformation. However, physical AI faces unique hurdles, often combined under the umbrella term "simulation-to-reality gap". These span from modeling shortcomings to the complexity of vertically integrating the highly heterogeneous hardware and software systems typically found in field robots. To address the latter, we introduce aerial-autonomy-stack, an open-source, end-to-end framework designed to streamline the pipeline from (GPU-accelerated) perception to (flight controller-based) action. Our stack allows the development of aerial autonomy using ROS2 and provides a common interface for two of the most popular autopilots: PX4 and ArduPilot. We show that it supports over 20x faster-than-real-time, end-to-end simulation of a complete development and deployment stack -- including edge compute and networking -- significantly compressing the build-test-release cycle of perception-based autonomy.

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.

Referee Report

2 major / 1 minor

Summary. The manuscript introduces aerial-autonomy-stack, an open-source ROS2 framework for end-to-end simulation and deployment of perception-based drone autonomy. It provides a common autopilot-agnostic interface to PX4 and ArduPilot, integrates GPU-accelerated perception with flight control and networking, and claims to support over 20x faster-than-real-time simulation of the full stack, thereby compressing the build-test-release cycle.

Significance. If the performance claims are substantiated with reproducible benchmarks and the simulation fidelity is validated, the framework could meaningfully accelerate development of perception-driven aerial autonomy by enabling rapid, hardware-in-the-loop iteration that transfers to physical systems. This would address a practical bottleneck in robotics software stacks where heterogeneous components (perception, ROS2 messaging, autopilots, networking) are difficult to co-simulate at scale.

major comments (2)
  1. [Abstract] Abstract: the central claim of 'over 20x faster-than-real-time, end-to-end simulation of a complete development and deployment stack—including edge compute and networking' is presented without any benchmark protocol, workload description, timing methodology (wall-clock vs. simulated time), perception model details, input resolution, or comparison baselines. This renders the quantitative speedup unevaluable and load-bearing for the paper's contribution.
  2. [Abstract] The sim-to-real transfer assumption (that faster-than-real-time results will carry over to physical deployment without major retuning) is stated but unsupported by any reported validation experiments comparing simulated versus real perception accuracy, latency, or control stability.
minor comments (1)
  1. The manuscript would benefit from a dedicated section or table explicitly listing the hardware platforms, ROS2 versions, and autopilot firmware versions used in any timing experiments.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation of our performance claims and assumptions. We address each point below and indicate the revisions we will make to the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of 'over 20x faster-than-real-time, end-to-end simulation of a complete development and deployment stack—including edge compute and networking' is presented without any benchmark protocol, workload description, timing methodology (wall-clock vs. simulated time), perception model details, input resolution, or comparison baselines. This renders the quantitative speedup unevaluable and load-bearing for the paper's contribution.

    Authors: We agree that the abstract would benefit from additional methodological context to make the 20x claim immediately evaluable. The full manuscript (Section 4) already details the benchmark protocol: a YOLOv5 perception model at 640x480 input resolution running on GPU-accelerated ROS2 nodes, PX4 and ArduPilot SITL instances, wall-clock timing against simulated time using the ROS2 clock, and direct comparison to real-time execution baselines on the same hardware. To address the referee's concern without expanding the abstract excessively, we will revise the abstract to concisely reference the key workload parameters (perception model and resolution), timing methodology, and baseline comparison. This revision will be made in the next version. revision: yes

  2. Referee: [Abstract] The sim-to-real transfer assumption (that faster-than-real-time results will carry over to physical deployment without major retuning) is stated but unsupported by any reported validation experiments comparing simulated versus real perception accuracy, latency, or control stability.

