Pith. sign in

REVIEW 3 major objections 4 minor 20 references

Understanding Fire Through Thermal Radiation Fields for Mobile Robots

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read A mobile robot can navigate around fires safely by building a real-time thermal radiation field from fused depth and thermal images and letting an A* planner trade distance against heat.

desk verdict A worthwhile proof-of-concept for fire-aware navigation that extends thermal maps into free space, but the quantitative claims need cleanup: hand-tuned constants, a 68% flux overprediction, and an inconsistency between the model equation and the algorithm pseudocode. read the letter →

arxiv 2602.19108 v2 pith:IOB3Y5ZF submitted 2026-02-22 cs.RO

Anton R. Wagner , Madhan Balaji Rao , Xuesu Xiao , Sören Pirk This is my paper · ORCID
classification cs.RO
keywords thermalradiationfieldfire-awarenavigationStefan-Boltzmannlawthermal-depthfusionA*pathplanningquadrupedrobotline-of-sightheatpropagationcostmapsafetymargins
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

Fire is usually mapped as surface temperatures, but the hazard a robot faces lives in the empty space between it and the flame. This paper tries to establish that a robot can construct that missing free-space hazard in real time: fused depth and thermal images yield a temperature-annotated 3D point cloud, a fitted hemisphere localizes the fire, and a Stefan–Boltzmann inverse-square model spreads the flame's radiative power through the environment, gated by line of sight. The resulting thermal radiation field plugs directly into an A* cost map, so the planner can avoid hot regions while still reaching goals. Experiments on a quadruped with a controlled propane fire show paths that widen or tighten with a safety margin and that exploit wall occlusion as a heat shield. A calorimetric check shows the model overpredicts incident heat by about 68% and pushes the estimated safe distance out by 0.22 m—conservative for this test fire, but a sign that the constants carry the quantitative weight.

What carries the argument

The mechanism that carries the argument is a Stefan–Boltzmann inverse-square radiation field: a scalar map T[x,y] = P/(4πr^2)X defined at every free-space cell with line of sight to a fitted hemisphere fire source. The field is generated from registered depth–thermal clouds (a cluster of points above 100 °C for fire detection, a hemisphere footprint for area A), multiplied by the empirical factor γ = 0.4, and blended into occupancy as O = max(geometric, min(T/q_danger, 1)). The named danger threshold q_danger = 2.5/max(0.1, ϕ) converts heat flux into an impassable boundary; the A* cost C = 1 + Oβ then performs the route search. This lets one physical formula turn sparse surface temperatures

What would settle it

Set up a heat-flux or thermocouple array at the planner-computed danger boundary (e.g., 0.71 m and 0.93 m from the flame used in the paper) and at several larger radii. If the measured flux at any cell the planner marks safe exceeds the q_danger threshold, or if the decay with distance diverges visibly from 1/r^2 (e.g., because convection dominates), the central claim that the radiation field predicts safe navigation fails.

Watch

Extended reading notes

Core claim

The paper's central claim is that fire, for navigation purposes, is best represented not as a surface-temperature overlay but as a continuous radiative heat-flux field in free space, recoverable in real time from onboard sensors. The method fuses thermal and depth images into a 3D point cloud whose hottest cluster is treated as the fire source; a hemisphere fitted to that cluster supplies the emitting area A, and the Stefan–Boltzmann law P = σAT0^4γ with fixed T0 = 1,473.15 K and empirical γ = 0.4 yields the fire's radiative power. That power is spread over the environment by an inverse-square law with radiative fraction X = 0.35, gated by line of sight to the fire, and written into a 2D occ

Load-bearing premise

The quantitative safety of every planned path rests on assumed fire constants—flame temperature T0 = 1,473.15 K, correction factor γ = 0.4, and radiative fraction X = 0.35—and the paper's own calorimetry shows the model overpredicts incident heat by 68%, so the computed 'safe' distance is only as trustworthy as those constants.

Editorial extensions

If this is right

  • If correct, any cost-based planner can become fire-aware simply by adding a thermal radiation layer computed from a depth camera and a low-cost thermal camera.
  • The adjustable safety margin ϕ gives operators direct, real-time control over the efficiency–safety trade-off, as shown by the wide and narrow detours around the same fire.
  • The line-of-sight treatment of radiation means walls and tall obstacles create genuinely cooler corridors; the wall-occlusion experiment confirms the planner exploits this.
  • The calorimetric comparison implies that safe-distance estimates are conservative for the tested propane device—the model predicts 0.93 m clearance where measured flux would allow 0.71 m—so the system errs on the side of caution at the cost of extra detour.
  • The method is not surface-centric: it extrapolates from the visible flame to empty space, which is exactly the region a thermal camera alone cannot measure.

