Contact-Aided Factor-Graph Localization for Underwater Sampling

arXiv:2608.26932 · cs.RO · Submitted 2026-08-27 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: I'm Rosa.

Dev: And I'm Dev, here with Taro.

Rosa: We're looking at the paper: "Contact-Aided Factor-Graph Localization for Underwater Sampling".

Dev: Here's the gist of the paper. Accurate state estimation for autonomous underwater vehicles performing close-range seafloor sampling remains challengingIn low-altitude operation, down-looking cameras over featureless planar seabeds produce scale ambiguity, lateral degeneracy, and inconsistent feature trackingWe propose a Contact-Aided Factor-Graph Localization framework that treats physical interaction as an informative geometric constraint within a smoothing-based localization formulationExperimental evaluation in tanks, harbor, and simulation environments demonstrates that contact-induced constraints significantly reduce trajectory drift and improve object revisit accuracy compared to filtering-based navigation and contactfree graph formulationsThese results highlight the role of embodied physical interaction as a localization primitive in perception-degraded underwater environments Core Problem Definition Autonomous underwater vehicles (AUVs) performing close-range seafloor sampling operate in severely perceptiondegraded conditionsGPS is unavailable, and navigation relies on inertial–DVL dead reckoning, which accumulates driftUnderwater imagery suffers from scattering, turbidity, low texture, and photometric distortionThe scene is predominantly planar and texture-sparse, rendering monocular visual odometry geometrically ill-conditionedAccurate localization is essential for object approach, suction-based grasping, and repeated samplingThe core challenge is therefore how to obtain structurally informative constraints in environments where conventional exteroceptive sensing is weakWe propose to treat physical interaction as a localization primitiveWhen the AUV establishes suction contact with a benthic object, the end-effector creates a metrically grounded, viewpoint-invariant geometric constraintWe introduce a Contact-Aided FactorGraph Localization framework that tightly fuses IMU, DVL, pressure, and magnetometer measurements with adaptive down-looking visual odometry and manipulator contact constraintsThe system addresses three challenges: Planar visual degeneracy by using adaptive uncertainty scaling to mitigate estimator inconsistency under weak monocular geometryDrift without natural loop closures is mitigated by formulating suction contact as a high-confidence geometric factor that induces implicit loop closureA hybrid slidingwindow strategy alternates between landmarkaugmented and pose-only refinement to preserve real-time robustness against intermittent visual structure Sensor Fusion and Constraint Modeling The framework formulates underwater state estimation as a contact-aided smoothing problem on a factor graph, implemented using GTSAMVehicle poses, velocities, IMU biases, and environmental landmarks are jointly estimated by fusing proprioceptive navigation measurements with visual observations and manipulator contact constraints within a unified probabilistic frameworkThe inertial sensors provide a tightly coupled backbone where IMU preintegration links consecutive states using the formula (3)The DVL provides a velocity factor by incorporating body-frame velocity corrected for rotational motionPressure measurements are fused as a prior on the vertical position using the temperature-compensated density modelThe visual odometry pipeline employs contrast enhancement and ORB feature extraction, with metric scale recovered from auxiliary sensors like DVL or pressureVisual uncertainty is adaptively scaled using the inlier ratio to prevent visually weak frames from destabilizing the estimator Contact-Induced Geometric Anchoring The manipulator-based contact sensing treats the end-effector as a proprioceptive sensor providing contact-based geometric constraints through a suction end-effectorWhen the suction cup engages, the current position of the end-effector pee is calculated and treated as a point measurement in the environmentIf an existing landmark lj lies within a proximity threshold, a bearing–range factor is added between the current pose xk and the landmarkThis mechanism yields an implicit loop closure when the end-effector contacts a previously mapped object without requiring explicit place recognitionThe manipulator thus acts as an active tactile sensor, providing sparse but high-confidence constraints that stabilize the factor graph during prolonged low-visibility operation Experimental Validation and Results In the first tank experiments, introducing two 180◦ rotations increases ATE to 0.2917 m and yaw RMSE to 0.1232 rad due to weak lateral and heading observabilityAdding a starfish landmark during rotation improves yaw stability (0.0867 rad) but further increases ATE (0.3514 m), indicating that a single visual landmark does not resolve global planar degeneracyThe combination of depth variation and landmark constraints yields the best performance with ATE 0.1651 m and yaw RMSE 0.0412 radIn the harbor deployment, task-level revisit accuracy improves from 1.23 m without landmarks to 0.90 m when pump activation events are includedThis reduction demonstrates that contact-induced constraints provide structurally informative corrections even under visually degraded conditions We presented a contact-aided factor-graph framework in which physical interaction helps resolve degeneracy in perception-limited underwater SLAMStructured motion and deliberate physical interaction introduce metrically anchored constraints that improve consistency where manipulation occursThis trend also holds in harbor conditions, where turbidity and illumination variability increase global ATE, but relative pose errors remain low and contact events provide repeatable correctionsDrift in perception-limited underwater missions is not only a sensing problem, but an observability problemFuture work will extend this framework toward dense, mission-time mapping by coupling contact-aided localization with high-rate volumetric approaches such as FRAGG-MAP to enable drift-aware reconstruction of benthic workspaces for occupancy and manipulation planning REFERENCES