    Authors: The manuscript's primary contribution is the simulation framework itself and its ability to accelerate the development cycle through faster-than-real-time execution while preserving the same ROS2 interfaces used on hardware. We do not claim or demonstrate direct sim-to-real equivalence experiments (e.g., side-by-side perception accuracy or closed-loop stability metrics) in the current work, as those would require separate physical flight tests outside the paper's scope. We will add an explicit limitations paragraph in the discussion section clarifying that the framework uses identical message types and autopilot interfaces to support transfer, but that quantitative sim-to-real validation remains future work. This addresses the concern by making the assumption transparent rather than unsupported. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical software framework claim with no derivation chain

full rationale

The paper presents an open-source ROS2 framework for perception-based drone simulation and deployment. Its key claim of supporting over 20x faster-than-real-time end-to-end simulation is stated as an empirical demonstration from implementation benchmarks, not as a mathematical prediction or first-principles derivation. No equations, fitted parameters, self-definitional constructs, or load-bearing self-citations appear in the abstract or description. The contribution is self-contained in its software architecture and reported timing results, which can be independently verified via code execution without reducing to inputs by construction. This is the expected outcome for a systems/framework paper.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The contribution is a software integration framework rather than a theoretical derivation, so the ledger contains only standard domain assumptions about middleware and simulation fidelity with no free parameters or new physical entities.

axioms (2)
  • domain assumption ROS2 middleware can reliably integrate GPU perception pipelines with flight controller commands without prohibitive latency
    Invoked implicitly in the design of the common interface for perception-to-action flow
  • domain assumption Simulation models of edge compute and networking are sufficiently representative for development purposes
    Required for the faster-than-real-time claim to be useful beyond pure timing benchmarks

pith-pipeline@v0.9.0 · 5570 in / 1388 out tokens · 33356 ms · 2026-05-16T06:17:40.215539+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

30 extracted references · 30 canonical work pages

  1. [1]

    Champion-level drone racing using deep reinforce- ment learning,

    E. Kaufmann, L. Bauersfeld, A. Loquercio, M. M ¨uller, V . Koltun, and D. Scaramuzza, “Champion-level drone racing using deep reinforce- ment learning,”Nature, vol. 620, no. 7976, pp. 982–987, 2023

    work page 2023

  2. [2]

    On your own: Pro-level autonomous drone racing in uninstrumented arenas,

    M. Bosello, F. Pinzarrone, S. Kiade, D. Aguiari, Y . Keuter, A. AlShe- hhi, G. Caminati, K. L. Wong, K. S. Chou, J. Halepota, F. Alneyadi, J. Panerati, and G. Pau, “On your own: Pro-level autonomous drone racing in uninstrumented arenas,”IEEE Robotics and Automation Letters, vol. 11, no. 3, pp. 2674–2681, 2026

    work page 2026

  3. [3]

    A compre- hensive survey on artificial intelligence for unmanned aerial vehicles,

    S. Sai, A. Garg, K. Jhawar, V . Chamola, and B. Sikdar, “A compre- hensive survey on artificial intelligence for unmanned aerial vehicles,” IEEE Open Journal of V ehicular Technology, vol. 4, pp. 713–738, 2023

    work page 2023

  4. [4]

    Science, technology and the future of small autonomous drones,

    D. Floreano and R. J. Wood, “Science, technology and the future of small autonomous drones,”Nature, vol. 521, no. 7553, pp. 460–466, 2015

    work page 2015

  5. [5]

    Cloud container technologies in military applications,

    C. Zhong, Y . Zheng, W. Lijun, and W. Shen, “Cloud container technologies in military applications,” in2025 21st International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery (ICNC-FSKD), 2025, pp. 727–736

    work page 2025

  6. [6]

    Up, Up, and Away: Adventures in Aerial Robotics,

    K. McGuire and R. Roche, “Up, Up, and Away: Adventures in Aerial Robotics,” Keynote speech at the Open Source Summit Europe, November 2025, available at: https://youtu.be/HTsXCDTch2I

    work page 2025

  7. [7]

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

    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 [survey],”IEEE Robotics & Automation Magazine, vol. 32, no. 2, pp. 153–166, 2025

    work page 2025

  8. [8]

    Simulation-based testing of unmanned aerial vehicles with aerialist,

    S. Khatiri, S. Panichella, and P. Tonella, “Simulation-based testing of unmanned aerial vehicles with aerialist,” in2024 IEEE/ACM 46th International Conference on Software Engineering: Companion Proceedings (ICSE-Companion), 2024, pp. 134–138

    work page 2024

  9. [9]