Reading between the lines

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

  • A testable extension would estimate the fire's radiative power online from measured fuel mass-flow rate, replacing the fixed T0 and γ and likely removing the 68% overprediction.
  • Because smoke and hot combustion gases also radiate and convect heat without line of sight to the flame, the safe-corridor-behind-a-wall behavior may not transfer to real smoke-filled fires; a heat-flux probe behind an occluding wall would test whether the line-of-sight assumption is conservative or unsafe.
  • The same representation should scale to multiple fires by summing P(r) contributions from independently detected clusters, provided line-of-sight occlusion is tested per source.
  • The 192.5 °C ceiling of the thermal sensor likely truncates the fire cluster and biases the hemisphere fit, so repeating with a higher-dynamic-range camera is a cheap way to test whether the power overestimate comes from the sensor or the physical constants.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes a fire-aware navigation framework for mobile robots. Depth and thermal images are registered into a temperature-annotated 3D point cloud; high-temperature points are clustered and fit with a hemisphere to localize the fire and estimate its radiative power via the Stefan–Boltzmann law. An inverse-square decay model is used to construct a 2D thermal radiation field with line-of-sight occlusion handling, which is fused into an occupancy grid and used to plan A* paths. The system is demonstrated on a Boston Dynamics Spot robot with a controlled propane fire, and a calorimetry experiment is used to validate the radiation magnitude.

Significance. The idea of extending surface-centric thermal maps to a navigable free-space radiation field is timely and practically relevant for fire response robotics. The paper includes real-robot experiments, a physical model with explicit assumptions, and an independent calorimetric check, which are strengths. However, the quantitative claims rest on several assumed constants (T0, γ, X) and the validation shows a 68% overprediction in flux. More importantly, there is a concrete mismatch between the model in Eq. (3) and the algorithm pseudocode in Algorithm 1. If the deployed planner follows Algorithm 1, the calibrated and validated field is not the field used during navigation. These issues need to be resolved before the central 'thermally safe paths' claim is fully supported. The qualitative detour behavior is plausible and useful, but the quantitative basis needs repair.

major comments (3)
  1. [§III-B, Algorithm 1, §IV-B] Algorithm 1 line 24 computes T[i,j] ← P/(4πd²), omitting the radiative fraction X=0.35 from Eq. (3). With P=78.5 kW and r=0.45 m, Eq. (3) gives 10.79 kW/m² and a safe distance of 0.93 m (matching §IV-B), whereas the Algorithm 1 expression gives 30.85 kW/m² and a safe distance of 1.58 m. The pseudocode and the validation therefore describe different radiation fields. Please clarify which expression is used in the deployed planner, and if the pseudocode is wrong, correct it. This is load-bearing because the claimed calibrated safety distance (0.93 m) does not correspond to the algorithm as written.
  2. [§III-B, Table I] There is a physical inconsistency in the definition of P. Table I labels P as 'Estimated radiative power of fire source,' but Eq. (3) multiplies P by X=0.35, described as the radiative fraction of total heat release. If P is already radiative output, X should be 1; if P is meant to be total heat release, Eq. (2) should be described as estimating heat release rate, not 'thermal radiation emitted from the fire surface.' This ambiguity changes the field by a factor of 1/X≈2.86 and affects every cost-map threshold.
  3. [§IV-B, §V] The validation shows the model overpredicts incident flux by 68% (10.79 vs 6.4 kW/m²) and safe distance by 0.22 m. Because the overprediction is conservative, it does not invalidate the avoidance demonstration, but the abstract and conclusions use the phrase 'thermally safe paths.' As the limitations section correctly notes, fire strength is assumed rather than estimated. I ask for either a margin/safety analysis demonstrating that the executed paths remain outside the measured hazard zones, or a more modest claim such as 'conservative avoidance of high-radiation regions.'
minor comments (4)
  1. [Eq. (7)] There is a typographical error: 'max(O[x, y],; min(...))' contains an extra semicolon after the comma.
  2. [Abstract, §IV] The paper claims 'real-time' thermal radiation fields, but no timing or latency measurements are reported. Please provide runtime data or remove the 'real-time' wording.
  3. [§I, §III-A, Algorithm 1] The sensor description is inconsistent: §III-A says depth is obtained from the stereo camera, while Algorithm 1 processes 'LiDAR scans.' Clarify which sensor(s) are used for occupancy and obstacle detection.
  4. [Throughout] Minor typos: 'Fore one' should be 'For one' (Discussion), 'UA V' should be 'UAV' (Introduction), and 'scaler' should be 'scalar' (Eq. 6). Also, some figure references are repeated or ambiguous (e.g., 'In Fig. 7 we show...' appears twice in §IV-A).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: empirical constants are acknowledged inputs; validation is an external calorimetric comparison.

