Olfactory Inertial Odometry: Sensor Calibration and Drift Compensation

Anirban Paul; Kordel K. France; Ovidiu Daescu; Shalini Prasad

REVIEW 3 major objections 6 minor 13 references

Olfactory Inertial Odometry: Sensor Calibration and Drift Compensation

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a two-stage calibration of measurement uncertainty lets a slow robotic arm fuse gas and inertial data to localize an odor source at centimeter level, across five sensor types and without deep learning.

desk verdict A clever VIO-to-olfaction transfer that is undermined by a Results section with no quantitative localization data. read the letter →

arxiv 2506.04539 v1 pith:UP646IIQ submitted 2025-06-05 cs.RO cs.ETcs.LGcs.SYeess.SY

Kordel K. France , Ovidiu Daescu , Anirban Paul , Shalini Prasad This is my paper

classification cs.ROcs.ETcs.LGcs.SYeess.SY

keywords olfactoryinertialodometrymachineolfactionsensorcalibrationextendedKalmanfilterdriftcompensationbeliefmapodorsourcelocalizationroboticarm

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that a visual-inertial-odometry-style fusion of olfactory and inertial measurements can reliably locate an odor source if the sensor fusion filter is seeded with two calibrated uncertainties. It defines a protocol that generalizes over five gas sensor families, using a bout-detection moving average to set sensor-level (Type A) noise and a trilaterated belief map to set spatial (Type B) uncertainty. These values populate the covariance matrices of an extended Kalman filter, and the authors demonstrate on a real robotic arm that this calibrated approach outperforms a cold-start task and reaches centimeter-level localization. The significance is that careful calibration and noise priors can substitute for high-computation deep learning in basic scent tracking, opening applications in robot-assisted surgery and touchless security screening. The authors also state that much remains to be proven and list additional experimental controls as future work.

What carries the argument

The key machinery is the two-uncertainty calibration feeding an extended Kalman filter. Type A uncertainty is $u_a = s/\sqrt{k}$, where $s$ is the standard deviation over a bout-detection moving-average window $k$ of the olfactory differential signal; it captures per-sensor, per-degree-of-freedom noise. Type B uncertainty is $u_b = 2v/\sqrt{m}$, where $v$ is the Euclidean distance from the tri-laterated Voronoi vertex to the true source and $m$ is the number of robot moves; it captures spatial belief-map error. These values are placed along the diagonals of $R$ and $Q$ respectively, so the EKF's fusion of gas response and inertial kinematics is tuned by physically measured noise rather than hand-set parameters.

What would settle it

Measure the reported source location against a known source while changing airflow direction between trials: if the localization error $v$ in the Type B formula grows whenever the plume is advected away from the sensor's line of travel, the equal-radius sphere model is false and the calibration cannot support centimeter-level claims.

Watch

Extended reading notes

Core claim

The central discovery is that olfactory inertial odometry (OIO) can be calibrated generically for multiple sensor types by separating uncertainty into two components: Type A, computed per degree of freedom from the standard deviation of a bout-detection window over five initial measurements, and Type B, computed from a wireless-signal-inspired belief map in which each olfactory reading defines a sphere of equal-probability source locations, with sphere intersections converging on a Voronoi vertex. The uncertainties are placed on the diagonals of the sensor noise covariance R and process noise covariance Q of an extended Kalman filter that fuses the olfactory signal with joint-level inertial data. On a real robotic arm carrying five different sensor types, the calibrated filter localizes an odor source at centimeter level, a precision the authors contrast with meter-level quadcopter localization, and this is achieved without deep learning at the edge.

Load-bearing premise

The belief-map stage assumes each sensor reading can be treated as the radius of a sphere whose surface points are equally likely to contain the odor source, so intersecting spheres narrow the source to a Voronoi vertex, even though real gas plumes are turbulent and anisotropic.