1: M.

Rosa: Ferrera, J.

Dev: Moras, P.

Rosa: Trouve-Peloux, and V.

Paper summary: Rosa: We just talked about the core idea of this paper, "Contact-Aided Factor-Graph Localization for Underwater Sampling," which is using physical contact to anchor localization constraints in environments where visual and inertial sensors are weak.

Dev: Right, so the authors propose a Contact-Aided FactorGraph Localization framework that treats physical interaction as an informative geometric constraint inside a smoothing formulation. They are fusing suction-based manipulator contact events with adaptive visual odometry and other on board sensors to estimate the state of the underwater vehicle.

Taro: From what I see, they're tackling three main things: first, planar visual degeneracy using adaptive uncertainty scaling to handle those weak monocular geometries; second, drift without natural loop closures by making suction contact a high-confidence geometric factor; and third, maintaining real-time robustness with a hybrid sliding window strategy that alternates between landmark augmented and pose only refinement.

Rosa: That adaptive scaling sounds like a smart way to manage the visual uncertainty when the visual features are sparse or bad.

Dev: And modeling contact as a high-confidence factor is what’s really compelling because it creates implicit loop closures that correct drift without needing explicit place recognition.

Conclusion: Rosa: So, looking at the full scope of "Contact-Aided Factor-Graph Localization for Underwater Sampling," the authors argue that structured motion and deliberate physical interaction introduce metrically anchored constraints that improve consistency precisely when manipulation is happening.

Dev: And they’ve shown this trend holds in harbor conditions too, where things like turbidity and lighting variability might increase global absolute translational error, but their relative pose errors stay low, and those contact events provide repeatable corrections.

Taro: It seems the main implication is that drift in perception-limited underwater missions isn't just a sensing problem anymore; it’s fundamentally an observability problem that can be solved by incorporating physical interaction as a localization primitive.

Rosa: Exactly. This approach moves beyond just looking for visual features to using the vehicle's physical actions to ground the estimation of its position, which is crucial for future work involving drift-aware reconstruction for occupancy and planning.

School of Engineering & Physical Sciences, Heriot-Watt University of Edinburgh and Engineering Department, Japan Agency for Marine-Earth Science and Technology

cs.RO

Submitted: 2026-08-27

Updated: 2026-11-16

Code: https://github.com/borglab/gtsam

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 98/100

The gist: Accurate state estimation for autonomous underwater vehicles performing close-range seafloor sampling remains challengingIn low-altitude operation…

Key concepts

Contact-Aided FactorGraph Localization
This is a method that estimates the vehicle's position by combining multiple sensor measurements (like IMU and DVL) with new information gained from physical contact. When the vehicle touches an object, this interaction provides a strong, metric geometric anchor that helps stabilize the estimation process.
Planar Visual Degeneracy
This occurs when a camera looking down onto a flat seafloor cannot determine its precise position because the scene lacks unique features or depth cues. The framework addresses this by using adaptive uncertainty scaling to ensure the visual estimates remain consistent even when visual information is weak.
Contact-Induced Geometric Anchoring
When the vehicle's manipulator makes suction contact with a benthic object, it creates a fixed point in space relative to the vehicle. This contact acts like an implicit loop closure, providing high-confidence geometric constraints that correct accumulated drift without needing explicit visual landmarks.
Factor Graph Localization
This is a mathematical approach to state estimation where the system models the vehicle's position and sensor measurements as interconnected factors in a graph. The framework uses this to tightly fuse data from inertial sensors, visual observations, and physical contact constraints into one consistent estimate.

