Feeling the Force: A Nuanced Physics-based Traversability Sensor for Navigation in Unstructured Vegetation

Johann Laconte; Roland Lenain; Romuald Aufrere; Zaar Khizar

REVIEW 5 major objections 4 minor 20 references

Feeling the Force: A Nuanced Physics-based Traversability Sensor for Navigation in Unstructured Vegetation

T0 review · 5 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A robot can measure how hard vegetation pushes back by reading the displacement of a tensioned wire and converting that single measurement into a force via a static equilibrium model.

desk verdict A genuinely new wire-based force sensor with clean derivations and insufficient experimental validation; worth reviewing, not yet citable as quantitative. read the letter →

arxiv 2507.19831 v1 pith:AW5PJMCS submitted 2025-07-26 cs.RO cs.SYeess.SY

Zaar Khizar , Johann Laconte , Roland Lenain , Romuald Aufrere This is my paper · ORCID

classification cs.ROcs.SYeess.SY

keywords forcesensingtraversabilityvegetationnavigationdeformableobstacleswiredisplacementsensorconstant-tensionspringfieldmappingoutdoorrobotics

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

This paper claims that a small wire sensor can tell a robot how hard vegetation is pushing back by measuring only how much wire has been pulled out. The wire is held under a constant tension $T$ by an internal spring; when grass, a sapling, or a shrub deflects the wire, the measured elongation $l$ is converted into a force using a static equilibrium model. For a single contact point the model gives a closed-form force that grows with elongation and saturates at $2T$; for homogeneous vegetation the wire becomes a circular arc and the force is $T$ times its curvature. Field trials on a mobile robot report force-distance profiles for a sapling, dense grass, and a shrub, with readings reaching the sensor's 4.4 N limit. The authors frame this as a direct, physics-based traversability metric that complements vision and learned methods, and they list ground-truth validation, speed effects, and automatic model selection as future work.

What carries the argument

The carrying object is a ratiometric potentiometer whose wire is held at a nominally constant tension $T$ by an internal coiled-metal-strip spring; the only raw measurement is the extra wire length $l$ pulled out during contact. The machinery is the static equilibrium model that converts $l$ into force: Eq. (7) for a point load from the triangular deformation profile, and $F_v = T\kappa$ for homogeneous vegetation with curvature obtained numerically from Eq. (15). This yields a closed-form or one-parameter numerical inversion per reading, which the paper says is lightweight enough for real-time use, with a maximum measurable force of $2T = 4.4\,\mathrm{N}$.

What would settle it

Mount the sensor on a fixed rig, apply calibrated weights or a force gauge to the wire at known points $x_0$, and compare the forces predicted by Eq. (7) with the applied loads across the full displacement range; a systematic mismatch, or a measured change in spring pull-back tension with extension, would show that the reported forces are biased.

Watch

Extended reading notes

Core claim

The central claim is that vegetation resistance can be recovered from a single wire displacement under two idealized loading regimes. For a concentrated contact at $x_0$, minimizing the potential energy of the tensioned wire yields $F_s = T\left(\frac{y_0}{\sqrt{x_0^2+y_0^2}} + \frac{y_0}{\sqrt{(L-x_0)^2+y_0^2}}\right)$, with $y_0$ fixed by the constraint that the deformed length is $L+l$, and this force saturates at $2T$ for large elongation. For homogeneous loading the Euler-Lagrange equation gives $F_v = T\kappa$, so the wire forms a circular arc whose curvature comes from $L = \frac{2}{\kappa}\sin\left(\frac{\kappa(L+l)}{2}\right)$. Field trials show a sapling force rising to about $3\,\mathrm{N}$ with a yield phase, dense grass peaking at $4.3\,\mathrm{N}$ with spatial variation, and a shrub reaching about $4\,\mathrm{N}$ under either model. The paper argues that these direct readings capture compliance that vision and learning methods only infer.