Editorial extensions

If this is right

The same calibration protocol works for NDIR, photoacoustic, electrochemical, and both MOX sensor families, so the fusion stack need not be redesigned when the gas sensor is swapped.
A slow robotic arm can localize an odor source at centimeter level, a precision the paper contrasts with meter-level quadcopter-based systems.
Calibrated OIO removes the need for high-computation deep learning in basic scent tracking, making edge deployment feasible on constrained hardware.
Seeding the EKF covariance matrices with the two calibrated uncertainties is sufficient to improve performance over a cold-start olfactory navigation task.

Reading between the lines

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

The same two-uncertainty calibration could be applied to non-olfactory sensors whose responses map to distance, such as acoustic or RF signal strength, whenever a trilaterated belief map can be built; the paper only demonstrates gas sensors.
If the isotropic sphere assumption fails under turbulent airflow, an unstated fix is to replace equal-radius spheres with anisotropic plume-shape priors, which would change the Type B formula.
The paper averages two redundant sensors but does not use the gradient between them; using that gradient as a directional cue, which the authors list as future work, could reduce the number of moves $m$ needed to reach the Voronoi vertex.
A testable extension is to run the same calibration on a mobile base at higher speed to see whether centimeter-level accuracy degrades with platform dynamics, since the paper explicitly targets slow robotic-arm motion.

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.

Desk Editor's Note

read the letter

Colleague,

This paper is an odd mix: a reasonable research idea and an empty Results section. The central claim—calibrated OIO gives centimeter-level odor source localization on a slow robot arm—is not actually demonstrated anywhere in the text. Figure 4 is captioned 'Results for EKF Pre- and Post-Calibration' but has no axis labels with numbers, no error bars, and no mention of trials. Table I lists calibration times and sensor errors in parts per million, not localization accuracy. The conclusion literally says 'future work will define additional experimental controls,' which is an admission that the essential controls are missing.

What is genuinely new: the authors port VIO-style EKF calibration to olfaction, define a protocol that covers five sensor types (EC, two MOX variants, PA, NDIR), and demonstrate it on a real robotic arm rather than a quadcopter or simulation. That is a useful step for machine olfaction and could reduce reliance on deep learning. The sensor diversity is a plus.

The soft spots are not minor. The Type B uncertainty formula u_b = 2v/sqrt(m) is asserted without derivation, and v is defined as the distance from the estimated Voronoi vertex to the true source on calibration runs. That value is then placed into the process noise Q used for future localization. Without a held-out benchmark, any improvement over cold-start is partly self-confirming. The belief-map model also assumes each RSSI reading defines a sphere of equiprobable source locations, which is hard to stomach for turbulent, time-varying gas plumes; if that assumption fails, the whole EKF fusion built on it is shaky.

There are minor issues too: reference [9] is an App Store link, and there is a stray sentence about 'simulation design' that does not match the actual experiments.

The conceptual direction is worth keeping alive. The paper as submitted should not be accepted, but it is not nonsense either. With a proper evaluation—numerical localization errors, a cold-start or uncalibrated baseline, trial counts, error bars, and a derivation or justification of the Type B term—this could become a useful contribution.

For peer review: I would send it to referees rather than desk reject. The idea is specific and testable, and the missing data is exactly what referees can request. Just don't expect it to survive in its current form.

Referee Report

3 major / 6 minor

Summary. The manuscript defines 'olfactory inertial odometry' (OIO), an analog of visual-inertial odometry in which gas sensor readings are fused with joint-level inertial data from a robotic arm through an extended Kalman filter (EKF). The proposed calibration protocol has two stages: Type A uncertainty, computed as the standard error of bout-detection differential measurements per degree of freedom, and Type B uncertainty, computed from the distance between a Voronoi-vertex estimate and the true source over a calibration run. These quantities are placed on the diagonals of the sensor-noise and process-noise covariance matrices R and Q. The authors claim that this calibration generalizes over five sensor types (EC, two MOX variants, PA, NDIR) and improves performance over a cold-start olfactory navigation task on a real robotic arm, targeting centimeter-level odor-source localization for surgical and security applications.