full rationale

The derivation chain—thermal-depth fusion, fire localization, Stefan–Boltzmann radiation model, cost-map integration, A* planning—is self-contained in the sense that each stage is defined from explicit inputs (T0, gamma, X, A, r) and none of these inputs is defined in terms of the outputs (safe distance, safe path). The empirical correction factor gamma=0.4 is a calibrated constant, but the paper does not claim it was fitted to the calorimetric validation; in fact, the validation shows the model overpredicts incident radiation by 68% (10.79 vs 6.4 kW/m^2), which is evidence that gamma was not tuned to that particular measurement. The paper explicitly acknowledges in the limitations that the effective fire strength was assumed rather than estimated directly, which is an honest statement of model dependence rather than a circular step. The radiative fraction X=0.35 is cited from independent NIST work (Hamins et al.), not from the present authors. The Algorithm 1 omission of X relative to Eq. (3) is an internal inconsistency between the model description and the pseudocode, and it is a correctness risk (the deployed field may be more conservative than the validated one), but it is not a circular derivation: the algorithm's output is not used to define its own inputs. No self-citation is load-bearing; references are contextual or independent. Therefore no circularity step meets the required evidentiary standard.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central quantitative model depends on several constants that are assumed, hand-tuned, or taken from different fire types; a 68% overprediction in validation confirms the miscalibration. No new physical entities are introduced. The navigation concept is independent of exact values, but the claimed 'thermally safe' distances are not.

free parameters (6)
  • T0 (effective flame temperature) = 1,473.15 K
    Assumed in Eq. (1); sets radiated power via T^4; not measured for the propane device.
  • γ (emissivity/geometry correction) = 0.4
    Chosen empirically in Eq. (2); 'empirically found to provide reasonable estimates'; this is hand-fitting.
  • X (radiative fraction) = 0.35
    Taken from Hamins et al. [19] for large pool fires; applied to a small propane training fire without in-situ measurement.
  • ϕ (caution scalar) = 1.0 (varied)
    User-set scaling in Eq. (6) that sets the danger threshold qdanger; not derived from robot thermal limits.
  • β (cost shaping exponent) = not specified (≥1)
    Eq. (8) weights thermal cost versus efficiency; value never reported, affecting path tradeoffs.
  • DBSCAN ε and min_samples = ε=0.5 m, min_samples=2
    Chosen for fire clustering in Sec. III-B; not justified or sensitivity-analyzed.
assumptions (6)
  • domain assumption Stefan-Boltzmann and inverse-square point-source decay model radiative heat in free space
    Sec. III-B Eqs. (2)-(3); ignores convection, plume dynamics, and reflection, acknowledged in Sec. V.
  • ad hoc to paper Fire can be approximated as a single hemisphere with uniform temperature T0
    Hemisphere fitted to the bounding rectangle of the largest high-temperature cluster; no independent geometric validation.
  • domain assumption Thermal camera pixels above 100°C localize the fire surface
    Sec. III-B; the phone thermal camera saturates at 192.5°C (Sec. V), so true flame temperatures are not measured.
  • domain assumption Line-of-sight (Bresenham) is sufficient to model radiation occlusion by obstacles
    Eq. (5) and Algorithm 1; no scattering or reflected radiation is modeled.
  • domain assumption Thermal-depth extrinsic calibration is accurate enough for fire localization
    Sec. III-A uses Calibmar with ArUco targets; no reported reprojection error or quantitative calibration accuracy.
  • domain assumption Radiative fraction X=0.35 from large pool fires transfers to this propane training device
    Value from [19]; the paper does not measure radiative fraction for its own fire.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Understanding Fire Through Thermal Radiation Fields for Mobile Robots." pith.science (2026). https://pith.science/paper/IOB3Y5ZF

