Pith. sign in

REVIEW 4 major objections 5 minor 42 references

Safe and Transparent Robots for Human-in-the-Loop Meat Processing

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a general-purpose robot arm, equipped with a hand-tracking safety stop, a force-sensing knife, a displacement-based failed-cut detector, and an editable plan viewer, can perform meat slicing and trimming tasks alongsi

desk verdict A credible cobot meat-cutting integration with one under-validated load-bearing claim: the displacement-based uncertainty proxy has no successful-cut baseline, so the transparency story rests on thinner evidence than the rest of the paper. read the letter →

arxiv 2508.14763 v1 pith:RXKDIR5K submitted 2025-08-20 cs.RO

Sagar Parekh , Casey Grothoff , Ryan Wright , Robin White , Dylan P. Losey This is my paper · ORCID
classification cs.RO
keywords collaborativerobotsmeatprocessingsafetytransparencyhuman-robotinteractionforcesensinguncertaintydetectionuserstudy
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 tries to establish that an ordinary robot arm, fitted with the right sensing and communication, can do meat-cutting tasks next to human workers without forcing the plant to buy specialized machinery. The central claim is that combining a real-time hand-tracking stop, an instrumented knife that halts on hard contact, a displacement-based detector for failed cuts, and interfaces that show and edit the robot's plan gives both safety and transparency. If the claim holds, small and medium meat processors could adopt flexible collaborative robots that take over repetitive slicing and trimming while a human stays in control. The paper supports the claim with component tests and a demonstration where 20 experts rated the automatic and collaborative modes above the industry-standard emergency-stop and black-box approaches.

What carries the argument

The load-bearing components are the instrumented knife and the uncertainty proxy. The knife mount's force sensor turns hard contact (bone, table, fixture) into an emergency stop by comparing the live reading against a threshold calibrated to the maximum force seen in normal cutting. The uncertainty proxy is the mean Euclidean distance between the four corners of the meat's bounding box before and after a cut, converted to a score by Ψ(d) = (e^(βd) - e^(-βd)) / (e^(βd) + e^(-βd)) — a hyperbolic tangent with a tunable sensitivity β — so that large meat displacement, taken to indicate the knife dragged the meat after hitting a bone, makes the robot signal red and call for inspection. The hand-t

What would settle it

Run the same uncertainty pipeline on a set of clean, successful cuts (no bone present) and record how often the displacement score crosses the red-LED threshold; if a substantial fraction of good cuts trigger 'uncertain', the transparency signal misleads workers. Separately, measure the knife force on different meat products (frozen, bone-in, connective tissue) to see whether the single fixed threshold separates meat from bone in all cases.

Watch

Extended reading notes

Core claim

Working from an industry survey that named safety and transparency as the barriers to adoption, the authors build a single framework that addresses both. Safety comes from a ceiling-mounted camera that tracks hand landmarks and zeroes the robot's velocity when a hand enters the workspace, plus a knife whose mount embeds a force sensor: when the measured force exceeds a threshold set above normal meat cutting, the robot stops. Transparency comes from computing, after each cut, the mean displacement of the meat's bounding-box corners between before and after images, mapping that displacement through a tunable function to a 0-1 uncertainty score; an LED signals normal operation, approach, and f

Load-bearing premise

That the amount the meat moves after a cut reliably tells whether the cut failed: a successful cut leaves the meat almost still, while a failed cut drags it enough to cross the uncertainty threshold; the paper only tests failed cuts, not the false-alarm rate of successful ones.

Editorial extensions

If this is right

  • If the framework transfers to plant conditions, a single general-purpose robot arm can switch between slicing and trimming tasks, replacing several single-purpose machines.
  • Workers no longer need to watch the robot continuously: automatic stopping plus LED alerts covers the main failure modes.
  • The click-and-drag planner puts the final decision about where to cut with the human, which the survey shows experts prefer over purely autonomous plans.
  • Because the automatic stop was preferred 17 to 3 among the 20 experts, proactive sensing may be a market requirement for cobot meat processing.
  • Expert concern about collaborative safety dropped after the demonstration (p = 0.052), suggesting exposure to such a system can shift industry attitudes.