Terminology

Summary

Accurate state estimation for autonomous underwater vehicles performing close-range seafloor sampling remains challengingIn low-altitude operation, down-looking cameras over featureless planar seabeds produce scale ambiguity, lateral degeneracy, and inconsistent feature trackingWe propose a Contact-Aided Factor-Graph Localization framework that treats physical interaction as an informative geometric constraint within a smoothing-based localization formulationExperimental evaluation in tanks, harbor, and simulation environments demonstrates that contact-induced constraints significantly reduce trajectory drift and improve object revisit accuracy compared to filtering-based navigation and contactfree graph formulationsThese results highlight the role of embodied physical interaction as a localization primitive in perception-degraded underwater environments

Core Problem Definition

Autonomous underwater vehicles (AUVs) performing close-range seafloor sampling operate in severely perceptiondegraded conditionsGPS is unavailable, and navigation relies on inertial–DVL dead reckoning, which accumulates driftUnderwater imagery suffers from scattering, turbidity, low texture, and photometric distortionThe scene is predominantly planar and texture-sparse, rendering monocular visual odometry geometrically ill-conditionedAccurate localization is essential for object approach, suction-based grasping, and repeated samplingThe core challenge is therefore how to obtain structurally informative constraints in environments where conventional exteroceptive sensing is weakWe propose to treat physical interaction as a localization primitiveWhen the AUV establishes suction contact with a benthic object, the end-effector creates a metrically grounded, viewpoint-invariant geometric constraintWe introduce a Contact-Aided FactorGraph Localization framework that tightly fuses IMU, DVL, pressure, and magnetometer measurements with adaptive down-looking visual odometry and manipulator contact constraintsThe system addresses three challenges: Planar visual degeneracy by using adaptive uncertainty scaling to mitigate estimator inconsistency under weak monocular geometryDrift without natural loop closures is mitigated by formulating suction contact as a high-confidence geometric factor that induces implicit loop closureA hybrid slidingwindow strategy alternates between landmarkaugmented and pose-only refinement to preserve real-time robustness against intermittent visual structure

Sensor Fusion and Constraint Modeling

The framework formulates underwater state estimation as a contact-aided smoothing problem on a factor graph, implemented using GTSAMVehicle poses, velocities, IMU biases, and environmental landmarks are jointly estimated by fusing proprioceptive navigation measurements with visual observations and manipulator contact constraints within a unified probabilistic frameworkThe inertial sensors provide a tightly coupled backbone where IMU preintegration links consecutive states using the formula (3)The DVL provides a velocity factor by incorporating body-frame velocity corrected for rotational motionPressure measurements are fused as a prior on the vertical position using the temperature-compensated density modelThe visual odometry pipeline employs contrast enhancement and ORB feature extraction, with metric scale recovered from auxiliary sensors like DVL or pressureVisual uncertainty is adaptively scaled using the inlier ratio to prevent visually weak frames from destabilizing the estimator

Contact-Induced Geometric Anchoring

The manipulator-based contact sensing treats the end-effector as a proprioceptive sensor providing contact-based geometric constraints through a suction end-effectorWhen the suction cup engages, the current position of the end-effector pee is calculated and treated as a point measurement in the environmentIf an existing landmark lj lies within a proximity threshold, a bearing–range factor is added between the current pose xk and the landmarkThis mechanism yields an implicit loop closure when the end-effector contacts a previously mapped object without requiring explicit place recognitionThe manipulator thus acts as an active tactile sensor, providing sparse but high-confidence constraints that stabilize the factor graph during prolonged low-visibility operation

Experimental Validation and Results

In the first tank experiments, introducing two 180◦ rotations increases ATE to 0.2917 m and yaw RMSE to 0. p. 4

Where the paper says it

Introducing two 180◦ rotations increases ATE to 0. 2917 m and yaw RMSE to 0.

