pith. sign in

arxiv: 2606.17394 · v1 · pith:IHFCGFVWnew · submitted 2026-06-16 · 💻 cs.RO · cs.LG

Damage Adaptation in Seconds for Architected Materials

James Avtges , Jake Ketchum , Helena Young , Taekyoung Kim , Ryan Truby , Todd Murphey This is my paper

Pith reviewed 2026-06-27 01:24 UTC · model grok-4.3

classification 💻 cs.RO cs.LG
keywords damage adaptationsoft roboticsproprioceptionarchitected materialsensemble methodreal-time adaptationHSA actuatorsLEAP
0
0 comments X

The pith

Latent damage representations combined with an ensemble method enable real-time adaptation to unseen damage in soft robots.

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

The paper establishes that architected materials in soft-actuated systems allow damage to be captured in latent representations, which with a simple ensemble method suffice to adapt to catastrophic unseen damage in under one minute. This works because actuator failure is gradual and damage occupies a low-dimensional discrete coordinate space. The authors also identify conditions where the sample complexity of learning these representations drops from exponential to linear, unlike rigid or continuum mechanisms. The LEAP algorithm demonstrates this on a 6DoF soft wrist with Handed Shearing Auxetic actuators, achieving simulation-free adaptation to cuts, burns, and repairs during a tracing task.

Core claim

Latent damage representations plus a simple yet robust ensemble method is sufficient for adapting to unseen damage in real-time. Moreover, we identify conditions under which exponential sample complexity collapses to linear sample complexity for learned representations of architected materials, a concrete advantage over rigid components or continuum soft mechanisms.

What carries the argument

Latent damage representations in a low-dimensional discrete coordinate space, paired with an ensemble method for adaptive proprioception.

If this is right

  • Enables simulation-free real-time adaptation to cuts, burns, and actuator repairs.
  • Sample complexity for damage learning reduces from exponential to linear under the identified conditions.
  • Supports long-term autonomy for soft robots performing tasks like tracing with a 6DoF wrist.
  • Provides a concrete advantage for architected materials over rigid or continuum soft systems.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach may extend to other gradual-damage scenarios in robotics where full modeling is impractical.
  • It could reduce reliance on pre-simulated damage models in field deployments.
  • Further tests on different actuator geometries would clarify the scope of the linear-complexity regime.

Load-bearing premise

Damage can be described in a low-dimensional, discrete coordinate space.

What would settle it

An experiment showing that adaptation to a new damage type requires more than linear samples or fails to reach real-time performance would disprove the claim.

Figures

Figures reproduced from arXiv: 2606.17394 by Helena Young, Jake Ketchum, James Avtges, Ryan Truby, Taekyoung Kim, Todd Murphey.