Reading between the lines

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

  • The displacement proxy is only validated on deliberately failed cuts (knife hitting an artificial bone); a natural next test is measuring how often successful cuts move the meat enough to trigger the red LED, which would give the false-alarm rate.
  • The fixed force threshold is calibrated on one knife and one meat setup; per-product calibration (frozen versus thawed, pork versus beef) is an implicit requirement that the paper does not test.
  • The same safety-plus-transparency pattern could extend beyond meat to other deformable-object cutting (fish, fruit, vegetables) where specialized automation is also rare.
  • The equality of expert preference between LED and graphical interfaces hints that either modality alone might suffice; a follow-up could test them independently rather than as a package.
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

4 major / 5 minor

Summary. The paper proposes an integrated safety and transparency framework for collaborative meat processing with a general-purpose robot arm. Safety is addressed by an overhead MediaPipe hand-detection system that stops the robot when a human hand enters the workspace, and by an instrumented knife whose embedded force sensor is intended to distinguish contact with meat from contact with bone or fixtures. Transparency is addressed by a post-cut uncertainty estimate based on displacement of the meat, an LED interface that signals anomalies, and a graphical interface that shows the planned cutting trajectory and allows human edits. The authors evaluate components in controlled experiments and then conduct a demonstration with 20 experts, comparing manual-stop vs. automatic-stop safety and fully autonomous vs. collaborative operation. Survey results show a preference for the proposed framework and a marginal reduction in safety concern after the demonstration.

Significance. If the system performs as claimed, it would be a useful, low-cost step toward flexible cobot deployment in small and medium meat processors. The paper has clear strengths: a concrete system integration, a lightweight hand-detection component with an objectively measured millisecond-level stop latency, instrumented-knife contact tests, a user study with domain experts, and a clear description of the graphical feedback interface. However, the two headline claims—'safe operation' and 'keeping humans in-the-loop'—are currently supported unevenly. The knife-contact stop has a mean latency of 2.095 s, and the uncertainty-detection evaluation is both circular and missing a successful-cut baseline. These gaps weaken the central claims, but they are addressable with additional experiments and a more conservative framing.