Load-bearing premise

The load-bearing premise is that the sensor's internal spring holds a constant tension $T$ while the wire extends, and that the wire is massless and frictionless, so the static-equilibrium formulas in Section III represent what the robot feels while moving at $1\,\mathrm{m/s}$.

Editorial extensions

If this is right

A robot can treat vegetation as a quantifiable force rather than a binary obstacle, so a planner can allow safe collisions with compliant plants while still avoiding rigid hazards.
Force readings can be accumulated into a spatial force field map, as the paper does with 0.15 m cells, giving a denser representation than an occupancy grid.
The saturation at $2T$ sets a clear operating envelope: the sensor is suited to grass, saplings, and small shrubs up to about 4.4 N, while larger obstacles fall outside its range.
Since the inversion is one nonlinear solve per sample, the sensor output can feed navigation or learning loops at or beyond the demonstrated 10 Hz sampling.
The two idealized models cover point contacts and uniform vegetation; an adaptive switch between them would require external sensing to locate contacts.

Reading between the lines

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

A direct lab calibration against a force gauge, pulling the wire with known weights at known contact points, would settle whether the constant-tension spring assumption holds across the full displacement range.
The sensor's 4.4 N ceiling suggests an obvious extension: an array of wires with different tensions could widen the measurable force range and localize contacts along the robot's front.
Because both models can produce similar peaks for mixed vegetation, fusing the sensor with a camera or lidar segmenter could let a planner choose the correct inversion automatically rather than fixing a model in advance.
Dynamic effects at the 1 m/s field speed are not separately characterized; instrumenting the rig with a load cell would reveal whether static equilibrium is an adequate approximation.

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

This one should be on your radar: a wire-based traversability sensor that converts measured wire deflection into a contact force using two clean analytical models — a point-load triangle for discrete stems and a circular-arc for homogeneous vegetation. The derivations in Section III are internally consistent, the force limits 2T and 2T/L follow from the models, and I don't see this exact hardware-model pairing in the cited prior art. The authors make a fair case that bumpers and pushbars lack the sensitivity needed for light vegetation.

The experimental side is the weak part, and the authors know it. They report peak forces (sapling ~3 N, grass ~4.3 N) with no repeated trials, no error bars, no ground-truth force sensor, and no calibration of the "constant tension" spring. All forces scale linearly with T, so if the spring's tension drifts over extension (or internal friction matters), every reported number shifts. That is an unverified load-bearing assumption, not a fatal one.

There is also a real units ambiguity in the homogeneous case. Eq. (12) gives Fv = T·κ, which has units of force per length (N/m), but the text and figures present values as N without saying whether L = 0.44 m was multiplied in. The peak 4.3 N either means ~1.9 N total if it's per-length, or ~9.8 N/m per-length if it's total — those are very different statements, and the max-force bound 2T/L only makes sense for the per-length quantity. The right fix is to state explicitly what is being plotted and give Fv·L for the total force.

What the paper does well: the models are parameter-free except for T and L, there is no fitting to data (no circularity), and the force field map in Experiment 2 is a sensible way to project sparse measurements into a planning grid. The limitations section is honest about future ground-truth validation and the need for adaptive model selection.

Who is this for: people working on traversability in vegetation, and robotics folks interested in cheap physical sensing complements to vision. It is not a paradigm shift, but it is a plausible new tool. A serious referee could push the authors to calibrate the spring, run repeats, resolve the units, and release data — all achievable revisions. I'd send it to review rather than desk-reject, and I'd want to see the revised version before citing it.

Referee Report

5 major / 4 minor

Summary. The paper proposes a wire-based compliant sensor for estimating interaction forces between a mobile robot and vegetation. The sensor measures the additional wire pulled out, l, under an assumed constant tension T. Section III derives a point-force model (Eqs. 7-9) and a homogeneous-loading model (Eqs. 12-16), giving force limits 2T and 2T/L. Section IV reports three field experiments (sapling, grass, shrub) and constructs a force-field map. The analytic machinery is mostly internally consistent, but the quantitative claims are undermined by missing calibration, a unit ambiguity in the homogeneous case, and the absence of ground-truth validation.