Figure 1
Figure 1. Figure 1: Latent Ensemble Adaptive Proprioception (LEAP): Modeling damage in architected materials enables simple yet robust ensemble methods to adapt to damage in seconds. We evaluate our approach on a soft wrist that completes a contour tracing task using proprioception alone. The wrist is damaged by melting actuators (left), cutting with scissors (right), and repairing them with glue. After adaptation, LEAP recov… view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of LEAP: Our proposed approach encodes damage states into a latent representation and uses the latent structure to produce ensembles— where online data collection and regression of ensemble weightings occurs within seconds. Our architecture contains two neural networks—the first produces latent encodings of the actuator damage state. The second uses the latent variables alongside the measured … view at source ↗
Figure 3
Figure 3. Figure 3: HSA-Actuated Soft Wrist: A. We demonstrate LEAP on a miniature, parallel 6-DoF soft wrist. Servomotors drive HSAs, and a proprioceptive camera determines platform pose via an AprilTag. B. The soft wrist mounted to a Franka Emika Panda robot arm. We demonstrate the wrist’s functionality via a tracing task where the objective is to traverse the contour of a 3D-printed surface while remaining in contact as mu… view at source ↗
Figure 4
Figure 4. Figure 4: LEAP Enables Robust, Rapid Damage Adaptation. A.) HSA Damage as Discrete Coordinates: The periodic structure of architected materials introduces degeneracies: in an HSA, severed links can be either on a vertical strut (A, B, C) highlighted in the undamaged HSA, or on the horizontal links connecting the major struts. Torching HSAs occurs in one of three locations along each strut. B.) Repair in Architected … view at source ↗
Figure 6
Figure 6. Figure 6: Tracing Task Demonstration: Representative trajectories from the proprioceptive tracing task. Without adaptation, the damaged wrist is not task capable, it fails to follow the contour and cannot recover. After adapting with LEAP, the system recovers up to 100% of its original performance. of the total trajectory. After adapting to burning one HSA and cutting a second (see the provided multimedia), the wris… view at source ↗
Figure 5
Figure 5. Figure 5: Wrist Adaptation Performance: Our whole system model fit to data from increasingly damaged wrist configurations. The confusion matrix diagonal shows that the system is able to adapt to a variety of system damage states using less than a minute of total data collection and computation time. The off-diagonal elements show a clear trend whereby increasing damage accumulation degrades model performance in the … view at source ↗
Figure 7
Figure 7. Figure 7: Normalized Error Before and After Adaptation: Pictured is a representative sample of force-torque errors over time—closer to zero is better. Data is plotted for two HSAs that each undergo multiple damages sequentially. apriltag is tag25h9:0 at size 0.0078m and is adhered to the underside of the upper platform. During manual testing the system is controlled with a 6DOF SpaceMouse Compact from 3DConnexion. T… view at source ↗
Figure 8
Figure 8. Figure 8: Wrist Block Diagram: The wrist is controlled and interfaced to LEAP via ROS2, associated nodes and hardware are shown. TABLE V ALL HSAS USED IN THIS WORK, AND EACH OF THEIR DAMAGE CONDITIONS. ‘VC’ INDICATES A VERTICAL CUT, ‘HC’ INDICATES A HORIZONTAL CUT, AND ‘B’ INDICATES A BURN. HIGHLIGHTED ROWS UNDERWENT REPAIRS. HSA Damage 1 Damage 2 Damage 3 1 VC A1 HC B3 VC B4 2 VC C3 - - 3 VC A4 - - 4 B C0 VC B3 VC … view at source ↗
Figure 9
Figure 9. Figure 9: All HSAs used in this work. Table V shows how each HSA was damaged [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
read the original abstract

Adaptation to damages and in-situ physical repairs is essential for long-term robot autonomy, yet challenging outside of narrowly defined and well-anticipated bounds. In this work we proprioceptively adapt to catastrophic damage in soft-actuated systems in under one minute. Architected materials are well equipped for adaptation: actuator failure occurs gradually rather than acutely, and damage can be described in a low-dimensional, discrete coordinate space. Surprisingly, latent damage representations plus a simple yet robust ensemble method is sufficient for adapting to unseen damage in real-time. Moreover, we identify conditions under which exponential sample complexity collapses to linear sample complexity for learned representations of architected materials, a concrete advantage over rigid components or continuum soft mechanisms. We demonstrate LEAP, our method for adaptive proprioception, via a tracing task for a 6DoF soft wrist based on Handed Shearing Auxetic (HSA) actuators. Our algorithm is able to adapt to cuts, burns, and actuator repairs, enabling simulation-free real-time adaptation that is critical for realizing the promise of soft robots outside the lab. Videos and more information are available at https://murpheylab.github.io/leap.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper presents LEAP, a proprioceptive adaptation method for soft-actuated architected materials that uses latent damage representations combined with a simple ensemble approach to achieve real-time adaptation to unseen damage (cuts, burns, actuator repairs) in under one minute. It demonstrates the approach on a 6DoF Handed Shearing Auxetic (HSA) wrist in a tracing task and claims that damage in such systems lives in a low-dimensional discrete coordinate space, enabling a collapse from exponential to linear sample complexity.

