Pith. sign in

REVIEW 4 major objections 6 minor 32 references

Breathless: An 8-hour Performance Contrasting Human and Robot Expressiveness

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Video-captured dance gave a robot arm an 8-hour duet.

desk verdict A real 8-hour human-robot duet, honestly described, but the joint-mapping pseudocode contradicts the prose and needs a fix before the technical details are reliable. read the letter →

arxiv 2411.12361 v2 pith:KXRHNNHR submitted 2024-11-19 cs.RO cs.CV

Catie Cuan , Tianshuang Qiu , Shreya Ganti , Ken Goldberg This is my paper · ORCID
classification cs.ROcs.CV
keywords robochoreographyrobotdancesinusoidalmotionplanninghumanposetrackingOpenUR5ehuman-robotinteractionartandrobotics
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 describes the robotics pipeline behind Breathless, an eight-hour performance in which a human dancer and a UR5e industrial robot arm share a stage for the length of an American workday. The central claim is that this pipeline—deep-learning human pose tracking from video, Fourier-domain smoothing of the resulting joint angles, fitting sinusoidal trajectories per joint, and combining these with teach-mode and force-sensing interactions—is sufficient to produce an expressive, safe, and continuous eight-hour duet. The authors care because the work turns a choreographic idea into a reproducible engineering recipe, and it demonstrates that an industrial manipulator can sustain long-duration artistic interaction with a human.

What carries the argument

The load-bearing mechanism is a staged motion-generation pipeline. OpenPose extracts human joint keypoints from video; a 1D blur convolution of size 15 smooths the angle trajectories; a fast Fourier transform with a hand-set threshold of 20 zeroes out high-frequency bins before inversion; and the resulting per-joint sinusoids are mapped onto the UR5e's six axes, with the unmapped shoulder-pan and wrist-2 joints filled by manually generated sinusoids. The pipeline's second half uses the UR5e's teachMode for compliant, dancer-guided recording at 500 Hz, forceMode with damping 0.2 to smooth the transition out of compliance, and a force-cue detector (movement-average threshold of 20 N) for triggered motifs such as the concluding bow. A URDF-based simulation library is used to verify trajectories for self-collision and sudden jolts before any motion is run on the robot.

What would settle it

Replay a fresh set of human dance videos through the exact OpenPose, FFT-threshold-20, and sinusoid-fitting pipeline on a UR5e and record whether any protective stop triggers, whether the robot enters a self-collision, or whether the resulting motion reads as a different motif; a single such failure would show the pipeline is not sufficient as described.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that human arm motions can be captured as 2D joint angles, reduced by an FFT threshold to a small set of sinusoidal components, and replayed on a six-axis robot to evoke recognizable human-labour gestures. The shoulder, elbow, and wrist angles from OpenPose are mapped to the UR5e's shoulder-lift, elbow, wrist-1, and wrist-3 joints, while the remaining joints are driven by manually generated sinusoids with tuned amplitude, frequency, phase, offset, and an exponential amplitude envelope to create grow-and-fade motions. Live sections are handled by teach-mode 'zero-G' recording, a force-mode transition stage with damping 0.2 to avoid protective stops, and a force-cued waiting state with a 20 N threshold. The paper reports that this combination ran for the full eight-hour premiere without robot replacement or significant debugging, with only a few timeouts requiring resets.

Load-bearing premise

The assumption that carries the whole claim is that the shoulder, elbow, and wrist angles read off a flat video of a dancer, after smoothing and removing high-frequency wiggles, will drive a robot arm's joints in a way that is both safe and recognizable as human movement, even though the remaining joints are just filled with simple back-and-forth motions.

Editorial extensions

If this is right

  • An artist with a UR5e and the released code suite can reproduce the full set of labor motifs without writing a trajectory planner.
  • The pipeline's reliance on sinusoidal primitives means the same captured human motion can be ported to other six-axis arms by adjusting the hand-tuned amplitude and offset parameters.
  • The teach-mode and force-cue interaction pattern gives a safe, non-verbal protocol for a human to improvise with an industrial robot in performance.
  • Eight hours of continuous operation without major failure demonstrates that long-duration installation art is feasible with commodity industrial hardware.