major comments (4)
  1. [Section 3.1 (Instrumented Knife), Figure 6] The reported contact-stop latency is 2.095 s (SD 1.62 s), with 75% of trials taking less than 2 s and three trials exceeding 4 s. This is not an 'immediate' stop, and for a robot-mounted knife, a 2-s continued contact with a hard object is not a safety stop in any meaningful sense. The paper's own Discussion attributes the delay to sliding over the bone and suggests threshold tuning, but no improved result is reported. Since safety is the first contribution stated in the abstract, this latency needs to be substantially reduced or the safety claim must be narrowed to 'contact detection and subsequent stop' with the latency reported as a limitation.
  2. [Section 2.2.1, Eq. (1); Section 3.1 (Uncertainty Detection); Table 1] The uncertainty evaluation is circular and incomplete. Psi(d) = tanh(beta*d) is, by construction, a strictly increasing function of the measured displacement d; Table 1 therefore only confirms that the transform increases with its input. The paper never measures the distribution of d for successful cuts, so the false-positive rate of the LED red signal is unknown. The Discussion even acknowledges that 'this movement could simply be the result of the meat deforming after a successful cut.' Moreover, beta and the uncertainty threshold used to trigger the LED are not reported. Without a successful-cut baseline and reported thresholds, the transparency claim is unsupported.
  3. [Section 3.1 (Uncertainty Detection), Table 1] The uncertainty experiments test only deliberately failed cuts in which a 3D-printed bone intersects the trajectory. There is no condition with successful cuts, and there is no control where the knife cuts meat without a bone to measure baseline displacement from knife drag, vibration, or deformation. Consequently, the system's ability to distinguish 'uncertain' from 'certain' cuts is not established. The authors should run a matched set of successful cuts, report the displacement distribution and the ROC-style separation, and then set the LED threshold accordingly.
  4. [Section 2.4.2 (Full System Demonstration), Section 3.2, Figure 7] The user study compares whole conditions that differ in multiple components and also differ in task: Manual Safety and Automatic Safety both slice a pork loin, while Fully Autonomous and Collaborative both trim fat from a pork chop. The statistical tests across all four conditions conflate task type with framework features, and the subjective ratings cannot isolate the contribution of the uncertainty signal. Participants rated the Collaborative condition as a whole, not the reliability of the LED uncertainty cue. A more targeted comparison—e.g., presenting the same cutting task with the uncertainty LED on vs. off, or measuring whether participants can correctly identify successful vs. failed cuts from the LED—is needed to validate the transparency claim.
minor comments (5)
  1. [Section 2.2.1, Eq. (1)] Psi(d) is the hyperbolic tangent of beta*d but is not named as such. More importantly, the numerical value of beta is not given anywhere, nor is the threshold for turning the LED red. These values are necessary for reproducibility.
  2. [Section 3.1 (Uncertainty Detection)] The text says 'Using Equation (2.2.1)' but the equation is numbered (1). Also, Table 1 has no indication of how many trials each row represents; if these are single trials, the reader cannot assess variability.
  3. [Figure 6 and Section 3.1 (Handbook)] Figure 6(c) shows a standard error bar, but Figure 6(e) does not, even though the reported standard deviation is large. A box plot or histogram with quartiles would better communicate the 75th percentile and outliers.
  4. [Section 3.2, Figure 8] The pre/post concern change is reported as p=0.052. This is borderline and should be described as 'not statistically significant' without implying a trend as if it were a significant effect. Effect sizes and confidence intervals would strengthen the reporting.
  5. [Section 2.1.2] The force threshold is said to be 'experimentally determined' but the actual threshold value is not reported, nor is the force range observed for meat vs. bone. Reporting these values would help readers assess the generality of the discrimination.

Circularity Check

1 steps flagged · score 4.0 of 10

Uncertainty-detection validation reduces to the defining equation; the rest of the framework is independently tested.

  1. self definitional [Section 2.2.1 (Eq. 1) and Section 3.1, Table 1]
    "To convert this displacement into the robot’s uncertainty, we define an uncertainty function Ψ(d) that maps the displacement to a normalized confidence score. ... Ψ(d) = (e^{β d} − e^{−β d})/(e^{β d} + e^{−β d}) (1) ... The uncertainty value increases sharply with larger displacement or rotation, indicating to the operator a higher concern over the accuracy of the cut."

    Ψ(d) is defined as tanh(βd), a strictly increasing function of the measured displacement d. The experiments in Section 3.1 record manual displacement/rotation values and then report the corresponding Ψ values in Table 1. The 'steep increase' in uncertainty with displacement is therefore the defining formula evaluated on its own input, not empirical evidence that displacement tracks cutting error. Moreover, all trials use deliberately failed cuts (bone in the trajectory); no successful-cut baseline is measured, so the threshold separating normal deformation from failure is never estimated. The paper itself concedes in the Discussion that 'this movement could simply be the result of the meat deforming after a successful cut,' which underscores that the proxy's validity is assumed, not demons

full rationale

The paper's central claims—safe hand monitoring, force-based contact detection, and user preference for the integrated framework—are supported by independent experiments and a user study. The hand-detection accuracy, precision, and latency are measured against ground-truth human entries; the instrumented knife is tested on deliberate bone contacts; and the user study compares conditions subjectively. The one circular element is the uncertainty-detection component: the uncertainty score is defined as a monotone function of displacement, so Table 1 merely recomputes the definition. Because this component feeds into the transparency claim but does not by itself force the user-study results, the circularity is partial and confined to a supporting module. No load-bearing self-citation or renamed-known-result pattern was found.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The framework depends on several hand-set thresholds and a tunable sensitivity parameter, none of which are reported numerically. The uncertainty proxy and force-threshold discriminator are domain assumptions validated only under narrow conditions, and the user study is the sole evidence for transparency and acceptance.

