pith. sign in

arxiv: 2511.17299 · v2 · pith:IQ5PV2RVnew · submitted 2025-11-21 · 💻 cs.RO

MonoSpheres: Large-Scale Monocular SLAM-Based UAV Exploration through Perception-Coupled Mapping and Planning

Tom\'a\v{s} Musil , Mat\v{e}j Petrl\'ik , Martin Saska This is my paper

Pith reviewed 2026-05-17 20:42 UTC · model grok-4.3

classification 💻 cs.RO
keywords monocular SLAMUAV explorationfrontier-based explorationperception-aware planningsparse depth mappingoutdoor navigationuncertainty handlingreal-world robotics
0
0 comments X

The pith

Monocular SLAM with oversampled free-space mapping and uncertainty-aware planning lets UAVs explore large unstructured 3D outdoor environments.

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

The paper introduces a complete exploration pipeline for UAVs that relies solely on a single monocular camera. It augments a sparse SLAM front-end so that mapping explicitly oversamples free space in low-texture regions and maintains obstacle-position uncertainty estimates. Planning then exploits this information through fast replanning and heading commands that keep new views parallax-rich. Frontier detection is adapted to respect both the parallax needs of the SLAM system and the risk of textureless surfaces. The result is demonstrated in real-world indoor and outdoor settings at scales previously considered out of reach for monocular systems.

Core claim

The central claim is that perception-coupled mapping and planning, built directly around the limitations of sparse monocular SLAM, suffices for safe, large-scale 3D exploration. Mapping oversamples free space where texture is sparse and tracks depth uncertainty; planning reacts with rapid replanning and perception-aware heading control. Frontier-based exploration is shown to remain viable once parallax requirements and textureless-surface risks are explicitly modeled. The authors state this yields the first real-world 3D monocular exploration of unstructured outdoor environments.

What carries the argument

Perception-coupled mapping and planning that augments sparse monocular SLAM with free-space oversampling, obstacle uncertainty tracking, rapid replanning, and perception-aware heading control.

If this is right

  • Frontier exploration works with only sparse monocular depth once parallax and texture constraints are modeled.
  • Rapid replanning plus perception-aware heading control compensates for the remaining depth uncertainty.
  • The same pipeline covers both large indoor and unstructured outdoor environments at scale.
  • Open-sourcing the implementation enables direct reuse and extension by other monocular UAV teams.

Where Pith is reading between the lines

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

  • The approach could be tested on ground robots in similar unstructured settings to check whether the same coupling of mapping and planning generalizes beyond flying platforms.
  • If the uncertainty model proves conservative, it might allow tighter integration with learned depth predictors that supply denser but still uncertain maps.
  • A direct comparison of exploration coverage per meter flown against stereo or LiDAR baselines in the same outdoor sites would quantify the sensor-cost savings.

Load-bearing premise

The augmented sparse monocular SLAM front-end produces free-space information that is reliable enough for safe planning even in textureless or low-parallax regions.

What would settle it

A single documented collision or stuck state during a real outdoor flight through a texture-poor or low-parallax region would falsify the safety claim.

Figures

Figures reproduced from arXiv: 2511.17299 by Martin Saska, Mat\v{e}j Petrl\'ik, Tom\'a\v{s} Musil.

Figure 1
Figure 1. Figure 1: Illustration of the proposed approach. The mapping pipeline [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Mapping pipeline overview. The OVDE and DBOF modules are illustrated in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Open-area virtual-depth estimation diagram. The thick green points xvir,k satisfy the condition defined in subsection IV-B and are added to the construction of Fd B. Open-Area Virtual-Depth Estimation (OVDE) When moving in open areas, such as a grassy field without tall obstacles in [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Top-down illustration of sphere sampling and obstacle point t [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: In principle, we employ the commonly used greedy next-best-view (NBV) [11] strategy, which has so far been deemed unsuitable for monocular exploration in previous works [4], [5]. In the following sections, we introduce several major differences in methodology that are critical for allowing the NBV strategy to be used for robots equipped with only a monocular camera for depth sensing. A. Frontier Sampling o… view at source ↗
Figure 7
Figure 7. Figure 7: This stops the UAV from getting stuck repeatedly trying to uncover such surfaces. VI. EXPERIMENTS In this section, we analyze the performance of the pro￾posed method in large-scale real-world (Sec. VI-A) and simulated (Sec. VI-B) environments. The presented imple￾mentation of the proposed mapping and exploration methods is currently written in python in two threads. With this implementation, the mapping ru… view at source ↗
Figure 6
Figure 6. Figure 6: The 4min orchard exploration experiment. Right: explored map (approx. 40x30m wide), the UAV’s trajectory (small black arrows) and trees (circled blue). A. Real-World UAV Monocular-Inertial Exploration In [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of the real-world large-scale exploration experiment [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Trajectories of the best MonoSpheres (black) and Grid-Based explorer (blue) simulation experiments per each of the 4 worlds (top) along with the resulting MonoSpheres mapped obstacle points (bottom) colored by height. X marks the starting position [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Ablation test results on the earthquake world. Left: visualization of exploration progress, best runs per method are thick and a marker signifies a crash or mission completion. Right: best-run trajectories for each method. the occupancy grid with the same minimum distance from obstacles and uknown space as the MonoSpheres method (1.5 m in these experiments). The robot then follows these paths with the came… view at source ↗
read the original abstract