Significance. If the empirical claims held, the paper would make a useful contribution to machine olfaction: a low-computation, sensor-agnostic calibration procedure that avoids deep learning and is demonstrated on physical hardware rather than in simulation. The two-stage Type A / Type B uncertainty structure feeding an EKF is a sensible and potentially transferable pipeline, and the authors deserve credit for stating concrete protocol steps, reporting per-sensor calibration times, and targeting a well-motivated application domain. However, the significance is conditional on the experimental demonstration, which the manuscript does not provide: there are no quantitative localization results, no trial counts, no error bars, and no held-out cold-start benchmark, so the paper's central quantitative claims are unverified. No code, datasets, or machine-checked proofs are released, and the most consequential formulas are asserted without derivation.

major comments (3)

[IV (Results), Fig. 4, Table I] The central claim that OIO calibration 'improves performance over a cold-start olfactory navigation task' and achieves centimeter-level localization is not supported by any quantitative result in Section IV. Table I reports calibration times and sensor errors in PPM, not tracking or localization error, and Figure 4, captioned 'Results for EKF Pre- and Post-Calibration,' contains no numerical error values, no trial counts, and no confidence intervals, and is not even referenced by number in the text. The Conclusion's statement that 'future work will define additional experimental controls' is an explicit acknowledgment that the essential cold-start baseline comparison is currently missing. Because the abstract and introduction assert the improvement, and the only claimed evidence is this unlabeled figure, the manuscript's primary empirical contribution cannot be verified as written.
[III-C, Eq. (3)] Equation (3), u_b = 2v/sqrt(m), is asserted without derivation, and it raises a circularity concern that is load-bearing for the calibration methodology: v is defined as the Euclidean distance from the estimated Voronoi vertex to the true odor source on calibration runs, and the resulting u_b is then placed on the diagonal of the EKF process noise Q used for subsequent localization. Without a held-out benchmark in which v is measured on runs that did not inform Q, the reported improvement is partly self-confirming. In addition, the belief-map construction in Section III-C assumes each RSSI reading defines a sphere of equal-probability source locations, which requires an isotropic and static sensor-to-distance mapping that is generally not satisfied by turbulent, time-varying gas plumes; if that mapping fails, both u_b and the EKF fusion built on it are invalid. A concrete test would be to localize sources at held-out positions and compare localization error between Q calibrated on a disjoint set of runs and Q obtained from a cold start.
[III-A, III-B, Table I] The claimed generalization 'over several olfaction sensor types' is not demonstrated: Table I lists only calibration times and PPM error per sensor, with no per-sensor-type localization accuracy, and Section III-A states that 'optimal windows for each DoF are 3, 5, 7, and 11,' meaning the moving-average window k is a tuned free parameter per DoF. Section IV further acknowledges that 'additional fine-tuning of the parameters may be valuable' for EC and MOX sensors, and no sensitivity analysis of k, of the Type B scale factor, or of the extra fine-tuning is reported. The dependence on these tuned parameters weakens the claim of a general calibration procedure, since the results could in principle be driven by per-sensor, per-DoF tuning rather than by the proposed calibration process itself.

minor comments (6)

[III-C] The word 'repeates' should be 'repeats,' and the spaced rendering 'V oronoi' appears throughout the manuscript, apparently from a LaTeX macro or rendering issue.
[III-C, Eq. (3)] Equation (3) is dimensionally inhomogeneous as written (a distance v divided by sqrt(m) and scaled by 2), and the origin of the factor 2, as well as the variance interpretation of the resulting quantity, should be stated even if the formula is ultimately retained.
[III-A, Eq. (2)] Equation (2) uses the same symbol k for the moving-average window and for the sample size in the standard error; the two roles should be distinguished, since the window length is stated to vary by DoF and sensor type.
[III-B] The encoder drift value is reported as '1e-3 deg/sec^2,' which is an angular acceleration rather than a drift rate; the intended quantity (drift per unit time or per move) should be clarified.
[III-B] The description of the 1-DoF search as a '3-armed bandit problem' is not elaborated: no action-reward definition, regret measure, or learning update is given, so the bandit framing does not add information to the described casting heuristic.
[References] Reference [12] is incomplete: no publication venue or arXiv identifier is given for 'Learning to fly in seconds.'