free parameters (4)
  • beta (uncertainty sensitivity)
    Shape parameter in Psi(d)=tanh(beta*d) in Section 2.2.1; controls how strongly displacement maps to uncertainty. Value not reported, chosen by hand.
  • Knife force threshold
    Set in Section 2.1.2 from the experimentally determined maximum force during normal cuts plus a safety margin. Numerical value not reported.
  • Uncertainty threshold for LED red
    Predefined threshold that triggers the red LED after a cut (Sections 2.2.2 and 3.2). Value not reported.
  • RGB segmentation thresholds
    Predefined RGB color thresholds for meat/fat segmentation used in planning (Section 2.2.3); taken from prior work [22] and not reported here.
assumptions (6)
  • domain assumption MediaPipe hand landmark detection reliably detects human hands from the overhead camera under meat-plant lighting.
    Section 2.1.1 uses Google MediaPipe as the sole human-presence sensor; the paper asserts robustness to lighting but provides no test under varied lighting or occlusion.
  • domain assumption Meat displacement after a cut is a valid proxy for cutting error.
    Section 2.2.1 hypothesizes that a successful cut causes little or no meat movement; only bone-collision failures were tested, never successful cuts.
  • domain assumption A fixed force threshold cleanly separates contact with meat from contact with hard objects.
    Section 2.1.2 sets a threshold from the maximum force observed during normal cutting plus a safety margin; no data distinguish bone, table, or fixture contacts.
  • domain assumption The vision-based planning and calibration from prior work [22] remains valid with the new two-camera setup.
    Section 2.4.1 builds on the previous planning system and calibration without re-deriving or re-validating them with the new mount.
  • domain assumption Kinematic constraints confine the robot to the operational workspace Wop.
    Section 2.1.1 relies on this constraint to define safe/warning zones; no quantitative verification of the constraint is provided.
  • domain assumption Likert ratings and rankings by 20 recruited experts measure the true safety and usefulness of the framework.
    Section 3.2 treats subjective survey responses as primary evidence; no objective behavioral or performance metrics for trust are included.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Safe and Transparent Robots for Human-in-the-Loop Meat Processing." pith.science (2026). https://pith.science/paper/RXKDIR5K

@misc{pith2026250814763,
  author       = {Pith},
  title        = {Pith review of: Safe and Transparent Robots for Human-in-the-Loop Meat Processing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RXKDIR5K}},
  note         = {Machine review of arXiv:2508.14763}
}
read the original abstract

Labor shortages have severely affected the meat processing sector. Automated technology has the potential to support the meat industry, assist workers, and enhance job quality. However, existing automation in meat processing is highly specialized, inflexible, and cost intensive. Instead of forcing manufacturers to buy a separate device for each step of the process, our objective is to develop general-purpose robotic systems that work alongside humans to perform multiple meat processing tasks. Through a recently conducted survey of industry experts, we identified two main challenges associated with integrating these collaborative robots alongside human workers. First, there must be measures to ensure the safety of human coworkers; second, the coworkers need to understand what the robot is doing. This paper addresses both challenges by introducing a safety and transparency framework for general-purpose meat processing robots. For safety, we implement a hand-detection system that continuously monitors nearby humans. This system can halt the robot in situations where the human comes into close proximity of the operating robot. We also develop an instrumented knife equipped with a force sensor that can differentiate contact between objects such as meat, bone, or fixtures. For transparency, we introduce a method that detects the robot's uncertainty about its performance and uses an LED interface to communicate that uncertainty to the human. Additionally, we design a graphical interface that displays the robot's plans and allows the human to provide feedback on the planned cut. Overall, our framework can ensure safe operation while keeping human workers in-the-loop about the robot's actions which we validate through a user study.