Significance. If the central claims hold with supporting derivations and validation, the work would be significant for soft robotics by offering a practical, simulation-free route to damage resilience outside controlled settings. The sample-complexity reduction result would constitute a concrete theoretical advantage over rigid or continuum systems if rigorously derived and tested.

major comments (2)
  1. [Abstract] Abstract: the central claim that 'latent damage representations plus a simple yet robust ensemble method is sufficient for adapting to unseen damage in real-time' rests on the unvalidated premise that damage occupies a low-dimensional discrete coordinate space; no independent check is supplied that the ensemble exhaustively covers this space or that the exponential-to-linear collapse survives when discreteness is relaxed.
  2. [Abstract] Abstract: the abstract asserts adaptation to cuts, burns, and repairs but supplies no derivation details, error bars, quantitative performance metrics, or data-exclusion rules, rendering the sufficiency claim for real-time adaptation unverifiable from the provided text.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments on our manuscript. We address each major comment point-by-point below, proposing targeted revisions to improve clarity and verifiability while preserving the core contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that 'latent damage representations plus a simple yet robust ensemble method is sufficient for adapting to unseen damage in real-time' rests on the unvalidated premise that damage occupies a low-dimensional discrete coordinate space; no independent check is supplied that the ensemble exhaustively covers this space or that the exponential-to-linear collapse survives when discreteness is relaxed.

    Authors: The manuscript motivates the low-dimensional discrete coordinate space from the discrete nature of actuator failures in architected materials and derives conditions under which sample complexity reduces from exponential to linear. Experimental results on the 6DoF HSA wrist demonstrate adaptation to multiple unseen damage instances using the ensemble, providing empirical support for coverage of the relevant space. We agree that an explicit independent validation of exhaustive coverage and robustness under relaxed (continuous) discreteness assumptions would strengthen the theoretical claim. We will revise the abstract to qualify the claim with reference to the derived conditions and add a dedicated paragraph in the discussion section addressing the assumptions and potential extensions to continuous damage models. revision: yes

  2. Referee: [Abstract] Abstract: the abstract asserts adaptation to cuts, burns, and repairs but supplies no derivation details, error bars, quantitative performance metrics, or data-exclusion rules, rendering the sufficiency claim for real-time adaptation unverifiable from the provided text.

    Authors: The abstract is intentionally concise and high-level, as is standard. The full manuscript contains the requested elements: quantitative metrics (adaptation times under one minute, success rates), error bars from repeated trials, derivation of the sample-complexity result, and details on experimental protocols including damage types and data handling. We will revise the abstract to incorporate key quantitative highlights (e.g., adaptation time and performance metrics) within length constraints and ensure all supporting details remain clearly signposted in the main text and supplementary material. revision: partial

Circularity Check

0 steps flagged

No circularity detected; claims rest on explicit stated assumption

full rationale

The abstract states the low-dimensional discrete damage coordinate space as a premise for architected materials and then shows that an ensemble on latent representations suffices for real-time adaptation under that premise. No equations are given, no fitted parameters are renamed as predictions, and no self-citations are invoked to justify uniqueness or an ansatz. The sample-complexity collapse is presented as following from the discrete framing rather than being smuggled in by definition. The derivation is therefore self-contained on the provided text; external validation would require the full methods section, but nothing in the visible chain reduces to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; all claims rest on unstated modeling assumptions about damage representation and ensemble robustness.

pith-pipeline@v0.9.1-grok · 5743 in / 1088 out tokens · 20754 ms · 2026-06-27T01:24:52.381610+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 28 canonical work pages

  1. [1]

    Gazzola, L

    M. Gazzola, L. H. Dudte, A. G. McCormick, and L. Ma- hadevan. Forward and inverse problems in the mechanics of soft filaments.Royal Society Open Science, page 171628, 2018. URL https://royalsocietypublishing.org/ doi/10.1098/rsos.171628

    work page doi:10.1098/rsos.171628 2018

  2. [2]

    EunBi Oh, Taekyoung Kim, Pranav Kaarthik, and Ryan L. Truby. Architected Materials for Soft Robotics. Journal of Materials Research, 2026. URL https://doi. org/10.1557/s43578-025-01778-2

    work page doi:10.1557/s43578-025-01778-2 2026

  3. [3]

    The case for an omni-bodied robot brain, 2025

    The Skild AI Team. The case for an omni-bodied robot brain, 2025. URL https://www.skild.ai/blogs/ omni-bodied