Autonomous exploration of unknown environments is a key capability for mobile robots, but it is largely unsolved for robots equipped with only a single monocular camera and no dense range sensors. In this paper, we present a novel approach to monocular vision-based exploration that can safely cover large-scale unstructured indoor and outdoor 3D environments by explicitly accounting for the properties of a sparse monocular SLAM frontend in both mapping and planning. The mapping module solves the problems of sparse depth data, free-space gaps, and large depth uncertainty by oversampling free space in texture-sparse areas and keeping track of obstacle position uncertainty. The planning module handles the added free-space uncertainty through rapid replanning and perception-aware heading control. We further show that frontier-based exploration is possible with sparse monocular depth data when parallax requirements and the possibility of textureless surfaces are taken into account. We evaluate our approach extensively in diverse real-world and simulated environments, including ablation studies. To the best of the authors' knowledge, the proposed method is the first to achieve 3D monocular exploration in real-world unstructured outdoor environments. We open-source our implementation to support future research.

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 / 2 minor

Summary. The manuscript presents MonoSpheres, a monocular SLAM-based system for large-scale 3D UAV exploration in unstructured indoor and outdoor environments. The mapping module augments sparse monocular depth with free-space oversampling in texture-sparse regions and explicit obstacle uncertainty tracking; the planning module uses rapid replanning and perception-aware heading to tolerate residual uncertainty. Frontier-based exploration is adapted to account for parallax and textureless surfaces. Extensive real-world and simulated evaluations with ablations are reported, and the implementation is open-sourced. The central claim is that this is the first method to achieve safe 3D monocular exploration in real-world unstructured outdoor settings.

Significance. If the results hold, the work would be significant for enabling lightweight, sensor-minimal exploration in challenging outdoor environments where dense range sensors are impractical. The explicit coupling of monocular SLAM properties into both mapping and planning, together with the open-sourced code, provides a reproducible baseline that could accelerate progress in vision-only robotics. The claim of first real-world outdoor 3D monocular exploration, if substantiated, would mark a notable advance over prior indoor or simulation-only monocular systems.

major comments (2)
  1. [Mapping module] Mapping module description (around the oversampling and uncertainty tracking paragraphs): the approach relies on oversampling free space and tracking obstacle uncertainty to convert sparse SLAM output into safe planning constraints, yet no quantitative error bounds, covariance analysis, or worst-case depth-error characterization is provided for correlated errors in low-texture or low-parallax outdoor regions. This is load-bearing for the central safety claim, as optimistic free-space labels could still produce colliding trajectories before replanning reacts.
  2. [Evaluation] Evaluation section (real-world outdoor experiments): while ablations and diverse environments are mentioned, the reported metrics do not include direct measurements of collision risk or free-space label accuracy specifically in textureless outdoor patches; without these, it is difficult to verify that the perception-coupled modules sufficiently mitigate the weakest assumption identified in the skeptic note.
minor comments (2)
  1. [Related Work] The related-work section should explicitly compare against the most recent monocular exploration methods that also handle uncertainty (e.g., those using probabilistic occupancy or learned depth priors) to strengthen the novelty claim.
  2. [Mapping module] Notation for uncertainty propagation (e.g., how obstacle position variance is updated across frames) could be clarified with a short equation or pseudocode snippet for reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed review of our manuscript on MonoSpheres. The comments raise valid points about substantiating the safety aspects of our monocular mapping and planning approach. We respond to each major comment below and indicate planned revisions to address the concerns.