Significance. If the central claims were fully validated, the contribution would be a lightweight, physics-based traversability sensor that complements exteroceptive learning-based methods, with a parameter-free force estimate (given T and x0) and clear saturation limits. The analytic derivations in Eqs. (7), (8), (12), and (15) are internally consistent, and the force limits 2T and 2T/L follow correctly from the models. The method is computationally light, requiring only one scalar equation per reading, which is a genuine practical advantage. However, the absence of calibration and ground-truth comparison, plus the unresolved N versus N/m ambiguity, means the paper does not currently demonstrate that it 'directly measures' vegetation resistance.

major comments (5)

[Section IV.B and IV.C, Eq. (12)] Equation (12) defines Fv as a force per unit length (N/m), and Eq. (16) gives the bound Fv <= 2T/L = 10 N/m for L = 0.44 m. The grass experiment nevertheless reports a peak of 4.3 N, and Section IV.C compares the homogeneous-model output with the point-force model as 'around 4 N'. If the plotted values are Fv in N/m they are not forces; if the plotted values are L*Fv in N, the multiplication by L = 0.44 m is not documented anywhere. This ambiguity must be resolved before the quantitative results can be interpreted.
[Section III and IV] The entire force estimate is proportional to T (Eqs. 7 and 12), and T is stated as 2.2 N, but no calibration of the spring's force-extension characteristic, internal friction, or spool dynamics is reported. The text asserts a 'constant pullback force' without presenting calibration evidence. Because every reported force scales with T, an unvalidated T biases all results; a spring calibration curve and an independent reference-force check (e.g., known weights or a load cell) are needed to substantiate the sensor's direct force measurement claim.
[Section IV] The experimental evaluation consists of one sapling, one grass patch, and one shrub traversal at a single speed with no repeated trials, no error bars, and no ground-truth force measurement. The conclusion even lists 'systematically validating results against ground truth' as future work. The qualitative force-profile interpretations (elastic phase, yield point, density variations) are therefore not quantitatively supported, and the claimed maximum measurable force of 4.4 N is not verified.
[Section IV.B] The force-field map is constructed by uniformly distributing each single integrated wire-force reading along the 0.44 m wire and accumulating it into grid cells, but this uniform-distribution assumption is not validated and is known to be false in general because Eq. (7) shows the point-force response depends on contact position x0. At minimum, a sensitivity analysis or an experiment with a known localized load is required before the map can be interpreted as a spatial force field.
[Section III and IV] The model is a static-equilibrium derivation, yet the robot travels at 1 m/s and data are sampled at 10 Hz. No evidence is given that wire inertia, vegetation impact dynamics, or spool dynamics are negligible at these speeds; without a quasi-static-validity check, the measured l may not correspond to the static equilibrium force assumed in Eq. (7) or (12).

minor comments (4)

[Section III.A, after Eq. (9)] The text says 'the maximum force the sensor in able to measure'; 'in' should be 'is'.
[Section IV intro] The phrase 'an real-time Kinematic Global Navigation Satellite System' should be 'a real-time kinematic Global Navigation Satellite System'.
[Section IV.B] The grid cell description '0.15 mcells' should read '0.15 m cells'.
[Figure 6] The colorbar is labeled in N, but if the homogeneous results are per-length values, the units in the figure must be updated to N/m or the total-force conversion must be documented.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: force estimates follow from a parameter-free static-equilibrium model of the wire sensor, with no fitted parameters or self-citation bearing on the derivation.