    2025

  4. [4]

    Kowalewski, Jiani Guo, Jacob F

    Molly Carton, Jakub F. Kowalewski, Jiani Guo, Jacob F. Alpert, Aman Garg, Daniel Revier, and Jeffrey Ian Lip- ton. Bridging hard and soft: Mechanical metamaterials enable rigid torque transmission in soft robots.Science Robotics, 2025. URL https://www.science.org/doi/full/ 10.1126/scirobotics.ads0548

    work page doi:10.1126/scirobotics.ads0548 2025

  5. [5]

    Z. Roth, B. Mooring, and B. Ravani. An overview of robot calibration.IEEE Journal on Robotics and Automa- tion, 1987. URL https://ieeexplore.ieee.org/document/ 1087124/

    1987

  6. [6]

    SpotCheck (joint and camera calibra- tion), 2022

    Boston Dynamics. SpotCheck (joint and camera calibra- tion), 2022. URL https://support.bostondynamics.com/s/ article/SpotCheck-Joint-and-Camera-Calibration-49935

    2022

  7. [7]

    MUKCa: Accurate and affordable cobot calibration without external measurement devices, 2025

    Giovanni Franzese, Max Spahn, Jens Kober, and Cosimo Della Santina. MUKCa: Accurate and affordable cobot calibration without external measurement devices, 2025. URL https://arxiv.org/abs/2503.12584

    arXiv 2025

  8. [8]

    Data-driven industrial robot arm calibration: A machine learning perspective

    Zhibin Li, Shuai Li, and Xin Luo. Data-driven industrial robot arm calibration: A machine learning perspective. In2021 IEEE International Conference on Networking, Sensing and Control (ICNSC), volume 1, pages 1–6,

  9. [9]

    doi: 10.1109/ICNSC52481.2021.9702246

    work page doi:10.1109/icnsc52481.2021.9702246 2021

  10. [10]

    Handedness in shearing auxetics creates rigid and compliant struc- tures.Science, 2018

    Jeffrey Ian Lipton, Robert MacCurdy, et al. Handedness in shearing auxetics creates rigid and compliant struc- tures.Science, 2018. URL https://www.science.org/doi/ abs/10.1126/science.aar4586

    work page doi:10.1126/science.aar4586 2018

  11. [11]

    Spatially pro- grammable architected materials inspired by the met- allurgical phase engineering.Advanced Materials,

    Chen Liu and Minh-Son Pham. Spatially pro- grammable architected materials inspired by the met- allurgical phase engineering.Advanced Materials,

  12. [12]

    URL https://advanced.onlinelibrary.wiley.com/doi/ abs/10.1002/adma.202305846

    work page doi:10.1002/adma.202305846

  13. [13]

    Dias, and Douglas P

    Yi Yang, Marcelo A. Dias, and Douglas P. Holmes. Multistable kirigami for tunable architected materials. Phys. Rev. Mater., Nov 2018. URL https://link.aps.org/ doi/10.1103/PhysRevMaterials.2.110601

    work page doi:10.1103/physrevmaterials.2.110601 2018

  14. [14]

    A flexible, ar- chitected soft robotic actuator for motorized extensional motion.Advanced Intelligent Systems, 2024

    Taekyoung Kim, Pranav Kaarthik, et al. A flexible, ar- chitected soft robotic actuator for motorized extensional motion.Advanced Intelligent Systems, 2024. URL https:// onlinelibrary.wiley.com/doi/abs/10.1002/aisy.202300866

    work page doi:10.1002/aisy.202300866 2024

  15. [15]

    Rubinstein, and Katia Bertoldi

    Ahmad Rafsanjani, Yuerou Zhang, Bangyuan Liu, Shmuel M. Rubinstein, and Katia Bertoldi. Kirigami skins make a simple soft actuator crawl.Science Robotics, 2018. URL https://www.science.org/doi/abs/ 10.1126/scirobotics.aar7555

    work page doi:10.1126/scirobotics.aar7555 2018

  16. [17]

    Automated gait generation for walking, soft robotic quadrupeds.IEEE International Conference on Intelligent Robots and Sys- tems (IROS), pages 10245–10251, 2023