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: Type B uses calibration ground-truth error to set EKF noise, and the claimed cold-start improvement is an empirical claim that lacks reported controls rather than a reduction to inputs.

full rationale

The paper's derivation chain is a calibration procedure rather than a self-referential derivation. Type A uncertainty in Eq. (2) is an ordinary standard-error estimate from the bout-detection window. Type B uncertainty in Eq. (3) does use v, the distance from the belief-map Voronoi vertex to the true source on calibration runs, and this value is placed on the diagonal of the EKF process-noise covariance Q in Section IV. That is a calibration step: ground-truth error is used to set a noise parameter, and no equation forces the later EKF output to equal v or makes the upstream Voronoi vertex the final EKF estimate. The paper's headline claim of centimeter-level accuracy and improvement 'over a cold-start olfactory navigation task' is an empirical claim, but Fig. 4 is captioned 'Results for EKF Pre- and Post-Calibration' without reporting numerical errors, trial counts, or a cold-start baseline, and the Conclusion concedes 'Future work will define additional experimental controls.' This is a serious evidentiary gap—the improvement is not demonstrated—but it is not a circular reduction: the reported calibration equations do not define the claimed result in terms of itself. The only self-citations ([6], [9]) are used for background or as a reference entry, not as load-bearing justification, so they do not create circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on an unvalidated geometric plume model and on several hand-chosen parameters. The ledger lists the moving-average window, the Type B scaling, and the assumptions that the RSSI-sphere model, diagonal covariance placement, and cross-sensor bout detection are valid.

free parameters (3)

Moving-average window length k per DoF = 3, 5, 7, 11 for 1, 2, 3, 5 DoF
The paper calls these optimal but gives no optimization procedure; the choice directly sets the bout detection sensitivity and the Type A uncertainty u_a = s/sqrt(k).
Type B uncertainty scale = ub = 2 v / sqrt(m)
The factor 2 and the square-root form are asserted without derivation; v is measured on calibration runs, so the formula is an ad hoc scaling rather than a derived quantity.
Extra EKF fine-tuning for EC and MOX sensors = unspecified
Section IV says additional fine-tuning of Q and R may be valuable for EC and MOX sensors, but the values and procedure are not given.

assumptions (4)

domain assumption Bout detection rules for metal-oxide sensors apply unchanged to electrochemical, photoacoustic, and NDIR sensors.
Section III-A states "we leverage the same principles for all five sensor types" based on [2] and [10], but the sampling dynamics differ substantially across these technologies.
domain assumption Each olfactory RSSI reading can be represented as the radius of a sphere whose surface points have equal probability of containing the odor source.
Section III-C builds the entire belief map and Voronoi vertex procedure on this isotropic, static plume assumption, which conflicts with known plume turbulence.
domain assumption Type A and Type B uncertainties can be placed on the diagonals of R and Q respectively with no off-diagonal cross-covariances.
Section IV says "we place our uncertainties along the diagonals", assuming the sensor and process noise are independent and fully described by these scalars.
domain assumption The robot's encoder drift is negligible at less than 1e-3 deg/s^2 as stated by the manufacturer.
Section III uses manufacturer-specified encoder drift as a given, without independent verification.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Olfactory Inertial Odometry: Sensor Calibration and Drift Compensation." pith.science (2026). https://pith.science/paper/UP646IIQ

@misc{pith2026250604539,
  author       = {Pith},
  title        = {Pith review of: Olfactory Inertial Odometry: Sensor Calibration and Drift Compensation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UP646IIQ}},
  note         = {Machine review of arXiv:2506.04539}
}

read the original abstract

Visual inertial odometry (VIO) is a process for fusing visual and kinematic data to understand a machine's state in a navigation task. Olfactory inertial odometry (OIO) is an analog to VIO that fuses signals from gas sensors with inertial data to help a robot navigate by scent. Gas dynamics and environmental factors introduce disturbances into olfactory navigation tasks that can make OIO difficult to facilitate. With our work here, we define a process for calibrating a robot for OIO that generalizes to several olfaction sensor types. Our focus is specifically on calibrating OIO for centimeter-level accuracy in localizing an odor source on a slow-moving robot platform to demonstrate use cases in robotic surgery and touchless security screening. We demonstrate our process for OIO calibration on a real robotic arm and show how this calibration improves performance over a cold-start olfactory navigation task.