read point-by-point responses

  1. Referee: [Mapping module] Mapping module description (around the oversampling and uncertainty tracking paragraphs): the approach relies on oversampling free space and tracking obstacle uncertainty to convert sparse SLAM output into safe planning constraints, yet no quantitative error bounds, covariance analysis, or worst-case depth-error characterization is provided for correlated errors in low-texture or low-parallax outdoor regions. This is load-bearing for the central safety claim, as optimistic free-space labels could still produce colliding trajectories before replanning reacts.

    Authors: We acknowledge that the manuscript does not include a formal derivation of quantitative error bounds or a full covariance analysis for correlated depth errors in low-texture or low-parallax regions. Our design instead uses conservative free-space oversampling and uncertainty tracking together with rapid replanning to maintain safety margins, which is supported by the collision-free results across all real-world outdoor experiments. To better address this point, we will revise the mapping module section to add an empirical analysis of observed depth errors from the monocular SLAM in texture-sparse outdoor areas, including how the oversampling parameters were selected relative to these uncertainties. This will provide additional support for the safety claims without altering the practical contributions of the work. revision: yes

  2. Referee: [Evaluation] Evaluation section (real-world outdoor experiments): while ablations and diverse environments are mentioned, the reported metrics do not include direct measurements of collision risk or free-space label accuracy specifically in textureless outdoor patches; without these, it is difficult to verify that the perception-coupled modules sufficiently mitigate the weakest assumption identified in the skeptic note.

    Authors: We agree that direct metrics on free-space label accuracy and collision risk within textureless outdoor patches would strengthen the evaluation and more explicitly verify the mitigation of mapping uncertainties. The current evaluation reports overall exploration success, coverage, and ablation results but does not break out these specific per-patch metrics. In the revised manuscript, we will add targeted analysis from the collected outdoor datasets, including the accuracy of free-space labels in identified texture-sparse regions and confirmation of zero collisions due to mapping errors. These additions will directly respond to the concern while preserving the focus on system-level performance. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation builds on independent modules and external SLAM foundations

full rationale

The paper augments an established sparse monocular SLAM frontend with separate mapping (oversampling and uncertainty tracking) and planning (rapid replanning and perception-aware control) modules whose correctness is evaluated empirically in real-world tests rather than derived tautologically from fitted parameters or self-referential definitions. No equations or claims reduce the central exploration result to its own inputs by construction, and the 'first to achieve' statement is an empirical claim supported by evaluation rather than a uniqueness theorem imported from prior self-work. The approach remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The central claim depends on standard domain assumptions about monocular SLAM behavior plus a small number of tunable parameters for uncertainty handling; no new invented entities are introduced.

free parameters (2)
  • free-space oversampling density in texture-sparse regions
    Chosen to compensate for sparse depth data; value not specified in abstract but required for the mapping module to function.
  • replan frequency threshold
    Set to enable rapid response to new depth information; appears as a design choice for handling uncertainty.
axioms (2)
  • domain assumption Monocular SLAM provides usable depth estimates when sufficient parallax and texture are present.
    Invoked when adapting frontier exploration to sparse data.
  • domain assumption Obstacle position uncertainty can be tracked and propagated into safe planning decisions.
    Core premise of the mapping module described in the abstract.

pith-pipeline@v0.9.0 · 5518 in / 1428 out tokens · 72374 ms · 2026-05-17T20:42:38.413160+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

26 extracted references · 26 canonical work pages

  1. [1]

    CERBERUS in the DARPA Subterranean Challenge,

    M. Tranzatto, T. Miki, M. Dharmadhikariet al., “CERBERUS in the DARPA Subterranean Challenge,”Science Robotics, vol. 7, 2022

    work page 2022

  2. [2]

    Heterogeneous Ground and Air Platforms, Homogeneous Sensing: Team CSIRO Data61’s Approach to the DARPA Subterranean Challenge,

    N. Hudson, F. Talbot, M. Coxet al., “Heterogeneous Ground and Air Platforms, Homogeneous Sensing: Team CSIRO Data61’s Approach to the DARPA Subterranean Challenge,”Field Robotics, vol. 2, pp. 595–636, 2021

    work page 2021

  3. [3]