    Jake Ketchum, Sophia Schiffer, et al. Automated gait generation for walking, soft robotic quadrupeds.IEEE International Conference on Intelligent Robots and Sys- tems (IROS), pages 10245–10251, 2023. doi: 10.1109/ IROS55552.2023.10342059

    arXiv 2023

  17. [18]

    Taekyoung Kim, Alexander Torres-Soto, Haodong Wang, and Ryan L. Truby. Architected elas- tomers as load bearing actuators for untethered soft robot walking.Advanced Materials Technologies,

  18. [19]

    URL https://advanced.onlinelibrary.wiley.com/doi/ abs/10.1002/admt.202500503

    work page doi:10.1002/admt.202500503

  19. [20]

    Force and speed in a soft stewart platform

    Jake Ketchum, James Avtges, et al. Force and speed in a soft stewart platform. In2025 IEEE 8th International Conference on Soft Robotics (RoboSoft), 2025. doi: 10. 1109/RoboSoft63089.2025.11020875

    arXiv 2025

  20. [21]

    Yuen, Jeffrey Lipton, Luis H

    Lillian Chin, Michelle C. Yuen, Jeffrey Lipton, Luis H. Trueba, Rebecca Kramer-Bottiglio, and Daniela Rus. A Simple Electric Soft Robotic Gripper with High- Deformation Haptic Feedback. InInternational Confer- ence on Robotics and Automation (ICRA), 2019. URL https://ieeexplore.ieee.org/abstract/document/8794098

    arXiv 2019

  21. [22]

    Dunn, and Ryan L

    Taekyoung Kim, Eliot A. Dunn, and Ryan L. Truby. Clutchable soft actuators produce rapid, high-power movements in robotic artificial musculoskeletal systems. IEEE Robotics and Automation Letters, 2026. doi: 10.1109/LRA.2026.3655266

    work page doi:10.1109/lra.2026.3655266 2026

  22. [23]

    Burden, and Jef- frey Ian Lipton

    Joseph Sullivan, Ian Good, Samuel A. Burden, and Jef- frey Ian Lipton. Spring-brake! handed shearing auxetics improve efficiency of hopping and standing, 2025. URL https://arxiv.org/abs/2505.22898

    arXiv 2025

  23. [24]

    Truby, and Todd D

    James Avtges, Jake Ketchum, Millicent Schlafly, Helena Young, Taekyoung Kim, Allison Pinosky, Ryan L. Truby, and Todd D. Murphey. Real-Time Reinforcement Learn- ing for Dynamic Tasks with a Parallel Soft Robot, 2025. URL http://arxiv.org/abs/2509.19525

    Pith/arXiv arXiv 2025

  24. [25]

    Sample- efficient online control policy learning with real-time recursive model updates

    Zixin Zhang, James Avtges, and Todd Murphey. Sample- efficient online control policy learning with real-time recursive model updates. InProceedings of The 9th Conference on Robot Learning, Proceedings of Machine Learning Research. PMLR, 27–30 Sep 2025. URL https://proceedings.mlr.press/v305/zhang25d.html

    2025

  25. [26]

    Tuthill and Eiman Azim

    John C. Tuthill and Eiman Azim. Proprioception.Cur- rent Biology, 2018. URL https://www.sciencedirect.com/ science/article/pii/S0960982218300976

    2018

  26. [27]

    Toward perceptive soft robots: Progress and challenges.Advanced Science, 2018

    Hongbo Wang, Massimo Totaro, and Lucia Beccai. Toward perceptive soft robots: Progress and challenges.Advanced Science, 2018. URL https://advanced.onlinelibrary.wiley.com/doi/abs/10. 1002/advs.201800541

    2018

  27. [28]

    Yuen, Rebecca Kramer-Bottiglio, and Jamie Paik

    Michelle C. Yuen, Rebecca Kramer-Bottiglio, and Jamie Paik. Strain sensor-embedded soft pneumatic actuators for extension and bending feedback. In2018 IEEE International Conference on Soft Robotics (RoboSoft),