Figures

Figures reproduced from arXiv: 2506.04539 by the authors.

**Figure 2.** A diagram showing the MQ-series metal oxide sensor and NDIR [PITH_FULL_IMAGE:figures/full_fig_p001_2.png] view at source ↗

**Figure 3.** Four belief map spheres intersecting to form the Voronoi vertex. [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Results for EKF Pre- and Post-Calibration. Axis units are in millimeters. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

13 extracted references · 11 canonical work pages

[9]

Scentience,

K. K. France, “Scentience,” https://apps.apple.com/us/app/scentience/ id6741092923, 2025, accessed: 2025-03-02. [Online]. Available: https://apps.apple.com/us/app/scentience/id6741092923

work page 2025
[1]

High-speed odor sensing using miniaturized electronic nose,

N. Dennler, D. Drix, T. P. A. Warner, S. Rastogi, C. D. Casa, T. Ackels, A. T. Schaefer, A. van Schaik, and M. Schmuker, “High-speed odor sensing using miniaturized electronic nose,” Science Advances, vol. 10, no. 45, p. eadp1764, 2024. [Online]. Available: https://www.science.org/doi/abs/10.1126/sciadv.adp1764

doi:10.1126/sciadv.adp1764 2024
[2]

Exploiting plume structure to decode gas source distance using metal-oxide gas sensors,

M. Schmuker, V . Bahr, and R. Huerta, “Exploiting plume structure to decode gas source distance using metal-oxide gas sensors,” Sensors and Actuators B: Chemical, vol. 235, pp. 636–646, 2016. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0925400516307833

work page 2016
[3]

Environmental chemical sensing using small drones: A review,

J. Burgu ´es and S. Marco, “Environmental chemical sensing using small drones: A review,” Science of The Total Environment , vol. 748, p. 141172, 2020. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S004896972034701X

work page 2020
[4]

Smelling nano aerial vehicle for gas source localization and mapping,

J. Burgu ´es, V . Hern´andez, A. J. Lilienthal, and S. Marco, “Smelling nano aerial vehicle for gas source localization and mapping,” Sensors, vol. 19, no. 3, 2019. [Online]. Available: https://www.mdpi.com/ 1424-8220/19/3/478

work page 2019
[5]

Sniffy bug: A fully autonomous swarm of gas-seeking nano quadcopters in cluttered environments,

B. P. Duisterhof, S. Li, J. Burgu ´es, V . J. Reddi, and G. C. H. E. de Croon, “Sniffy bug: A fully autonomous swarm of gas-seeking nano quadcopters in cluttered environments,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 2021, pp. 9099– 9106

work page 2021
[6]

Emergent behavior in evolutionary swarms for machine olfaction,

K. K. France, A. Paul, I. Banga, and S. Prasad, “Emergent behavior in evolutionary swarms for machine olfaction,” in Proceedings of the Genetic and Evolutionary Computation Conference , ser. GECCO ’24. New York, NY , USA: Association for Computing Machinery, 2024, p. 30–38. [Online]. Available: https://doi.org/10.1145/3583131.3590376

work page arXiv 2024
[7]

Integrating vision and olfaction via multi-modal llm for robotic odor source localization,

S. Hassan, L. Wang, and K. R. Mahmud, “Integrating vision and olfaction via multi-modal llm for robotic odor source localization,” Sensors, vol. 24, no. 24, 2024. [Online]. Available: https://www.mdpi. com/1424-8220/24/24/7875

work page 2024

Show all 13 references

[8]

Emergent behaviour and neural dynamics in artificial agents tracking odour plumes,

S. H. Singh, F. van Breugel, R. P. N. Rao, and B. W. Brunton, “Emergent behaviour and neural dynamics in artificial agents tracking odour plumes,” Nature Machine Intelligence , vol. 5, no. 1, pp. 58–70, Jan
[10]

Drift in a popular metal oxide sensor dataset reveals limitations for gas classification benchmarks,

N. Dennler, S. Rastogi, J. Fonollosa, A. van Schaik, and M. Schmuker, “Drift in a popular metal oxide sensor dataset reveals limitations for gas classification benchmarks,” Sensors and Actuators B: Chemical, vol. 361, p. 131668, 2022. [Online]. Available: https: //www.scienced...