Figures

Figures reproduced from arXiv: 2508.14763 by the authors.

Figure 1
Figure 1. Our framework for ensuring safety, maintaining transparency, and integrating human feedback into collaborative robotic systems. The robot works in close proximity to human workers, typically working on processing the same meat product. We propose an instrumented knife and a human hand monitoring system for anticipating in real-time the chances of unwanted interaction between the robot and the human or the environmen… view at source ↗
Figure 2
Figure 2. Safety mechanism of our robot framework. (Left) The vision based hand monitoring system. A camera is mounted above the robot arm to obtain a top-down view of the robot’s workspace Wop and the surrounding environment. The camera constantly monitors the region to detect human hand entering the workspace. (Middle-left) We divide the region into two zones. A safe zone where the robot cannot reach on account of the kinem… view at source ↗
Figure 3
Figure 3. The LED interface used as a communication channel between robot and human. Here, we demonstrate the LED interface communicating the different situations which might be encountered during the robot’s execution of the cuts. (Left) When the human is farther away from the robot, it is in normal operation mode and the LED lights are green. (Middle) When the robot approaches the robot, entering the safe zone, the light tu… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: The graphical interface for maintaining transparency about the robot’s planning procedure. The robot displays the sequence of steps for how it will cut this piece of meat. (Left) A camera is mounted on the robot arm which captures the image of the meat placed on the cu…
Figure 5
Figure 5. Figure 5: Robot arm setup used in our framework. (Left) The setup includes two RGB cameras, one mounted on the robot for planning cutting trajectories and another mounted overhead for hand monitoring, our instrumented knife, an LED interface controlled by a micro-controller, and…
Figure 6
Figure 6. Figure 6: Results of individual component experiments. The first three plots show the results of the experiments evaluating the hand monitoring system; the last three plots show the results for the instrumented knife experiments. (a) shows the detection accuracy of our hand moni…
Figure 7
Figure 7. Figure 7: Results of the survey conducted after each of the four sections. The experts rated these on three measures, namely safety, less need for monitoring, and usefulness. A higher value indicates a more favorable rating. (Left) The plot shows the participants’ ratings for th…
Figure 8
Figure 8. Figure 8 [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages

  1. [1]

    Fao (2024)

    Food & Organization, A. Fao (2024). https://openknowledge.fao.org/handle/20.500.14283/cd5077en

    work page 2024

  2. [2]

    A., Boulos, C

    Taylor, C. A., Boulos, C. & Almond, D. Livestock plants and COVID-19 transmission. Proc. Natl. Acad. Sci. (2020)

    work page 2020

  3. [3]

    & Jones, J

    Bir, C., Peel, D., Holcomb, R., Raper, K. & Jones, J. The impact of covid-19 on meat processing, and the renewed interest in local processing capabilities. In Western Economics F orum(2021)

    work page 2021

  4. [4]

    & Barnard, A

    Victor, K. & Barnard, A. Slaughtering for a living: A hermeneutic phenomenological perspective on the well-being of slaughterhouse employees. Int. journal qualitative studies on health well-being (2016)

    work page 2016

  5. [5]

    & Harrison, W

    Gastón, M. & Harrison, W. Meatpacking workers’ perceptions of working conditions, psychological contracts, and organizational justice. J. Latino/Latin Am. Stud. (2012)

    work page 2012

  6. [6]

    P., Estévez-Moreno, L

    Pastrana-Camacho, A. P., Estévez-Moreno, L. X. & Miranda-de la Lama, G. C. What slaughterhouse workers’ attitudes and knowledge reveal about human-pig relationships during pre-slaughter operations: A profile-based approach. Meat science (2023)

    work page 2023

  7. [7]