1232 rad due to weak lateral and heading observabilityAdding a starfish landmark during rotation improves yaw stability (0. p. 4
Where the paper says it

1232 rad, consistent with weak lateral and heading observability in planar, down-looking configurations. Adding a starfish landmark during rotation improves yaw stability (0.

0867 rad) but further increases ATE (0. p. 4
Where the paper says it

Adding a starfish landmark during rotation improves yaw stability (0. 0867 rad) but further increases ATE (0.

3514 m), indicating that a single visual landmark does not resolve global planar degeneracyThe combination of depth variation and landmark constraints yields the best performance with ATE 0. p. 4
Where the paper says it

3514 m), indicating that a single visual landmark does not resolve global planar degeneracy. In contrast, introducing depth variation reduces ATE to 0.

1651 m and yaw RMSE 0. p. 4
Where the paper says it

The combination of depth variation and landmark constraints yields the best performance (ATE 0. 1651 m, yaw RMSE 0.

0412 radIn the harbor deployment, task-level revisit accuracy improves from 1. p. 4
Where the paper says it

RESULTS We evaluate the proposed framework across two controlled tank environments and a real harbor deployment. The experiments are designed to assess (i) estimator behavior under planar degeneracy, (ii) the contribution of landmarks and contact events, and (iii) task-level revisit consistency during repeated benthic interactions.

23 m without landmarks to 0. p. 0
Where the paper says it

The scene is predominantly planar and texture-sparse, rendering monocular visual odometry geometrically ill-conditioned: translation parallel to the seabed is weakly observable, metric scale is ambiguous without reliable altitude cues, and feature tracking is unstable. While visual landmarks can provide constraints, benthic organisms are small, sparsely distributed, and only briefly visible.

90 m when pump activation events are includedThis reduction demonstrates that contact-induced constraints provide structurally informative corrections even under visually degraded conditions

Conclusion and Future Work

We presented a contact-aided factor-graph framework in which physical interaction helps resolve degeneracy in perception-limited underwater SLAMStructured motion and deliberate physical interaction introduce metrically anchored constraints that improve consistency where manipulation occursThis trend also holds in harbor conditions, where turbidity and illumination variability increase global ATE, but relative pose errors remain low and contact events provide repeatable correctionsDrift in perception-limited underwater missions is not only a sensing problem, but an observability problemFuture work will extend this framework toward dense, mission-time mapping by coupling contact-aided localization with high-rate volumetric approaches such as FRAGG-MAP to enable drift-aware reconstruction of benthic workspaces for occupancy and manipulation planning

REFERENCES

[1] M. Ferrera, J. p. 7

Where the paper says it

REFERENCES [1] M. Ferrera, J.

Moras, P. Trouve-Peloux, and V. p. 7
Where the paper says it

Ferrera, J. Trouve-Peloux, and V.

Creuze, “Real-time ´monocular visual odometry for turbid and dynamic underwater environments,” Sensors, vol. p. 7
Where the paper says it

Trouve-Peloux, and V. Creuze, “Real-time ´ monocular visual odometry for turbid and dynamic underwater environments,” Sensors, vol.

19, no. 3, 2019

[2] M. Grimaldi, D. p. 0

Where the paper says it

Monocular range estimates are noisy, and re-identification is unreliable, yielding intermittent constraints that do not ensure persistent global consistency 1School of Engineering & Physical Sciences, Heriot-Watt University, Edinburgh, UK. grimaldi@hw.

Nakath, M. She, and K. Koser, “Investigation of the ¨challenges of underwater-visual-monocular-slam,” ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, vol. p. 7
Where the paper says it

She, and K. Koser, “Investigation of the ¨ challenges of underwater-visual-monocular-slam,” ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences, vol.