full rationale

The paper's derivation chain converts measured wire pull-out l into an inferred force using a variational static-equilibrium model: Eq. (7) for a concentrated load and Eqs. (12)+(15) for homogeneous loading. The tension T is declared as a hardware constant, not fitted to the experimental force profiles. The maximum force 2T is obtained as an analytical limit of Eq. (7), and the homogeneous maximum 2T/L follows from the circular-segment geometry of Eq. (15), neither being an empirical fit. No experimental data are used to define the model parameters, so the reported force profiles are not predictions forced by a fitted input. The only self-citation, Laconte et al. [5], appears in related work on Lambda Fields and is not used to justify the sensor model or the uniqueness of any choice. Unvalidated assumptions, such as constant spring tension, absence of friction, quasi-static operation at 1 m/s, and the unit interpretation of Fv (N/m vs N), are legitimate correctness and validation concerns about whether the sensor truly measures force, but they do not make the derivation circular. Therefore no specific circular step can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central derivation rests on the constant-tension spring model, a static idealization of the wire, and two idealized interaction patterns (point and homogeneous). No new physical entities are postulated. The force field mapping adds a uniform-distribution assumption that is convenient but unvalidated.

free parameters (2)

T (constant wire tension) = 2.2 N (stated hardware value)
Stated as a known property of the internal constant-tension spring; all force estimates scale linearly with T, but no calibration certificate or independent measurement is provided. Any error in T propagates directly into every reported force.
x0 (contact point position) = L/2 in Experiment 1
The sapling force is computed assuming the point force acts at the wire midpoint 'for simplicity'; this assumption is not verified and the paper acknowledges future work on detecting the actual contact point. Off-center contacts would change the computed force via Eq. (7).

assumptions (4)

domain assumption The internal spring of the wire sensor provides a constant tension T independent of extension.
Invoked in Section III to define Us = (l+L)T and to treat T as constant in Eqs. (7) and (12). Real spring-loaded potentiometers can exhibit tension variation and friction.
domain assumption The wire is massless, frictionless, and in static equilibrium during interaction.
The derivation minimizes a static potential energy and ignores dynamics, damping, and wire mass. The robot moves at 1 m/s, so inertial effects may be non-negligible.
domain assumption Vegetation interaction can be represented by either a single point force or a constant distributed force across the whole wire.
Equations (4) and the homogeneous case in Section III-B are idealizations; real vegetation often has graded, non-uniform compliance. The paper acknowledges this in Experiment 3.
ad hoc to paper In the force-field map, the single integrated force reading is uniformly distributed along the wire and assigned to grid cells.
Section IV-B uses this to build the 0.15 m force field map, but the spatial distribution of force along the wire is not measured; this directly affects the map's validity.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Feeling the Force: A Nuanced Physics-based Traversability Sensor for Navigation in Unstructured Vegetation." pith.science (2026). https://pith.science/paper/AW5PJMCS

@misc{pith2026250719831,
  author       = {Pith},
  title        = {Pith review of: Feeling the Force: A Nuanced Physics-based Traversability Sensor for Navigation in Unstructured Vegetation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AW5PJMCS}},
  note         = {Machine review of arXiv:2507.19831}
}

read the original abstract

In many applications, robots are increasingly deployed in unstructured and natural environments where they encounter various types of vegetation. Vegetation presents unique challenges as a traversable obstacle, where the mechanical properties of the plants can influence whether a robot can safely collide with and overcome the obstacle. A more nuanced approach is required to assess the safety and traversability of these obstacles, as collisions can sometimes be safe and necessary for navigating through dense or unavoidable vegetation. This paper introduces a novel sensor designed to directly measure the applied forces exerted by vegetation on a robot: by directly capturing the push-back forces, our sensor provides a detailed understanding of the interactions between the robot and its surroundings. We demonstrate the sensor's effectiveness through experimental validations, showcasing its ability to measure subtle force variations. This force-based approach provides a quantifiable metric that can inform navigation decisions and serve as a foundation for developing future learning algorithms.

Figures

Figures reproduced from arXiv: 2507.19831 by the authors.