    & Alleyne, E

    Slade, J. & Alleyne, E. The psychological impact of slaughterhouse employment: A systematic literature review. Trauma, Violence, & Abus. (2023)

    work page 2023

  8. [8]

    A., Low, T., Long, D., Baillie, C

    Aly, B. A., Low, T., Long, D., Baillie, C. & Brett, P. Robotics and sensing technologies in red meat processing: A review. Trends F ood Sci. & Technol.(2023)

    work page 2023

Show all 42 references
  1. [9]

    Madsen, N., Nielsen, J. U. & Mønsted, J. Automation–the meat factory of the future. In 52nd International Congress of Meat Science and Technology (2006). 14/17

    2006

  2. [10]

    Wakholi, C. et al. Economic analysis of an image-based beef carcass yield estimation system in korea. Animals (2021)

    2021

  3. [11]

    Kim, J. et al. Economic analysis of the use of vcs2000 for pork carcass meat yield grading in korea. Animals (2021)

    2021

  4. [12]

    Manufacturing technology in the danish pig slaughter industry

    Hinrichsen, L. Manufacturing technology in the danish pig slaughter industry. Meat science (2010)

    2010

  5. [13]

    & Umbarkar, A

    Kakade, S., Patle, B. & Umbarkar, A. Applications of collaborative robots in agile manufacturing: a review. Robotic Syst. Appl. (2023)

    2023

  6. [14]

    & Pirayesh, A

    Keshvarparast, A., Battini, D., Battaia, O. & Pirayesh, A. Collaborative robots in manufacturing and assembly systems: literature review and future research agenda. J. Intell. Manuf. (2024)

    2024

  7. [15]

    & Losey, D

    Parekh, S. & Losey, D. P. Learning latent representations to co-adapt to humans. Auton. Robots (2023)

    2023

  8. [16]

    & Galetto, M

    Puttero, S., Verna, E., Genta, G. & Galetto, M. Collaborative robots for quality control: an overview of recent studies and emerging trends. J. Intell. Manuf. (2025)

    2025

  9. [17]

    Romanov, D., Korostynska, O., Lekang, O. I. & Mason, A. Towards human-robot collaboration in meat processing: Challenges and possibilities. J. F ood Eng.(2022)

    2022

  10. [18]

    M., Kathiravan, S., Lokesh, M

    Sahan, A. M., Kathiravan, S., Lokesh, M. & Raffik, R. Role of cobots over industrial robots in industry 5.0: A review. In 2023 2nd International Conference on Advancements in Electrical, Electronics, Communication, Computing and Automation (ICAECA) (2023)

    2023

  11. [19]

    Fournier, É. et al. Human-cobot collaboration’s impact on success, time completion, errors, workload, gestures and acceptability during an assembly task. Appl. Ergonomics (2024)

    2024

  12. [20]

    & Gauchard, G

    Bouillet, K., Lemonnier, S., Clanche, F. & Gauchard, G. Effects of pace on productivity and physical and mental workloads in a human–cobot collaboration. Int. J. Occup. Saf. Ergonomics (2025)

    2025

  13. [21]

    Su, B. et al. Exploring the impact of human-robot interaction on workers’ mental stress in collaborative assembly tasks. Appl. Ergonomics (2024)

    2024

  14. [22]

    & Losey, D

    Wright, R., Parekh, S., White, R. & Losey, D. P. Safely and autonomously cutting meat with a collaborative robot arm. Sci. Reports (2024)

    2024

  15. [23]

    & Marwala, T

    Xing, B. & Marwala, T. Smart maintenance for human–robot interaction. Stud. Syst. Decis. Control. Springer (2018)

    2018

  16. [24]

    & Šabanovi´c, S

    Mutlu, B., Roy, N. & Šabanovi´c, S. Cognitive human–robot interaction. Springer handbook robotics (2016)