  28. [29]

    doi: 10.1109/ROBOSOFT.2018.8404920

    work page doi:10.1109/robosoft.2018.8404920 2018

  29. [30]

    Truby, Lillian Chin, et al

    Ryan L. Truby, Lillian Chin, et al. Fluidic innervation sensorizes structures from a single build material.Sci- ence Advances, 2022. doi: 10.1126/sciadv.abq4385

    work page doi:10.1126/sciadv.abq4385 2022

  30. [31]

    Johnson, Mantas Naris, Maxwell P

    Vani Sundaram, Khoi Ly, Brian K. Johnson, Mantas Naris, Maxwell P. Anderson, James Sean Humbert, Niko- laus Correll, and Mark Rentschler. Embedded magnetic sensing for feedback control of soft hasel actuators.IEEE Transactions on Robotics, pages 808–822, 2023. doi: 10.1109/TRO.2022.3200164

    work page doi:10.1109/tro.2022.3200164 2023

  31. [32]

    Somatosensory actuator based on stretchable conductive photothermally responsive hydrogel.Science Robotics, 2021

    Yusen Zhao, Chiao-Yueh Lo, Lecheng Ruan, Chen-Huan Pi, Cheolgyu Kim, Yousif Alsaid, Imri Frenkel, Rossana Rico, Tsu-Chin Tsao, and Ximin He. Somatosensory actuator based on stretchable conductive photothermally responsive hydrogel.Science Robotics, 2021. doi: 10. 1126/scirobotics.abd5483. URL https://www.science.org/ doi/abs/10.1126/scirobotics.abd5483

    work page doi:10.1126/scirobotics.abd5483 2021

  32. [33]

    7643–7650

    Annan Zhang, Ryan L. Truby, Lillian Chin, Shuguang Li, and Daniela Rus. Vision-based sensing for electrically- driven soft actuators.IEEE Robotics and Automation Letters, 7(4):11509–11516, 2022. doi: 10.1109/LRA. 2022.3201604

    work page doi:10.1109/lra 2022

  33. [34]

    Yifan Zhu, Mei Hao, Xupeng Zhu, Quentin Bateux, Alex Wong, and Aaron M. Dollar. Forces for free: Vision- based contact force estimation with a compliant hand. Science Robotics, page eadq5046, 2025. URL https:// www.science.org/doi/full/10.1126/scirobotics.adq5046

    work page doi:10.1126/scirobotics.adq5046 2025

  34. [35]

    Durability of soft pneumatic actuators: A review and benchmarking protocol.Ad- vanced Robotics Research, 2026

    Dickson Chiu Yu Wong, Yucheng Luo, and Rob Bernar- dus Nicolaas Scharff. Durability of soft pneumatic actuators: A review and benchmarking protocol.Ad- vanced Robotics Research, 2026. URL https://advanced. onlinelibrary.wiley.com/doi/abs/10.1002/adrr.202500172

    work page doi:10.1002/adrr.202500172 2026

  35. [36]

    Robiul Islam

    Abhirup Sarker, Tamzid Ul Islam, and Md. Robiul Islam. A review on recent trends of bioinspired soft robotics: Actuators, control methods, materials selection, sensors, challenges, and future prospects.Advanced Intelligent Systems, 2025. URL https://advanced.onlinelibrary.wiley. com/doi/abs/10.1002/aisy.202400414

    work page doi:10.1002/aisy.202400414 2025

  36. [37]

    Soft robot percep- tion using embedded soft sensors and recurrent neural networks.Science Robotics, 2019

    Thomas George Thuruthel, Benjamin Shih, Cecilia Laschi, and Michael Thomas Tolley. Soft robot percep- tion using embedded soft sensors and recurrent neural networks.Science Robotics, 2019. URL https://www. science.org/doi/abs/10.1126/scirobotics.aav1488

    work page doi:10.1126/scirobotics.aav1488 2019

  37. [38]

    Shep- herd

    Hedan Bai, Young Seong Kim, and Robert F. Shep- herd. Autonomous self-healing optical sensors for dam- age intelligent soft-bodied systems.Science Advances,

  38. [39]