X-1/W1-2023

[3] M. Kaess, H. Johannsson, R. p. 7

Where the paper says it

5194/isprs-annals-X-1-W1-2023-1113-2023 [3] M. Johannsson, R.

Roberts, V. p. 7
Where the paper says it

Johannsson, R. Roberts, V.

Ila, J. Leonard, and F. p. 7
Where the paper says it

Roberts, V. Leonard, and F.

Dellaert, “isam2: Incremental smoothing and mapping with fluid relinearization and incremental variable reordering,” in 2011 IEEE International Conference on Robotics and Automation

[4] J. C. Kinsey, R. M. Eustice, and L. p. 7

Where the paper says it

3281–3288. Eustice, and L.

L. Whitcomb, “A survey of underwater vehicle navigation: Recent advances and new challenges,” 2006

[5] X. Wang, X. Fan, P. Shi, J. Ni, and Z. Zhou, “An overview of key slam technologies for underwater scenes,” Remote Sensing, 2023

[6] H. Huang, Z. Liu, L. Zhang, D. Wang, and J. p. 7

Where the paper says it

Zhou, “An overview of key slam technologies for underwater scenes,” Remote Sensing, 2023. Wang, and J.

Wang, “Via-slam: An underwater visual–inertial–acoustic slam with integrated dvl,” IEEE Transactions on Instrumentation and Measurement, 2025

[7] S. Ding, T. Zhang, Y. Li, S. Xu, and M. Lei, “Underwater multi-sensor fusion localization with visual-inertial-depth using hybrid residuals and efficient loop closing,” Measurement, vol. p. 7

Where the paper says it

Wang, “Via-slam: An underwater visual–inertial–acoustic slam with integrated dvl,” IEEE Transactions on Instrumentation and Measurement, 2025. Lei, “Underwater multi-sensor fusion localization with visual-inertial-depth using hybrid residuals and efficient loop closing,” Measurement, vol.

238

[8] S. Xu, K. Zhang, and S. p. 7

Where the paper says it

115245, 2024. Zhang, and S.

Wang, “Aqua-slam: Tightly-coupled underwater acoustic-visual-inertial slam with sensor calibration,” 2025

[9] S. Rahman, A. Q. Li, and I. Rekleitis, “Svin2: An underwater slam system using sonar, visual, inertial, and depth sensor,” in 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2019

[10] S. Archieri, J. p. 7

Where the paper says it

1861–1868. Archieri, J.

Drupt, A. Cinar, M. Grimaldi, I. p. 0
Where the paper says it

Monocular range estimates are noisy, and re-identification is unreliable, yielding intermittent constraints that do not ensure persistent global consistency 1School of Engineering & Physical Sciences, Heriot-Watt University, Edinburgh, UK. grimaldi@hw.

Carlucho, J. p. 7
Where the paper says it

Grimaldi, I. Carlucho, J.

S. Willners, and Y. p. 7
Where the paper says it

Carlucho, J. Willners, and Y.

R. Petillot, “3dssdf: Underwater 3d sonar reconstruction using signed distance functions,” in 2025 IEEE International Conference on Robotics and Automation (ICRA), 2025

[11] G. Wang, X. Liu, Z. Ye, Z. Liu, and P. p. 7

Where the paper says it

5306–5312. Liu, and P.

Huang, “Contact slam: An active tactile exploration policy based on physical reasoning utilized in robotic fine blind manipulation tasks,” 2025

[12] F. Dellaert and G. p. 7

Where the paper says it

Leonard, and F. Dellaert, “isam2: Incremental smoothing and mapping with fluid relinearization and incremental variable reordering,” in 2011 IEEE International Conference on Robotics and Automation, 2011, pp.

Contributors, “borglab/gtsam,” May 2022

[13] F. Dellaert and M. p. 7

Where the paper says it

Leonard, and F. Dellaert, “isam2: Incremental smoothing and mapping with fluid relinearization and incremental variable reordering,” in 2011 IEEE International Conference on Robotics and Automation, 2011, pp.

Kaess, Factor Graphs for Robot Perception, Foundations and Trends in Robotics, Vol. p. 7
Where the paper says it

Kaess, Factor Graphs for Robot Perception. Foundations and Trends in Robotics, Vol.