    2016

  17. [25]

    Wright, R., Parekh, S., Losey, D. P. & White, R. Surveying processor perceptions of automation in meat processing (2025). Under review

    2025

  18. [26]

    & Croft, E

    Haddadin, S. & Croft, E. Physical human–robot interaction. In Springer handbook of robotics , 1835–1874 (Springer, 2016)

    2016

  19. [27]

    A., Fong, T

    Lasota, P. A., Fong, T. & Shah, J. A. A survey of methods for safe human-robot interaction. F oundations Trends Robotics (2017)

    2017

  20. [28]

    Habibian, S., Alvarez Valdivia, A., Blumenschein, L. H. & Losey, D. P. A survey of communicating robot learning during human-robot interaction. The Int. J. Robotics Res. (2025)

    2025

  21. [29]

    J., Unhelkar, V

    Yang, X. J., Unhelkar, V . V ., Li, K. & Shah, J. A. Evaluating effects of user experience and system transparency on trust in automation. In Proceedings of the 2017 ACM/IEEE international conference on human-robot interaction (2017)

    2017

  22. [30]

    & Wang, Z

    Yao, M., Li, J. & Wang, Z. Robot transparency and employees’ acceptance: The roles of trust and anthropomorphism. Int. J. Human–Computer Interact. (2025)

    2025

  23. [31]

    & Bicchi, A

    De Santis, A., Siciliano, B., De Luca, A. & Bicchi, A. An atlas of physical human–robot interaction. Mech. Mach. Theory (2008)

    2008

  24. [32]

    Lugaresi, C. et al. Mediapipe: A framework for perceiving and processing reality. In Third Workshop on Computer Vision for AR/VR at IEEE Computer Vision and Pattern Recognition (CVPR) 2019 (2019). 15/17

    2019

  25. [33]

    Flexiforce a401 sensor

    Tekscan. Flexiforce a401 sensor. https://www.tekscan.com/products-solutions/force-sensors/flexiforce-a401-sensor

  26. [34]

    Winfield, A. F. et al. Ieee p7001: A proposed standard on transparency. Front. Robotics AI (2021)

    2021

  27. [35]

    UR10 Applications

    UniversalRobots. UR10 Applications. https://www.universal-robots.com/applications/

  28. [36]

    UR10 Compliance

    UniversalRobots. UR10 Compliance. https://www.universal-robots.com/articles/ur/safety/safety-faq/

  29. [37]

    Qiao, S., Wang, Y . & Li, J. Real-time human gesture grading based on openpose. In2017 10th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI) (2017)

    2017

  30. [38]

    A., Neverova, N

    Güler, R. A., Neverova, N. & Kokkinos, I. Densepose: Dense human pose estimation in the wild. In Proceedings of the IEEE conference on computer vision and pattern recognition (2018)

    2018

  31. [39]

    & Farhadi, A

    Redmon, J., Divvala, S., Girshick, R. & Farhadi, A. You only look once: Unified, real-time object detection. InProceedings of the IEEE conference on computer vision and pattern recognition (2016)

    2016

  32. [40]

    Ali, U. et al. Interactive distance field mapping and planning to enable human-robot collaboration. IEEE Robotics Autom. Lett. (2024)

    2024

  33. [41]

    A., Low, T., Long, D., Brett, P

    Aly, B. A., Low, T., Long, D., Brett, P. & Baillie, C. Tactile sensing for tissue discrimination in robotic meat cutting: A feasibility study. J. F ood Eng.(2024)

    2024

  34. [42]

    A., Brett, P., Low, T

    Aly, B. A., Brett, P., Low, T. & Long, D. Fundamental studies on tactile feedback in robotic striploin fat trimming task. J. F ood Eng.(2025). 6 Appendix 6.1 Survey Questions In this section we provide the details of the questions we asked the participants in our demonstration...

    2025

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.