@misc{pith2026260219108,
  author       = {Pith},
  title        = {Pith review of: Understanding Fire Through Thermal Radiation Fields for Mobile Robots},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IOB3Y5ZF}},
  note         = {Machine review of arXiv:2602.19108}
}
read the original abstract

Safely moving through environments affected by fire is a critical capability for autonomous mobile robots deployed in disaster response. In this work, we present a novel approach for mobile robots to understand fire through building real-time thermal radiation fields. We register depth and thermal images to obtain a 3D point cloud annotated with temperature values. From these data, we identify fires and use the Stefan-Boltzmann law to approximate the thermal radiation in empty spaces. This enables the construction of a continuous thermal radiation field over the environment. We show that this representation can be used for robot navigation, where we embed thermal constraints into the cost map to compute collision-free and thermally safe paths. We validate our approach on a Boston Dynamics Spot robot in controlled experimental settings. Our experiments demonstrate the robot's ability to avoid hazardous regions while still reaching navigation goals. Our approach paves the way toward mobile robots that can be autonomously deployed in fire-affected environments, with potential applications in search-and-rescue, firefighting, and hazardous material response.

Figures

Figures reproduced from arXiv: 2602.19108 by the authors.

Figure 1
Figure 1. Experimental setup with the Boston Dynamics Spot [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of our framework: we use stereo depth and thermal sensor data to capture a fire (a) and to compute a [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. To register the right (a) and left (b) RGB frames [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Registration result: we show the registered RGB [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Experimental setup showing the Boston Dynamics [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Fire-aware navigation results for the Boston Dynamics Spot robot. When no fire is present the robot walks over the [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Thermal occlusion: In case of an obstacle blocks the fire source (e.g., a wall), our approach dynamically adjusts and [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 8
Figure 8. Figure 8: Navigation results for different fire sizes and levels [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 9
Figure 9. Figure 9: Thermal radiation profile of our model: we show [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: To estimate the thermal radiation, we performed a [PITH_FULL_IMAGE:figures/full_fig_p007_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 2 linked inside Pith

  1. [1]

    Portable fire evacuation guide robot system,

    Y .-D. Kim, Y .-G. Kim, S.-H. Lee, J.-H. Kang, and J. An, “Portable fire evacuation guide robot system,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2009

    2009

  2. [2]

    Analysis and design of human–robot swarm interaction in firefighting,

    A. Naghsh, J. Gancet, A. Tanoto, and C. R. Roast, “Analysis and design of human–robot swarm interaction in firefighting,” inIEEE RO-MAN, 2008

    2008

  3. [3]

    Extinguishing wildfires in large scale scenarios using swarms of uavs,

    G. P. Tzoumas, L. R. Salinas, A. McConville, T. G. Richardson, and S. Hauert, “Extinguishing wildfires in large scale scenarios using swarms of uavs,” inInternational Conference on Swarm Intelligence, 2024

    2024

  4. [4]

    Adoption of uav swarm technology: Survey and opinions of firefighters,

    A. McConville, G. P. Tzoumas, L. R. Salinas, M. C. Munera Ramirez, and S. Hauert, “Adoption of uav swarm technology: Survey and opinions of firefighters,”arXiv preprint, 2024

    2024

  5. [5]

    Autonomous firefighting using a quadruped robot,

    C. Baird and S. Nokleby, “Autonomous firefighting using a quadruped robot,”Transactions of the Canadian Society for Mechanical Engineer- ing, 2024. TABLE I: Summary of symbols and parameters used throughout the paper. Symbol Description Units / Notes rDistance from fire center to querypoint m T0 Effective flame temperature K AEstimated emitting surface ar...

    2024

  6. [6]

    M2p2: A multi- modal passive perception dataset for off-road mobility in extreme low- light conditions,

    A. Datar, A. Pokhrel, M. Nazeri, M. B. Rao, C. Pan, Y . Zhang, A. Harrison, M. Wigness, P. R. Osteen, J. Yeet al., “M2p2: A multi- modal passive perception dataset for off-road mobility in extreme low- light conditions,” inIEEE IROS, 2025

    2025

  7. [7]

    Targetless extrinsic calibration of stereo cameras, thermal cameras, and laser sensors in the wild,