    UA Vs Beneath the Sur- face: Cooperative Autonomy for Subterranean Search and Rescue in DARPA SubT,

    M. Petrl ´ık, P. Petr ´aˇcek, V . Kr ´atk´yet al., “UA Vs Beneath the Sur- face: Cooperative Autonomy for Subterranean Search and Rescue in DARPA SubT,”Field Robotics, vol. 3, pp. 1–68, 2022

    work page 2022

  4. [4]

    From monocular SLAM to autonomous drone exploration,

    L. von Stumberg, V . C. Usenko, J. J. Engel, J. St ¨uckler, and D. Cre- mers, “From monocular SLAM to autonomous drone exploration,” 2017 European Conference on Mobile Robots (ECMR), pp. 1–8, 2016

    work page 2017

  5. [5]

    Monocular 3D Exploration using Lines-of-Sight and Local Maps,

    D. Pittol, M. Mantelli, R. Maffei, M. L. Kolberg, and E. Prestes, “Monocular 3D Exploration using Lines-of-Sight and Local Maps,” Journal of Intelligent & Robotic Systems, vol. 100, pp. 465 – 481, 2020

    work page 2020

  6. [6]

    MonoNav: MA V Navigation via Monoc- ular Depth Estimation and Reconstruction,

    N. Simon and A. Majumdar, “MonoNav: MA V Navigation via Monoc- ular Depth Estimation and Reconstruction,” inSymposium on Experi- mental Robotics (ISER), 2023

    work page 2023

  7. [7]

    CNN-Based Dense Monocular Visual SLAM for Real-Time UA V Exploration in Emergency Conditions,

    A. Steenbeek and F. Nex, “CNN-Based Dense Monocular Visual SLAM for Real-Time UA V Exploration in Emergency Conditions,” Drones, 2022

    work page 2022

  8. [8]

    Deep learning for monocular depth estimation: A review,

    Y . Ming, X. Meng, C. Fan, and H. Yu, “Deep learning for monocular depth estimation: A review,”Neurocomputing, vol. 438, pp. 14–33, 2021

    work page 2021

  9. [9]

    Deep Learning-based Depth Estimation Methods from Monocular Image and Videos: A Comprehensive Survey,

    U. Rajapaksha, F. Sohel, H. Laga, D. Diepeveen, and M. Bennamoun, “Deep Learning-based Depth Estimation Methods from Monocular Image and Videos: A Comprehensive Survey,”ACM Comput. Surv., vol. 56, no. 12, Oct. 2024

    work page 2024

  10. [10]

    A Framework for Autonomous UA V Navigation Based on Monocular Depth Estimation,

    J. Gaigalas, L. Perkauskas, H. Gricius, T. Kanapickas, and A. Kriˇsˇci¯unas, “A Framework for Autonomous UA V Navigation Based on Monocular Depth Estimation,”Drones, 2025

    work page 2025

  11. [11]

    A frontier-based approach for autonomous exploration,

    B. Yamauchi, “A frontier-based approach for autonomous exploration,” inProceedings 1997 IEEE International Symposium on Computational Intelligence in Robotics and Automation CIRA’97. ’Towards New Computational Principles for Robotics and Automation’, 1997, pp. 146–151

    work page 1997

  12. [12]

    High resolution maps from wide angle sonar,

    H. P. Moravec and A. Elfes, “High resolution maps from wide angle sonar,”Proceedings. 1985 IEEE International Conference on Robotics and Automation, vol. 2, pp. 116–121, 1985

    work page 1985

  13. [13]

    OctoMap: An Efficient Probabilistic 3D Mapping Framework Based on Octrees,

    A. Hornung, K. M. Wurm, M. Bennewitz, C. Stachniss, and W. Bur- gard, “OctoMap: An Efficient Probabilistic 3D Mapping Framework Based on Octrees,”Autonomous Robots, 2013

    work page 2013

  14. [14]

    V oxblox: Incremental 3D Euclidean Signed Distance Fields for on- board MA V planning,

    H. Oleynikova, Z. Taylor, M. Fehr, R. Y . Siegwart, and J. I. Nieto, “V oxblox: Incremental 3D Euclidean Signed Distance Fields for on- board MA V planning,”IEEE/RSJ International Conference on Intelli- gent Robots and Systems (IROS), pp. 1366–1373, 2017

    work page 2017

  15. [15]