6, 2017

[14] S. Pizer, R. Johnston, J. p. 7

Where the paper says it

Foundations and Trends in Robotics, Vol. Johnston, J.

Ericksen, B. p. 7
Where the paper says it

Johnston, J. Ericksen, B.

Yankaskas, and K. p. 7
Where the paper says it

Ericksen, B. Yankaskas, and K.

Muller, “Contrast-limited adaptive histogram equalization: speed and effectiveness,” in [1990] Proceedings of the First Conference on Visualization in Biomedical Computing

[15] C. Campos, R. Elvira, J. J. G. Rodriguez, J. p. 7

Where the paper says it

Muller, “Contrast-limited adaptive histogram equalization: speed and effectiveness,” in [1990] Proceedings of the First Conference on Visualization in Biomedical Computing, 1990, pp. Rodriguez, J.

M. M. Montiel, and J. p. 7
Where the paper says it

Rodriguez, J. Montiel, and J.

D. Tardos, “Orb-slam3: An accurate open-source library for visual, visual–inertial, and multimap slam,” IEEE Transactions on Robotics

[16] S. Leutenegger, M. p. 7

Where the paper says it

Tardos, “Orb-slam3: An accurate open-source library for visual, visual–inertial, and multimap slam,” IEEE Transactions on Robotics, 2021. Leutenegger, M.

Chli, and R. p. 7
Where the paper says it

Leutenegger, M. Chli, and R.

Y. Siegwart, “Brisk: Binary robust invariant scalable keypoints,” in 2011 International Conference on Computer Vision

[17] Y. Maeda, H. Kakami, and T. p. 7

Where the paper says it

2548–2555. Kakami, and T.

Aso, “Development of an electric robotic arm for the hadal zone,” in OCEANS 2024 - Halifax, 2024

[18] M. Grimaldi, Y. p. 0

Where the paper says it

Monocular range estimates are noisy, and re-identification is unreliable, yielding intermittent constraints that do not ensure persistent global consistency 1School of Engineering & Physical Sciences, Heriot-Watt University, Edinburgh, UK. grimaldi@hw.

Maeda, H. Kakami, I. Carlucho, Y. p. 7
Where the paper says it

Grimaldi, I. Carlucho, J.

p. 7
Where the paper says it

Grimaldi, I. Carlucho, J.

Petillot, and T. p. 7
Where the paper says it

Willners, and Y. Petillot, “3dssdf: Underwater 3d sonar reconstruction using signed distance functions,” in 2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp.

Inoue, “Towards reliable subsea object recovery: a simulation study of an auv with a suction-actuated end effector,” 2026

[19] M. Grimaldi, P. p. 0

Where the paper says it

Monocular range estimates are noisy, and re-identification is unreliable, yielding intermittent constraints that do not ensure persistent global consistency 1School of Engineering & Physical Sciences, Heriot-Watt University, Edinburgh, UK. grimaldi@hw.

Cieslak, E. p. 7
Where the paper says it

Grimaldi, P. Cieslak, E.

Ochoa, V. Bharti, H. Rajani, I. Carlucho, M. p. 7
Where the paper says it

Grimaldi, I. Carlucho, J.

Koskinopoulou, Y. p. 7
Where the paper says it

Carlucho, M. Koskinopoulou, Y.

R. Petillot, and N. p. 7
Where the paper says it

Willners, and Y. Petillot, “3dssdf: Underwater 3d sonar reconstruction using signed distance functions,” in 2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp.

Gracias, “Stonefish: Supporting machine learning research in marine robotics,” 2025

[20] J. A. Barnes, A. R. Chi, L. S. Cutler, D. J. Healey, D. B. Leeson, T. E., McGunigal, J. p. 7

Where the paper says it

Gracias, “Stonefish: Supporting machine learning research in marine robotics,” 2025. McGunigal, J.

, A., Mullen, W., Smith, R., Sydnor, R., C., Vessot,, and G. p. 7
Where the paper says it

McGunigal, J. Vessot, and G.