    Aerostack2: A software framework for developing multi-robot aerial systems,

    M. Fernandez-Cortizas, M. Molina, P. Arias-Perez, R. Perez-Segui, D. Perez-Saura, and P. Campoy, “Aerostack2: A software framework for developing multi-robot aerial systems,” 2024. [Online]. Available: https://arxiv.org/abs/2303.18237

    work page arXiv 2024

  10. [10]

    Ag- ilicious: Open-source and open-hardware agile quadrotor for vision- based flight,

    P. Foehn, E. Kaufmann, A. Romero, R. Penicka, S. Sun, L. Bauersfeld, T. Laengle, G. Cioffi, Y . Song, A. Loquercio, and D. Scaramuzza, “Ag- ilicious: Open-source and open-hardware agile quadrotor for vision- based flight,”Science Robotics, vol. 7, no. 67, p. eabl6259, 2022

    work page 2022

  11. [11]

    Crazyswarm: A large nano-quadcopter swarm,

    J. A. Preiss, W. Honig, G. S. Sukhatme, and N. Ayanian, “Crazyswarm: A large nano-quadcopter swarm,” in2017 IEEE International Confer- ence on Robotics and Automation (ICRA), 2017, pp. 3299–3304

    work page 2017

  12. [12]

    Unmanned aerial vehicle abstraction layer: An abstraction layer to operate unmanned aerial vehicles,

    F. Real, A. Torres-Gonz ´alez, P. R. Soria, J. Capit ´an, and A. Ollero, “Unmanned aerial vehicle abstraction layer: An abstraction layer to operate unmanned aerial vehicles,”International Journal of Advanced Robotic Systems, vol. 17, no. 4, pp. 1–13, 2020

    work page 2020

  13. [13]

    Fast, autonomous flight in gps-denied and cluttered environments,

    K. Mohta, M. Watterson, Y . Mulgaonkar, S. Liu, C. Qu, A. Makineni, K. Saulnier, K. Sun, A. Zhu, J. Delmerico, D. Thakur, K. Karydis, N. Atanasov, G. Loianno, D. Scaramuzza, K. Daniilidis, C. J. Taylor, and V . Kumar, “Fast, autonomous flight in gps-denied and cluttered environments,”Journal of Field Robotics, vol. 35, no. 1, pp. 101–120, 2018

    work page 2018

  14. [14]

    The mrs uav system: Pushing the frontiers of reproducible research, real-world deployment, and education with autonomous unmanned aerial vehicles,

    T. Baca, M. Petrlik, M. Vrba, V . Spurny, R. Penicka, D. Hert, and M. Saska, “The mrs uav system: Pushing the frontiers of reproducible research, real-world deployment, and education with autonomous unmanned aerial vehicles,”Journal of Intelligent & Robotic Systems, vol. 102, no. 1, p. 26, 2021

    work page 2021

  15. [15]

    Xtdrone: A customizable multi-rotor uavs simulation platform,

    K. Xiao, S. Tan, G. Wang, X. An, X. Wang, and X. Wang, “Xtdrone: A customizable multi-rotor uavs simulation platform,” in2020 4th Inter- national Conference on Robotics and Automation Sciences (ICRAS), 2020, pp. 55–61

    work page 2020

  16. [16]

    What is the impact of releasing code with publications?: Statistics from the machine learning, robotics, and control communities,

    S. Zhou, L. Brunke, A. Tao, A. W. Hall, F. P. Bejarano, J. Panerati, and A. P. Schoellig, “What is the impact of releasing code with publications?: Statistics from the machine learning, robotics, and control communities,”IEEE Control Systems, vol. 44, no. 4, pp. 38– 46, 2024

    work page 2024

  17. [17]

    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. Schoellig, “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), 2021, pp. 7512–7519

    work page 2021

  18. [18]