    T. Fu, H. Yu, W. Yang, Y . Hu, and S. Scherer, “Targetless extrinsic calibration of stereo cameras, thermal cameras, and laser sensors in the wild,”arXiv preprint arXiv:2109.13414, 2021

    arXiv 2021

  8. [8]

    The project thermomap: Indoor mapping with thermal and rgb-d cameras,

    D. Borrmann, J. Elseberg, K. Lingemann, A. N ¨uchter, and J. Hertzberg, “The project thermomap: Indoor mapping with thermal and rgb-d cameras,” inIEEE/RSJ International Conference on Intelli- gent Robots and Systems (IROS), 2014

    2014

Show all 20 references
  1. [9]

    3d thermal mapping of building interiors using an rgb-d and thermal camera,

    S. Vidas, P. Moghadam, and M. Bosse, “3d thermal mapping of building interiors using an rgb-d and thermal camera,” inIEEE International Conference on Robotics and Automation (ICRA), 2013, pp. 2303–2310

    2013

  2. [10]

    3d radiometric mapping by means of lidar slam and thermal camera data fusion,

    D. De Pazzi, M. Pertile, and S. Chiodini, “3d radiometric mapping by means of lidar slam and thermal camera data fusion,”Sensors, vol. 22, no. 21, p. 8512, 2022

    2022

  3. [11]

    Sparse depth enhanced direct thermal-infrared slam beyond the visible spectrum,

    Y . Shin and A. Kim, “Sparse depth enhanced direct thermal-infrared slam beyond the visible spectrum,”IEEE Robotics and Automation Letters, vol. 4, no. 3, pp. 2911–2918, 2019

    2019

  4. [12]

    Eil-slam: Depth-enhanced edge-based infrared–lidar slam,

    W. Chen, Y . Wang, H. Chen, and Y . Liu, “Eil-slam: Depth-enhanced edge-based infrared–lidar slam,”Journal of Field Robotics, vol. 39, no. 8, pp. 1200–1217, 2021

    2021

  5. [13]

    Firebotslam: Thermal slam to increase situational awareness in smoke-filled environments,

    B. R. van Manen, V . Sluiter, and A. Y . Mersha, “Firebotslam: Thermal slam to increase situational awareness in smoke-filled environments,” Sensors, vol. 23, no. 17, p. 7611, 2023

    2023

  6. [14]

    Graph-based thermal–inertial slam with probabilistic neural networks,

    M. R. U. Saputra, C. X. Lu, P. P. B. de Gusmao, B. Wang, A. Markham, and N. Trigoni, “Graph-based thermal–inertial slam with probabilistic neural networks,”IEEE Transactions on Robotics, 2021

    2021

  7. [15]

    Evaluation of navigation sensors in fire smoke environments,

    J. W. Starr and B. Y . Lattimer, “Evaluation of navigation sensors in fire smoke environments,”Fire Technology, vol. 50, pp. 1459–1481, 2014

    2014

  8. [16]

    Physics informed neural fields for smoke reconstruction with sparse data,

    M. Chu, L. Liu, Q. Zheng, A. Franz, H.-P. Seidel, C. Theobalt, and R. Zayer, “Physics informed neural fields for smoke reconstruction with sparse data,”ACM Transactions on Graphics, vol. 41, no. 4, 2022

    2022

  9. [17]

    Activegs: Active scene reconstruction using gaussian splatting,

    L. Jin, X. Zhong, Y . Pan, J. Behley, C. Stachniss, and M. Popovi ´c, “Activegs: Active scene reconstruction using gaussian splatting,”IEEE Robotics and Automation Letters, 2025, arXiv:2412.17769

    2025 arXiv

  10. [18]

    A calibration tool for refractive underwater vision,

    F. Seegr ¨aber, M. She, F. Woelk, and K. K ¨oser, “A calibration tool for refractive underwater vision,” 05 2024

    2024

  11. [19]

    Thermal radiation from large pool fires,

    A. Hamins, K. McGrattan, and T. Kashiwagi, “Thermal radiation from large pool fires,” National Institute of Standards and Technology, Gaithersburg, MD, Tech. Rep. NISTIR 6546, 1999. [Online]. Available: https://nvlpubs.nist.gov/nistpubs/Legacy/IR/nistir6546.pdf

    1999

  12. [20]

    Merci and T

    B. Merci and T. Beji,Fluid Mechanics Aspects of Fire and Smoke Dynamics in Enclosures, 2nd ed. London (CRC Press/Taylor & Francis): CRC Press, 2022

    2022

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.