Reading between the lines

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

  • A natural extension is to replace the hand-set FFT threshold and manual sinusoids with an automated model that learns per-joint sinusoidal parameters from a corpus of dance videos, which would make the pipeline less dependent on human tuning.
  • The force-cue and damping transitions could be adapted to collaborative manufacturing, where a worker briefly nudges a robot into a new pose and it resumes its task smoothly.
  • The paper's mapping from 2D image-plane angles to robot joints ignores the third dimension, so motions with strong depth variation would likely need a multi-camera pose estimator to preserve expressiveness.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. This paper describes the technology behind 'Breathless,' an eight-hour dance performance that paired the human dancer Catie Cuan with a Universal Robots UR5e industrial arm at National Sawdust Theater on December 16, 2023. The system combines several elements: an OpenPose-based pipeline that extracts 2D human joint angles, smooths them with a 15-tap convolution, applies FFT-domain filtering with a hand-tuned threshold of 20, and maps the resulting trajectories onto the UR5e joints; a library of manually constructed sinusoidal motifs (stirring, hammering, mopping, and others) with parameters A, omega, phi, gamma tuned in the urdfpy simulator; live interactive sections that use teach-mode recording and 500 Hz replay, a forceMode damping stage to smooth the compliant-to-rigid transition, and a 20 N force-threshold tap detector for cuing; and a Google Sheets, AppleScript, Keynote, and Zoom based stage-cueing system for music and lighting coordination. The paper claims a novel application of AI-based human pose tracking to sinusoidal robot motion planning, reports that the robot ran the full performance with occasional resets and that roughly 600 audience members attended, and offers the project website's video, code, and data as documentation.

Significance. If the system works as described, the paper documents a rare and useful data point: an industrial robot arm sustaining an eight-hour mixed scripted-and-interactive choreography in front of a live audience without hardware failure. The most valuable parts are the practical engineering details, including the damped forceMode transition around teach-mode replay, the moving-average tap detector with a 20 N threshold, and the simulation-first safety workflow, all of which are transferable to other human-robot performance and interaction settings. The stated plan to release video, code, and data is a genuine strength for reproducibility, and the Limitations section is candid about the operational complexity, which required three technical personnel, a stage manager, and a lighting technician. The significance is nonetheless bounded: the technical novelty beyond existing human-to-robot motion transfer and sinusoidal control work is incremental; expressiveness and operational robustness are asserted rather than measured; and the central signal-processing step is specified imprecisely. This is better framed as a systems and performance document than as a generalizable method.