**Figure 2.** Our custom force sensor mounted on the robot, with [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** Schematic of the sensor model. The diagram [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: Force profiles and physical representation of a single [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

**Figure 6.** Figure 6: Force field map of the traversed grass patch. The [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]

**Figure 5.** Figure 5: Force profile and physical representation of a dense [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

**Figure 7.** Figure 7: Comparative force profiles and physical representation of the shrub obstacle. Left: Force–distance curves for the homogeneous model (Equation 12, orange) and the unique-displacement model (Equation 7, red). Right: Reference image of the shrub structure. The proposed wi…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 18 canonical work pages

[1]

Fast traversability estimation for wild visual navigation

J Frey et al. “Fast traversability estimation for wild visual navigation”. In: Robotics: Science and Systems Proceedings 19 (2023)

work page 2023
[2]

Vegetation detection for mobile robot navigation

D Bradley et al. “Vegetation detection for mobile robot navigation”. In: Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, Tech. Rep. CMU-RI-TR-04-12 (2004)

work page 2004
[3]

Learning object deformation models for robot motion planning

Barbara Frank et al. “Learning object deformation models for robot motion planning”. In: Robotics and Autonomous Systems 62.8 (2014), pp. 1153–1174

work page 2014
[4]

Path planning for autonomous mobile robots: A review

Jose Ricardo Sanchez-Ibanez, Carlos J Perez-del-Pulgar, and Alfonso Garcia-Cerezo. “Path planning for autonomous mobile robots: A review”. In: Sensors 21.23 (2021), p. 7898

work page 2021
[5]

A novel occupancy mapping framework for risk-aware path planning in unstructured environments

Johann Laconte et al. “A novel occupancy mapping framework for risk-aware path planning in unstructured environments”. In: Sensors 21.22 (2021), p. 7562

work page 2021
[6]

Step: Stochastic traversability evaluation and planning for risk-aware off-road navigation

David D Fan et al. “Step: Stochastic traversability evaluation and planning for risk-aware off-road navigation”. In: arXiv preprint arXiv:2103.02828 (2021)

arXiv 2021
[7]

Complex terrain navigation via model error prediction

Adam Polevoy et al. “Complex terrain navigation via model error prediction”. In: International Conference on Robotics and Automation (ICRA) . IEEE. 2022, pp. 9411–9417

work page 2022
[8]

Risk-aware off-road navigation via a learned speed distribution map

Xiaoyi Cai et al. “Risk-aware off-road navigation via a learned speed distribution map”. In: 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE. 2022, pp. 2931–2937

work page 2022

Show all 20 references

[9]

Using lidar intensity for robot navigation

Adarsh Jagan Sathyamoorthy et al. “Using lidar intensity for robot navigation”. In: arXiv preprint arXiv:2309.07014 (2023)

2023 arXiv
[10]

Wayfast: Navigation with predictive traversability in the field

Mateus V Gasparino et al. “Wayfast: Navigation with predictive traversability in the field”. In: IEEE Robotics and Automation Letters 7.4 (2022), pp. 10651–10658

2022
[11]

Wayfaster: a self-supervised traversability prediction for increased navigation awareness

Mateus V Gasparino, Arun N Sivakumar, and Girish Chowdhary. “Wayfaster: a self-supervised traversability prediction for increased navigation awareness”. In: IEEE International Conference on Robotics and Automation (ICRA) . IEEE. 2024, pp. 8486–8492

2024
[12]

AMCO: Adaptive Multimodal Coupling of Vision and Proprioception for Quadruped Robot Navigation in Outdoor Environments

Mohamed Elnoor et al. “AMCO: Adaptive Multimodal Coupling of Vision and Proprioception for Quadruped Robot Navigation in Outdoor Environments”. In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . 2024, pp. 7687–7694

2024
[13]

VERN: Vegetation-Aware Robot Navigation in Dense Unstructured Outdoor Environments

Adarsh Jagan Sathyamoorthy et al. “VERN: Vegetation-Aware Robot Navigation in Dense Unstructured Outdoor Environments”. In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 2023, pp. 11233–11240

2023
[14]