    URL https://www.science.org/doi/10.1126/sciadv. abq2104

    work page doi:10.1126/sciadv

  39. [40]

    Hal Daumé III, John Langford, and Daniel Marcu

    Antoine Cully, Jeff Clune, Danesh Tarapore, and Jean- Baptiste Mouret. Robots that can adapt like animals. Nature, 2015. URL https://doi.org/10.1038/nature14422

    work page doi:10.1038/nature14422 2015

  40. [41]

    Reset-free trial-and-error learning for robot damage recovery.Robotics and Autonomous Systems, 100:236–250, 2018

    Konstantinos Chatzilygeroudis, Vassilis Vassiliades, and Jean-Baptiste Mouret. Reset-free trial-and-error learning for robot damage recovery.Robotics and Autonomous Systems, 100:236–250, 2018

    2018

  41. [42]

    Shah, Michael Levin, Rebecca Kramer-Bottiglio, and Josh Bon- gard

    Sam Kriegman, Stephanie Walker, Dylan S. Shah, Michael Levin, Rebecca Kramer-Bottiglio, and Josh Bon- gard. Automated shapeshifting for function recovery in damaged robots. InRobotics: Science and Systems XV. Robotics: Science and Systems Foundation, 2019. URL http://dx.doi.org/10.15607/RSS.2019.XV .028

    work page doi:10.15607/rss.2019.xv 2019

  42. [43]

    Bongard and H

    J.C. Bongard and H. Lipson. Automated damage di- agnosis and recovery for remote robotics. InIEEE International Conference on Robotics and Automation (ICRA), volume 4, pages 3545–3550, 2004. doi: 10.1109/ ROBOT.2004.1308802

    arXiv 2004

  43. [44]

    Resilient machines through continuous self-modeling.Science,

    Josh Bongard, Victor Zykov, and Hod Lipson. Resilient machines through continuous self-modeling.Science,

  44. [45]

    URL https://www.science.org/doi/abs/10.1126/ science.1133687

  45. [46]

    URL https://www

    Robert Kwiatkowski and Hod Lipson. Task-agnostic self- modeling machines.Science Robotics, 2019. URL https: //www.science.org/doi/abs/10.1126/scirobotics.aau9354

    work page doi:10.1126/scirobotics.aau9354 2019

  46. [47]

    Self-supervised learning of state estimation for manipulating deformable linear objects.IEEE Robotics and Automation Letters, 5(2):2372–2379, 2020

    Mengyuan Yan, Yilin Zhu, Ning Jin, and Jeannette Bohg. Self-supervised learning of state estimation for manipulating deformable linear objects.IEEE Robotics and Automation Letters, 5(2):2372–2379, 2020. doi: 10.1109/LRA.2020.2969931

    work page doi:10.1109/lra.2020.2969931 2020

  47. [48]

    LyTimeT: Towards robust and interpretable state-variable discovery, 2025

    Kuai Yu, Crystal Su, Xiang Liu, Judah Goldfeder, Mingyuan Shao, and Hod Lipson. LyTimeT: Towards robust and interpretable state-variable discovery, 2025. URL https://arxiv.org/abs/2510.19716

    arXiv 2025

  48. [49]

    Fully body visual self-modeling of robot morphologies.Science Robotics, 2022

    Boyuan Chen, Robert Kwiatkowski, Carl V ondrick, and Hod Lipson. Fully body visual self-modeling of robot morphologies.Science Robotics, 2022

    2022

  49. [50]

    Zero-shot adap- tation to robot structural damage via natural language- informed kinodynamics modeling, 2026

    Anuj Pokhrel, Aniket Datar, Mohammad Nazeri, Francesco Cancelliere, and Xuesu Xiao. Zero-shot adap- tation to robot structural damage via natural language- informed kinodynamics modeling, 2026. URL https: //arxiv.org/abs/2602.12385

    arXiv 2026

  50. [51]

    Data-driven inverse design of spinodoid architected materials.GAMM-Mitteilungen, 48, 2025