major comments (4)
  1. [Section 3.1] The pseudocode's final joint assignment contradicts the mapping stated in the text. The text says 'H1 is mapped to shoulder lift, H2 to elbow, and H3 to wrist 1, and H4 to wrist 3,' but the pseudocode defines H0 as the absolute chest-shoulder angle, H1 as the relative shoulder angle, H2 as the relative elbow angle, H3 as wrist flexion, and H4 as wrist rotation, and then constructs theta_i = [theta_0(t), H_i0, H_i1, H_i2, theta_4(t), H_i3]. As written, shoulder lift follows the absolute chest-shoulder angle (which includes torso orientation) rather than shoulder flexion, wrist 3 follows wrist flexion rather than wrist rotation, and H4 never enters the joint vector. If the released code follows the pseudocode, the described wrist-rotation capability was never realized; if the code corrects the indexing, the pseudocode misrepresents the implemented system. The manuscript must align text, pseudocode, and code, and state which mapping actually produced the motions shown in the performance.
  2. [Section 3.1] The FFT-filtering step is internally inconsistent and not reproducible as written. The pseudocode computes H' = FFT^-1(f) after thresholding, but the final assignment uses H_i, not H'_i, so the inverse transform has no effect on the output. Moreover, the text describes low-pass filtering ('setting high-frequency bins to 0'), whereas the pseudocode zeroes components with |f_i| < 20, which is magnitude thresholding; the two operations are different, and the value 20 is not given in units that allow replication (a bin index, an amplitude in radians, or a pixel-derived unit). Since the paper's first contribution rests on converting tracked poses into sinusoidal trajectories, this step must be specified exactly and applied consistently.
  3. [Section 4] The Results section does not support the paper's central claims of expressiveness and of an eight-hour successful run. The evidence is one audience quote, a Forbes citation, and the statement that the robot, 'In a few cases, it timed out and needed to be reset,' while it 'largely ran' for eight hours. There are no operational statistics for the performance: no count of resets or protective stops, no downtime duration, no detection rate or latency for the force-tap cue across the full show, and no evaluation of expressiveness by a defined protocol (structured audience responses, movement-expert ratings, or even a documented observation rubric). Because the stated goal is to keep the arm moving 'safely and expressively for a full working day,' the manuscript should either add these basic measurements or explicitly reframe the paper as an artistic demonstration without robustness and expressiveness claims.
  4. [Sections 3.1-3.2] The claimed contribution of applying 'AI-based human pose tracking to robot motion planning using sinusoidal functions' is not exactly what the described pipeline does. The OpenPose branch smooths and FFT-thresholds the extracted joint trajectories and replays them; the paper reports no explicit fit of a small set of sinusoids of the form A*cos(omega*t + phi) + gamma to the human data, and no fit-quality measure (retained harmonics, residual error, variance explained) is given. The manual motifs in Section 3.2 are explicit sinusoids, but their parameter values (A, omega, phi, gamma, and the e^(-Bx) decay used for the bartender sequence) are not reported in a table or appendix, so that half of the pipeline is also not reproducible from the text. The authors should state how the FFT-filtered trajectories relate to the sinusoidal parameterization and provide representative parameter values for the motifs.
minor comments (6)
  1. [Section 3.1] The net-change computation for the right arm reads 'sl = sum_{i=1}^{n-1} = |H_ri - H_l(i+1)|'; it should be 'sr = sum_{i=1}^{n-1} |H_ri - H_r(i+1)|', and the stray equals sign in the summation notation should be removed.
  2. [Section 3.1] The manuscript mixes indexing conventions, listing the human joints as 1 through 6 in the extraction list but using 0 through 5 in the pseudocode; adopting a single convention throughout would prevent the kind of off-by-one confusion that occurs in the joint-mapping step.
  3. [Section 3.1] In the pseudocode, 'cordinates' should be 'coordinates', and the loop header 'j in [0,1,2,3,4]' includes an upper index that the loop body never uses.
  4. [Section 3.3] The sentence 'we add an section between the compliant and rigid sections of the robot' contains a grammar error, and the choice of the 20 N threshold and the 10-sample running-average window deserves a brief justification relative to the sensor-noise level illustrated in Figure 7.
  5. [Figure 7 caption] The phrase 'correctly identify the cue' should read 'correctly identified the cue'; the figure shows a single rehearsal trial, and one sentence on how representative that trace is across rehearsals would strengthen the claim.
  6. [Section 4] The phrase 'without significant injury, only taking two 15 minute breaks' uses 'only' as an editorial judgment, and it is unclear whether the robot continued to run during the dancer's breaks; stating the robot's behavior during those breaks would clarify the 'eight-hour' claim.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the motion-generation pipeline is data-driven fitting and system description, not a derivation that reduces to its inputs.

full rationale

The paper's central technical content is a robot-choreography pipeline: OpenPose extracts human joint angles, the angles are smoothed and low-pass filtered in the Fourier domain, sinusoids are fit or manually generated, and the resulting trajectories are replayed on a UR5e. This is an empirical fitting and engineering description, not a derivation whose output is equivalent to its input by construction. The FFT threshold of 20 and the force-cue threshold of 20 N are manually tuned operating parameters, and the paper does not present either as a predicted quantity; they do not make the central claim circular. The self-citations in the related-work and acknowledgement sections provide historical and artistic context and are not load-bearing for any technical result. The pseudocode discrepancy noted by a skeptical reader (θi = [θ0(t), Hi0, Hi1, Hi2, θ4(t), Hi3] versus the prose mapping H1→shoulder lift, H2→elbow, H3→wrist 1, H4→wrist 3) is a real reproducibility and correctness concern about whether the implemented mapping matches the described mapping, but it is not a circularity: the human-derived angles are inputs to the robot joint trajectories, and no output is reintroduced as an input to define the same quantity. The successful eight-hour performance is an external outcome, and the paper's claims are explicitly about having produced and staged that performance rather than about deriving it from first principles. No circular step, self-definitional reduction, fitted-input-as-prediction, or load-bearing self-citation chain is present.