    UFOMap: An Efficient Probabilistic 3D Mapping Framework That Embraces the Unknown,

    D. Duberg and P. Jensfelt, “UFOMap: An Efficient Probabilistic 3D Mapping Framework That Embraces the Unknown,”IEEE Robotics and Automation Letters, vol. 5, pp. 6411–6418, 2020

    work page 2020

  16. [16]

    Inverse Depth Parametrization for Monocular SLAM,

    J. Civera, A. J. Davison, and J. M. M. Montiel, “Inverse Depth Parametrization for Monocular SLAM,”IEEE Transactions on Robotics, vol. 24, pp. 932–945, 2008

    work page 2008

  17. [17]

    FLaME: Fast Lightweight Mesh Estima- tion Using Variational Smoothing on Delaunay Graphs,

    W. N. Greene and N. Roy, “FLaME: Fast Lightweight Mesh Estima- tion Using Variational Smoothing on Delaunay Graphs,”2017 IEEE International Conference on Computer Vision (ICCV), pp. 4696–4704, 2017

    work page 2017

  18. [18]

    Obstacle Avoidance Based-Visual Navigation for Micro Aerial Vehicles,

    W. G. A. Castillo, V . P. Casaliglla, and J. L. P´olit, “Obstacle Avoidance Based-Visual Navigation for Micro Aerial Vehicles,”Electronics, vol. 6, p. 10, 2017

    work page 2017

  19. [19]

    Monocular vision SLAM for indoor aerial vehicles,

    K. Celik, S.-J. Chung, M. Clausman, and A. K. Somani, “Monocular vision SLAM for indoor aerial vehicles,” in2009 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems, 2009, pp. 1566– 1573

    work page 2009

  20. [20]

    Information-Driven Autonomous Exploration for a Vision-Based Mav,

    E. Palazzolo and C. Stachniss, “Information-Driven Autonomous Exploration for a Vision-Based Mav,”ISPRS Annals of the Photogram- metry, Remote Sensing and Spatial Information Sciences, pp. 59–66, 2017

    work page 2017

  21. [21]

    Sphere-Graph: A Compact 3D Topological Map for Robotic Navigation and Segmentation of Complex Environments,

    M. Spencer, R. Sawtell, and S. Kitchen, “Sphere-Graph: A Compact 3D Topological Map for Robotic Navigation and Segmentation of Complex Environments,”IEEE Robotics and Automation Letters, vol. 9, pp. 2567–2574, 2024

    work page 2024

  22. [22]

    SphereMap: Dynamic Multi- Layer Graph Structure for Rapid Safety-Aware UA V Planning,

    T. Musil, M. Petrl ´ık, and M. Saska, “SphereMap: Dynamic Multi- Layer Graph Structure for Rapid Safety-Aware UA V Planning,”IEEE Robotics and Automation Letters, vol. 7, pp. 11 007–11 014, 2022

    work page 2022

  23. [23]

    Bubble Planner: Planning High- speed Smooth Quadrotor Trajectories using Receding Corridors,

    Y . Ren, F. Zhu, W. Liuet al., “Bubble Planner: Planning High- speed Smooth Quadrotor Trajectories using Receding Corridors,”2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 6332–6339, 2022

    work page 2022

  24. [24]

    The MRS UA V System: Pushing the Frontiers of Reproducible Research, Real-world Deployment, and Education with Autonomous Unmanned Aerial Vehicles,

    T. Baca, M. Petrlik, M. Vrbaet al., “The MRS UA V 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. 26, pp. 1–28, May 2021

    work page 2021

  25. [25]

    Open- VINS: A Research Platform for Visual-Inertial Estimation,

    P. Geneva, K. Eckenhoff, W. Lee, Y . Yang, and G. P. Huang, “Open- VINS: A Research Platform for Visual-Inertial Estimation,”2020 IEEE International Conference on Robotics and Automation (ICRA), pp. 4666–4672, 2020

    work page 2020

  26. [26]

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

    N. P. Koenig and A. Howard, “Design and use paradigms for Gazebo, an open-source multi-robot simulator,”2004 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) (IEEE Cat. No.04CH37566), vol. 3, pp. 2149–2154 vol.3, 2004

    work page 2004