TerraPN: Unstructured Terrain Navigation using Online Self-Supervised Learning

Adarsh Jagan Sathyamoorthy et al. “TerraPN: Unstructured Terrain Navigation using Online Self-Supervised Learning”. In: IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . 2022, pp. 7197–7204

2022
[15]

Learned ultra-wideband RADAR sensor model for augmented LIDAR-based traversability mapping in vegetated environments

Juhana Ahtiainen et al. “Learned ultra-wideband RADAR sensor model for augmented LIDAR-based traversability mapping in vegetated environments”. In: 18th International Conference on Information Fusion (Fusion) . IEEE. 2015, pp. 953–960

2015
[16]

Traversability analysis with vision and terrain probing for safe legged robot navigation

Garen Haddeler et al. “Traversability analysis with vision and terrain probing for safe legged robot navigation”. In: Frontiers in Robotics and AI 9 (2022), p. 887910

2022
[17]

Modeling and traversal of pliable materials for tracked robot navigation

Camilo Ordonez et al. “Modeling and traversal of pliable materials for tracked robot navigation”. In: Unmanned Systems Technology XX. V ol. 10640. SPIE. 2018, pp. 76–83

2018
[18]

Characterization and traversal of pliable vegetation for robot navigation

Camilo Ordonez et al. “Characterization and traversal of pliable vegetation for robot navigation”. In: International Symposium on Experimental Robotics . Springer. 2020, pp. 293–304

2020
[19]

Sensitive Device for Probing and Recognition of Obstacles in a Natural Environment

Lama Al Bassit et al. “Sensitive Device for Probing and Recognition of Obstacles in a Natural Environment”. In: Agritech Day . 2023, pp. 66–75

2023
[20]

Measurement and Prediction of Override F orce of Clumps of Small V egetation in Off-Road Autonomous Navigation

Chris Goodin et al. Measurement and Prediction of Override F orce of Clumps of Small V egetation in Off-Road Autonomous Navigation . Tech. rep. SAE Technical Paper, 2024

2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Fast traversability estimation for wild visual navigation

J Frey et al. “Fast traversability estimation for wild visual navigation”. In: Robotics: Science and Systems Proceedings 19 (2023)

work page 2023

[2] [2]

Vegetation detection for mobile robot navigation

D Bradley et al. “Vegetation detection for mobile robot navigation”. In: Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, Tech. Rep. CMU-RI-TR-04-12 (2004)

work page 2004

[3] [3]

Learning object deformation models for robot motion planning

Barbara Frank et al. “Learning object deformation models for robot motion planning”. In: Robotics and Autonomous Systems 62.8 (2014), pp. 1153–1174

work page 2014

[4] [4]

Path planning for autonomous mobile robots: A review

Jose Ricardo Sanchez-Ibanez, Carlos J Perez-del-Pulgar, and Alfonso Garcia-Cerezo. “Path planning for autonomous mobile robots: A review”. In: Sensors 21.23 (2021), p. 7898

work page 2021

[5] [5]

A novel occupancy mapping framework for risk-aware path planning in unstructured environments

Johann Laconte et al. “A novel occupancy mapping framework for risk-aware path planning in unstructured environments”. In: Sensors 21.22 (2021), p. 7562

work page 2021

[6] [6]

Step: Stochastic traversability evaluation and planning for risk-aware off-road navigation

David D Fan et al. “Step: Stochastic traversability evaluation and planning for risk-aware off-road navigation”. In: arXiv preprint arXiv:2103.02828 (2021)

arXiv 2021

[7] [7]

Complex terrain navigation via model error prediction

Adam Polevoy et al. “Complex terrain navigation via model error prediction”. In: International Conference on Robotics and Automation (ICRA) . IEEE. 2022, pp. 9411–9417

work page 2022

[8] [8]

Risk-aware off-road navigation via a learned speed distribution map

Xiaoyi Cai et al. “Risk-aware off-road navigation via a learned speed distribution map”. In: 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE. 2022, pp. 2931–2937

work page 2022