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

The paper relies on several hand-tuned constants (FFT threshold of 20, force threshold of 20 N, smoothing kernel of 15) and on manually selected sinusoidal parameters per motif. It also assumes that OpenPose keypoint angles are accurate enough, that the human-to-robot joint mapping is meaningful, and that simulation in urdfpy predicts real-robot safety. No conceptual entities (such as forces or mediators) are newly postulated.

free parameters (4)
  • FFT high-frequency threshold = 20
    Set to 20 to generate trajectories that are interesting and within safety constraints (Section 3.1).
  • Force threshold for cue detection = 20 N
    Running average of end-effector force above 20 N triggers the bow motion (Section 3.3).
  • Smoothing kernel size = 15
    1D blur convolution of size 15 to reduce OpenPose noise (Section 3.1).
  • Sinusoid parameters (A, omega, phi, gamma) per motif = manually tuned
    Each movement motif is defined by manually chosen amplitude, frequency, phase, and offset, including exponential amplitude modulation in the bartender sequence (Section 3.2).
assumptions (3)
  • domain assumption OpenPose 2D keypoints accurately reflect human joint positions
    The pipeline relies on OpenPose [30] for keypoint extraction without error analysis (Section 3.1).
  • domain assumption Human arm joint angles computed via atan2 and differencing map meaningfully to UR5e joints
    The human shoulder/elbow/wrist angles are mapped to UR5e shoulder lift, elbow, wrist 1, and wrist 3, while shoulder pan and wrist 2 are filled with manually generated sinusoids (Section 3.1).
  • domain assumption Simulation in urdfpy reliably predicts safety of the physical robot
    All motions are verified in simulation before execution, and safety hazards are fixed in simulation (Section 3.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Breathless: An 8-hour Performance Contrasting Human and Robot Expressiveness." pith.science (2026). https://pith.science/paper/KXRHNNHR

@misc{pith2026241112361,
  author       = {Pith},
  title        = {Pith review of: Breathless: An 8-hour Performance Contrasting Human and Robot Expressiveness},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KXRHNNHR}},
  note         = {Machine review of arXiv:2411.12361}
}
read the original abstract

This paper describes the robot technology behind an original performance that pairs a human dancer (Cuan) with an industrial robot arm for an eight-hour dance that unfolds over the timespan of an American workday. To control the robot arm, we combine a range of sinusoidal motions with varying amplitude, frequency and offset at each joint to evoke human motions common in physical labor such as stirring, digging, and stacking. More motions were developed using deep learning techniques for video-based human-pose tracking and extraction. We combine these pre-recorded motions with improvised robot motions created live by putting the robot into teach-mode and triggering force sensing from the robot joints onstage. All motions are combined with commercial and original music using a custom suite of python software with AppleScript, Keynote, and Zoom to facilitate on-stage communication with the dancer. The resulting performance contrasts the expressivity of the human body with the precision of robot machinery. Video, code and data are available on the project website: https://sites.google.com/playing.studio/breathless

Figures

Figures reproduced from arXiv: 2411.12361 by the authors.

Figure 1
Figure 1. Catie Cuan with UR5e industrial robot arm at the National Sawdust Theater, Dec 16, 2023. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Dancer Catie Cuan with the output of the OpenPose software, indicating automatically [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Three snapshots from phase two, where the human dancer returns to the lab to dance with [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Visualization of the UR5e Arm. 6 joints on the UR5e utilized by the urdfpy simulation, a Python library for loading, manipulating, and exporting URDF files and robot specifications. Figure 4b showcases the UR5e with shoulder lift joint at an angle of -π/4 radians and e…
Figure 5
Figure 5. Figure 5: Methodology Behind Stirring Action Development Example of pipeline for setting sinusoidal parameter values for robot motions that suggest the robot arm is stirring a pot of soup. The process of developing a motif is illustrated in [PITH_FULL_IMAGE:figures/full_fig_p00…
Figure 6
Figure 6. Figure 6: Live performance photo, National Sawdust Theater, 2023. [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: The raw signal vs the running average detected on the end effector. The robot did not react [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Live performance photo, National Sawdust Theater, 2023. [PITH_FULL_IMAGE:figures/full_fig_p011_8.png]
Figure 9
Figure 9. Figure 9: The interview after the performance featuring Elena Park, Catie Cuan, and Ken Goldberg [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

32 extracted references · 32 canonical work pages

  1. [1]

    Tinguely, M´eta-matic no

    J. Tinguely, M´eta-matic no. 10, 1959. [Online]. Available: https://www.youtube. com/watch?v=CIEYlUosKUY

    work page 1959

  2. [2]

    Robot k-456,

    N. J. Paik, “Robot k-456,” Performance Art, 1964

    work page 1964

  3. [3]