    Aerial gym simulator: A framework for highly parallelized simulation of aerial robots,

    M. Kulkarni, W. Rehberg, and K. Alexis, “Aerial gym simulator: A framework for highly parallelized simulation of aerial robots,”IEEE Robotics and Automation Letters, pp. 1–8, 2025

    work page 2025

  19. [19]

    Flightmare: A flexible quadrotor simulator,

    Y . Song, S. Naji, E. Kaufmann, A. Loquercio, and D. Scaramuzza, “Flightmare: A flexible quadrotor simulator,” inConference on Robot Learning, 2020

    work page 2020

  20. [20]

    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, M. Hutter and R. Siegwart, Eds. Springer International Publishing, 2018, pp. 621–635

    work page 2018

  21. [21]

    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,”

    work page

  22. [22]

    Available: https://arxiv.org/abs/2306.04485

    [Online]. Available: https://arxiv.org/abs/2306.04485

    work page arXiv

  23. [23]

    Aerostack: An architecture and open-source software framework for aerial robotics,

    J. L. Sanchez-Lopez, R. A. Su ´arez Fern´andez, H. Bavle, C. Sampedro, M. Molina, J. Pestana, and P. Campoy, “Aerostack: An architecture and open-source software framework for aerial robotics,” in2016 International Conference on Unmanned Aircraft Systems (ICUAS), 2016, pp. 332–341

    work page 2016

  24. [24]

    Ros-based multi-domain swarm framework for fast prototyping,

    J. Martin and S. Esteban, “Ros-based multi-domain swarm framework for fast prototyping,”Aerospace, vol. 12, no. 8, p. 702, 2025

    work page 2025

  25. [25]

    The reality gap in robotics: Challenges, solutions, and best practices,

    E. Aljalbout, J. Xing, A. Romero, I. Akinola, C. R. Garrett, E. Heiden, A. Gupta, T. Hermans, Y . Narang, D. Fox, D. Scaramuzza, and F. Ramos, “The reality gap in robotics: Challenges, solutions, and best practices,”Annual Review of Control, Robotics, and Autonomous Systems, 2025

    work page 2025

  26. [26]

    Advancing reproducibility, benchmarks, and education with remote sim2real: Remote simulation to real robot hardware,

    S. Teetaert, W. Zhao, A. Loquercio, S. Zhou, L. Brunke, M. Schuck, W. H¨onig, J. Panerati, and A. P. Schoellig, “Advancing reproducibility, benchmarks, and education with remote sim2real: Remote simulation to real robot hardware,”IEEE Robotics & Automation Magazine, vol. 32, no. 1, 2025

    work page 2025

  27. [27]

    Neural lander: Stable drone landing control using learned dynamics,

    G. Shi, X. Shi, M. O’Connell, R. Yu, K. Azizzadenesheli, A. Anand- kumar, Y . Yue, and S.-J. Chung, “Neural lander: Stable drone landing control using learned dynamics,” in2019 International Conference on Robotics and Automation (ICRA), 2019, pp. 9784–9790

    work page 2019

  28. [28]

    A plea for lean software,

    N. Wirth, “A plea for lean software,”Computer, vol. 28, no. 2, 1995

    work page 1995

  29. [29]

    A review on yolov8 and its advancements,

    M. Sohan, T. Sai Ram, and C. V . Rami Reddy, “A review on yolov8 and its advancements,” inData Intelligence and Cognitive Informatics, I. J. Jacob, S. Piramuthu, and P. Falkowski-Gilski, Eds. Singapore: Springer Nature Singapore, 2024, pp. 529–545

    work page 2024

  30. [30]

    Kiss-icp: In defense of point-to-point icp – simple, accurate, and robust registration if done the right way,

    I. Vizzo, T. Guadagnino, B. Mersch, L. Wiesmann, J. Behley, and C. Stachniss, “Kiss-icp: In defense of point-to-point icp – simple, accurate, and robust registration if done the right way,”IEEE Robotics and Automation Letters, vol. 8, no. 2, pp. 1029–1036, 2023

    work page 2023