2022
[11]

Review—room-temperature ionic liquids for electrochemical application with special focus on gas sensors,

A. Paul, S. Muthukumar, and S. Prasad, “Review—room-temperature ionic liquids for electrochemical application with special focus on gas sensors,” Journal of The Electrochemical Society , vol. 167, no. 3, p. 037511, dec 2019. [Online]. Available: https: //dx.doi.org/10.1149/2.0...

2019 doi
[12]

Learning to fly in seconds,

J. Eschmann, D. Albani, and G. Loianno, “Learning to fly in seconds,” 2023. 4

2023
[2023]

Available: https://doi.org/10.1038/s42256-022-00599-w

[Online]. Available: https://doi.org/10.1038/s42256-022-00599-w

doi

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [9]

Scentience,

K. K. France, “Scentience,” https://apps.apple.com/us/app/scentience/ id6741092923, 2025, accessed: 2025-03-02. [Online]. Available: https://apps.apple.com/us/app/scentience/id6741092923

work page 2025

[2] [1]

High-speed odor sensing using miniaturized electronic nose,

N. Dennler, D. Drix, T. P. A. Warner, S. Rastogi, C. D. Casa, T. Ackels, A. T. Schaefer, A. van Schaik, and M. Schmuker, “High-speed odor sensing using miniaturized electronic nose,” Science Advances, vol. 10, no. 45, p. eadp1764, 2024. [Online]. Available: https://www.science.org/doi/abs/10.1126/sciadv.adp1764

doi:10.1126/sciadv.adp1764 2024

[3] [2]

Exploiting plume structure to decode gas source distance using metal-oxide gas sensors,

M. Schmuker, V . Bahr, and R. Huerta, “Exploiting plume structure to decode gas source distance using metal-oxide gas sensors,” Sensors and Actuators B: Chemical, vol. 235, pp. 636–646, 2016. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0925400516307833

work page 2016

[4] [3]

Environmental chemical sensing using small drones: A review,

J. Burgu ´es and S. Marco, “Environmental chemical sensing using small drones: A review,” Science of The Total Environment , vol. 748, p. 141172, 2020. [Online]. Available: https://www.sciencedirect.com/ science/article/pii/S004896972034701X

work page 2020

[5] [4]

Smelling nano aerial vehicle for gas source localization and mapping,

J. Burgu ´es, V . Hern´andez, A. J. Lilienthal, and S. Marco, “Smelling nano aerial vehicle for gas source localization and mapping,” Sensors, vol. 19, no. 3, 2019. [Online]. Available: https://www.mdpi.com/ 1424-8220/19/3/478

work page 2019

[6] [5]

Sniffy bug: A fully autonomous swarm of gas-seeking nano quadcopters in cluttered environments,

B. P. Duisterhof, S. Li, J. Burgu ´es, V . J. Reddi, and G. C. H. E. de Croon, “Sniffy bug: A fully autonomous swarm of gas-seeking nano quadcopters in cluttered environments,” in 2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 2021, pp. 9099– 9106

work page 2021

[7] [6]

Emergent behavior in evolutionary swarms for machine olfaction,

K. K. France, A. Paul, I. Banga, and S. Prasad, “Emergent behavior in evolutionary swarms for machine olfaction,” in Proceedings of the Genetic and Evolutionary Computation Conference , ser. GECCO ’24. New York, NY , USA: Association for Computing Machinery, 2024, p. 30–38. [Online]. Available: https://doi.org/10.1145/3583131.3590376

work page arXiv 2024

[8] [7]

Integrating vision and olfaction via multi-modal llm for robotic odor source localization,

S. Hassan, L. Wang, and K. R. Mahmud, “Integrating vision and olfaction via multi-modal llm for robotic odor source localization,” Sensors, vol. 24, no. 24, 2024. [Online]. Available: https://www.mdpi. com/1424-8220/24/24/7875

work page 2024