    S. R. Labs, What is real? 1998. [Online]. Available: https://www.srl.org/shows/ web98/

    work page 1998

  4. [4]

    Stelarc, Augmented and amplified body: Event for extended arm & sensor bracelet, 2023

    work page 2023

  5. [5]

    Robot choreography: Moving in a new direction,

    M. K. Apostolos, “Robot choreography: Moving in a new direction,”Leonardo, vol. 23, no. 1, pp. 25–29, 1990

    work page 1990

  6. [6]

    Black flags

    W. Forsythe. “Black flags.” (2014), [Online]. Available: https://www.williamforsythe. com/installations.html

    work page 2014

  7. [7]

    Forsythe, Choreographic objects, https://www.williamforsythe.com/essay

    W. Forsythe, Choreographic objects, https://www.williamforsythe.com/essay. html, 2017. 14 Catie Cuan, Tianshuang Qiu, Shreya Ganti, Ken Goldberg

    work page 2017

  8. [8]

    Babyface,

    K. Ladenheim and A. LaViers, “Babyface,” inProceedings of the 7th International Conference on Movement and Computing, ser. MOCO ’20, Jersey City/Virtual, NJ, USA: Association for Computing Machinery, 2020, I S B N: 9781450375054. D O I: 10.1145/3401956.3404253. [Online]. Available: https://doi.org/10.1145/ 3401956.3404253

    work page arXiv 2020

Show all 32 references
  1. [9]

    Style-based robotic motion in contem- porary dance performance,

    A. LaViers, L. Teague, and M. Egerstedt, “Style-based robotic motion in contem- porary dance performance,” Controls and Art, pp. 205–229, 2014

    2014

  2. [10]

    Choreographic and somatic approaches for the development of expressive robotic systems,

    A. LaViers, C. Cuan, C. Maguire, et al., “Choreographic and somatic approaches for the development of expressive robotic systems,” inArts, MDPI, vol. 7, 2018, p. 11

    2018

  3. [11]

    Do you love me?

    B. Dynamics. “Do you love me?” Boston Dynamics. (2020), [Online]. Available: https://www.youtube.com/watch?v=fn3KWM1kuAw

    2020

  4. [12]

    Choreography moves reference,

    B. Dynamics. “Choreography moves reference,” Boston Dynamics. (2021), [On- line]. Available: https://dev.bostondynamics.com/docs/concepts/choreography/ move reference

    2021

  5. [13]

    Uptown spot,

    B. Dynamics. “Uptown spot,” Boston Dynamics. (2018), [Online]. Available: https://www.youtube.com/watch?v=kHBcVlqpvZ8

    2018

  6. [14]

    Modulation of simple sinusoidal patterns by a coupled oscillator model for biped walking,

    J. Morimoto, G. Endo, J. Nakanishi, et al., “Modulation of simple sinusoidal patterns by a coupled oscillator model for biped walking,” in Proceedings 2006 IEEE International Conference on Robotics and Automation, 2006. ICRA 2006., 2006, pp. 1579–1584. D O I: 10.1109/ROBOT.200...

    2006 arXiv

  7. [15]

    A biologically inspired biped locomotion strategy for humanoid robots: Modulation of sinusoidal patterns by a coupled oscillator model,

    J. Morimoto, G. Endo, J. Nakanishi, and G. Cheng, “A biologically inspired biped locomotion strategy for humanoid robots: Modulation of sinusoidal patterns by a coupled oscillator model,” IEEE Transactions on Robotics, vol. 24, no. 1, pp. 185–191, 2008. D O I: 10.1109/TRO.2008.915457

    2008

  8. [16]

    Nonholonomic motion planning: Steering using sinusoids,

    R. M. Murray and S. S. Sastry, “Nonholonomic motion planning: Steering using sinusoids,” IEEE T. Automatic Control, vol. 38, no. 5, 1993

    1993

  9. [17]

    Robotics, Ur5e robot arm, https://www.universal-robots.com/products/ur5- robot/, 2008

    U. Robotics, Ur5e robot arm, https://www.universal-robots.com/products/ur5- robot/, 2008

    2008

  10. [18]

    Interactive jamming with shimon: A social robotic musician,

    G. Weinberg, A. Raman, and T. Mallikarjuna, “Interactive jamming with shimon: A social robotic musician,” in ACM/IEEE International Conference on Human- Robot Interaction, 2009, pp. 233–234

    2009

  11. [19]