    Alexandra Otto, Max Rosenkranz, Karl Alexander Kalina, and Markus K ¨astner. Data-driven inverse design of spinodoid architected materials.GAMM-Mitteilungen, 48, 2025. doi: 10.1002/gamm.70008

    work page doi:10.1002/gamm.70008 2025

  51. [52]

    Multifunc- tional mechanical metamaterials with tunable double- negative isotropic properties.Materials & Design, 232: 112146, 2023

    Zuyu Li, Wei Gao, Nicole Kessissoglou, Sebastian Oberst, Michael Yu Wang, and Zhen Luo. Multifunc- tional mechanical metamaterials with tunable double- negative isotropic properties.Materials & Design, 232: 112146, 2023. doi: 10.1016/j.matdes.2023.112146

    work page doi:10.1016/j.matdes.2023.112146 2023

  52. [53]

    beta-V AE: Learning basic visual concepts with a constrained variational framework.International Conference on Learning Rep- resentations, 2017

    Irina Higgins, Loic Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick, Shakir Mo- hamed, and Alexander Lerchner. beta-V AE: Learning basic visual concepts with a constrained variational framework.International Conference on Learning Rep- resentations, 2017. URL https://openreview.net/forum? id=Sy2fzU9gl

    2017

  53. [54]

    On the Continuity of Rotation Representations in Neural Networks, 2020

    Yi Zhou, Connelly Barnes, Jingwan Lu, Jimei Yang, and Hao Li. On the Continuity of Rotation Representations in Neural Networks, 2020. URL http://arxiv.org/abs/1812. 07035

    2020

  54. [55]

    Slaughter.The Linearized Theory of Elas- ticity

    William S. Slaughter.The Linearized Theory of Elas- ticity. Birkh ¨auser, Boston, MA, 2002. doi: 10.1007/ 978-1-4612-0093-2

    2002

  55. [56]

    Nonparametric regression using linear combinations of basis functions

    Robert Kohn, Michael Smith, et al. Nonparametric regression using linear combinations of basis functions. Statistics and Computing, 11(4):313–322, 2001

    2001

  56. [57]

    Optimal construction of Koopman eigenfunctions for prediction and control

    Milan Korda and Igor Mezi ´c. Optimal construction of Koopman eigenfunctions for prediction and control. IEEE Transactions on Automatic Control, 65(12):5114– 5129, 2020

    2020

  57. [58]

    DayDreamer: World models for physical robot learning.6th Annual Conference on Robot Learning, 2022

    Philipp Wu, Alejandro Escontrela, Danijar Hafner, Pieter Abbeel, and Ken Goldberg. DayDreamer: World models for physical robot learning.6th Annual Conference on Robot Learning, 2022

    2022

  58. [59]

    Berrueta, Allison Pinosky, et al

    Thomas A. Berrueta, Allison Pinosky, et al. Maximum diffusion reinforcement learning.Nature Machine Intel- ligence, 2024. doi: 10.1038/s42256-024-00829-3

    work page doi:10.1038/s42256-024-00829-3 2024

  59. [60]

    lower” platform that connects to a robot arm, and an “upper

    Rolf Pfeifer and Josh Bongard.How the body shapes the way we think: a new view of intelligence. MIT Press, 2006. APPENDIX A. Case Study of Twice-Damaged HSAs Figure 7 shows force-torque prediction errors over time for two HSAs that each undergo two damages sequentially; improved performance is closer to zero. As HSAs produce forces in response to torques ...

    2006

  60. [61]

    Kinematics:A conventional rigid Stewart platform can be kinematically constrained with four values:a l the angle from each anchor point to its associated corner on the lower platform,a u the angle from each anchor point to its associated corner on the upper platform,r l the lower anchor point radius andr u the upper anchor point radius. However, since thi...

  61. [62]

    An 0V981 UVC Arducam with an M12 lens provides Apriltag tracking at 100hz and 720p

    Control and Electronics:The soft wrist is actuated by six Feetech FT1117M-FB servos that are driven by a Mini Maestro 18 servo driver, which also measures position feedback from each of the servos. An 0V981 UVC Arducam with an M12 lens provides Apriltag tracking at 100hz and 720p. Video data is ingested using the usb cam node, and processed using the apri...