-M.: “Characterization of frequency stability,” IEEE Transactions on Instrumentation and Measurement, 1971

[21] J. L. Schonberger, E. p. 7

Where the paper says it

Winkler, “Characterization of frequency stability,” IEEE Transactions on Instrumentation and Measurement, 1971. Schonberger, E.

Zheng, M. Pollefeys, and J. p. 7
Where the paper says it

Schonberger, E. Pollefeys, and J.

-M. Frahm, “Pixel- ¨wise view selection for unstructured multi-view stereo,” in European Conference on Computer Vision (ECCV), 2016

[22] J. L. Schonberger and J. p. 7

Where the paper says it

Winkler, “Characterization of frequency stability,” IEEE Transactions on Instrumentation and Measurement, 1971. Schonberger, E.

-M. Frahm, “Structure-from-motion revisited,” in Conference on Computer Vision and Pattern Recognition (CVPR), 2016

[23] M. Bloesch, S. p. 7

Where the paper says it

Frahm, “Structure-from-motion revisited,” ¨ in Conference on Computer Vision and Pattern Recognition (CVPR), 2016. Bloesch, S.

Omari, M. Hutter, and R. p. 7
Where the paper says it

Bloesch, S. Hutter, and R.

Siegwart, “Robust visual inertial odometry using a direct ekf-based approach,” 2015

[24] K. Deb, Multiobjective Optimization Using Evolutionary Algorithms

[25] M. Grimaldi, N. p. 0

Where the paper says it

Monocular range estimates are noisy, and re-identification is unreliable, yielding intermittent constraints that do not ensure persistent global consistency 1School of Engineering & Physical Sciences, Heriot-Watt University, Edinburgh, UK. grimaldi@hw.

Palomeras, I. p. 7
Where the paper says it

Grimaldi, N. Palomeras, I.

Carlucho, Y. R. Petillot, and P. p. 7
Where the paper says it

Willners, and Y. Petillot, “3dssdf: Underwater 3d sonar reconstruction using signed distance functions,” in 2025 IEEE International Conference on Robotics and Automation (ICRA), 2025, pp.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Contact-Aided Factor-Graph Localization framework allows AUVs to obtain metrically grounded, viewpoint-invariant geometric constraint from suction contact events, which induces implicit loop closures without appearance-based place recognition.

  2. The system can fully initialize online during motion, addressing the challenge of needing stationarity or a dedicated bootstrapping phase for initialization.

  3. Adaptive uncertainty scaling based on inlier statistics prevents visually weak frames from destabilizing the estimator by mitigating inconsistency in planar visual environments where monocular odometry is ill-conditioned.

  4. The hybrid sliding window Local Bundle Adjustment strategy alternates between landmark-augmented and pose-only refinement to preserve real-time robustness when dealing with intermittent visual structure.

  5. Object detections contribute bearing–range factors, enabling the creation of persistent landmarks even in low-texture environments, which are then anchored to contact events for contact-consistent mapping.

Abstract

Accurate state estimation for autonomous underwater vehicles performing close-range seafloor sampling remains challenging. In low-altitude operation, down-looking cameras over featureless planar seabeds produce scale ambiguity, lateral degeneracy, and inconsistent feature tracking. Meanwhile, inertial-Doppler Velocity Log (DVL) fusion alone provides no mechanism for structural drift correction. We propose a Contact-Aided Factor-Graph Localization framework that treats physical interaction as an informative geometric constraint within a smoothing-based localization formulation. The method tightly fuses suction-based manipulator contact events with adaptive visual odometry, learned object detections, and on-board sensors. Visual odometry relative-pose factors and landmark bearing-range factors are uncertainty-scaled according to inlier statistics to prevent visually weak frames from destabilizing the estimator, while contact events are modeled as high-confidence factors that induce implicit loop closures without appearance-based place recognition. Furthermore, the system can fully initialize online during motion. Experimental evaluation in tanks, harbor, and simulation environments demonstrates that contact-induced constraints significantly reduce trajectory drift and improve object revisit accuracy compared to filtering-based navigation and contact-free graph formulations. These results highlight the role of embodied physical interaction as a localization primitive in perception-degraded underwater environments

Related papers