    Time to compile: An interactive art installa- tion,

    C. Cuan, I. Pakrasi, and A. LaViers, “Time to compile: An interactive art installa- tion,” in Biennial Symposium on Arts & Technology, vol. 51, 2018, p. 19

    2018

  12. [20]

    Live dance performance investigating the feminine cyborg metaphor with a motion-activatedwearable robot,

    K. Ladenheim, R. McNish, W. Rizvi, and A. LaViers, “Live dance performance investigating the feminine cyborg metaphor with a motion-activatedwearable robot,” in ACM/IEEE International Conference on Human-Robot Interaction , 2020, pp. 243–251

    2020

  13. [21]

    Beyonc ´e’s ‘renaissance’ tour has freaking robots,

    J. Blistein. “Beyonc ´e’s ‘renaissance’ tour has freaking robots,” Rolling Stone. (2023), [Online]. Available: https://www.rollingstone.com/music/music-news/ beyonce-perform-cozy-robots-renaissance-tour-opener-1234733103/

    2023

  14. [22]

    All together now

    C. Hennick. “All together now.” (2021), [Online]. Available: https : / / www. bostondynamics.com/resources/blog/all-together-now

    2021

  15. [23]

    When i’m gone,

    K. Perry. “When i’m gone,” Katy Perry. (2022), [Online]. Available: https://www. youtube.com/watch?v=N-4YMlihRf4. Breathless: An 8-hour Human-Robot Dance Performance 15

    2022

  16. [24]

    Designing robots with movement in mind,

    G. Hoffman and W. Ju, “Designing robots with movement in mind,”Journal of Human-Robot Interaction, vol. 3, no. 1, pp. 91–122, 2014

    2014

  17. [25]

    Curtain and time to compile: A demonstration of an experimental testbed for human-robot interaction,

    C. Cuan, I. Pakrasi, E. Berl, and A. LaViers, “Curtain and time to compile: A demonstration of an experimental testbed for human-robot interaction,” in IEEE International Symposium on Robot and Human Interactive Communication, 2018, pp. 255–261

    2018

  18. [26]

    Human-robot artistic co-creation: A study in improvised robot dance,

    O. Th ¨orn, P. Knudsen, and A. Saffiotti, “Human-robot artistic co-creation: A study in improvised robot dance,” in 2020 29th IEEE International Conference on Robot and Human Interactive Communication (RO-MAN), IEEE, Aug. 2020, pp. 845–850

    2020

  19. [27]

    Towards symbiotic creativity: A methodological approach to compare human and ai robotic dance creations,

    A. De Filippo, L. Giuliani, E. Mancini, A. Borghesi, P. Mello, and M. Milano, “Towards symbiotic creativity: A methodological approach to compare human and ai robotic dance creations,” in IJCAI, Aug. 2023, pp. 5806–5814

    2023

  20. [28]

    Dance teaching by a robot: Combining cognitive and physical human–robot inter- action for supporting the skill learning process,

    D. F. P. Granados, B. A. Yamamoto, H. Kamide, J. Kinugawa, and K. Kosuge, “Dance teaching by a robot: Combining cognitive and physical human–robot inter- action for supporting the skill learning process,” IEEE Robotics and Automation Letters, vol. 2, no. 3, pp. 1452–1459, 2017

    2017

  21. [29]

    Music mode: Transforming robot movement into music increases likability and perceived intelligence,

    C. Cuan, E. Fisher, A. Okamura, and T. Engbersen, “Music mode: Transforming robot movement into music increases likability and perceived intelligence,” arXiv preprint arXiv:2306.02632, 2023

    2023 arXiv

  22. [30]

    Realtime multi-person 2d pose estimation using part affinity fields,

    Z. Cao, T. Simon, S. E. Wei, and Y . Sheikh, “Realtime multi-person 2d pose estimation using part affinity fields,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, IEEE, 2017, pp. 7291–7299

    2017

  23. [31]

    [Online]

    OpenRR, Urdfpy, Github repository. [Online]. Available: https://github.com/ openrr/urdfpy

  24. [32]

    Teaching machines to be human, and humans to live with machines,

    B. Wolff. “Teaching machines to be human, and humans to live with machines,” Forbes Magazine. (2024), [Online]. Available: https://www.forbes.com/sites/ benjaminwolff/2024/03/24/teaching-machines-to-be-human-and-humans-to- live-with-machines/

    2024

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.