pith. sign in

arxiv: 2604.20468 · v2 · submitted 2026-04-22 · 💻 cs.RO · cs.AI· cs.CL· cs.HC· cs.LG

MOMO: A framework for seamless physical, verbal, and graphical robot skill learning and adaptation

Markus Knauer , Edoardo Fiorini , Maximilian M\"uhlbauer , Stefan Schneyer , Promwat Angsuratanawech , Florian Samuel Lay , Timo Bachmann , Samuel Bustamante
show 5 more authors
Korbinian Nottensteiner Freek Stulp Alin Albu-Sch\"affer Jo\~ao Silv\'erio Thomas Eiband
This is my paper

Pith reviewed 2026-05-10 00:00 UTC · model grok-4.3

classification 💻 cs.RO cs.AIcs.CLcs.HCcs.LG
keywords robot skill adaptationkinesthetic teachingnatural language interfacesergodic controlmovement primitivesvirtual fixtureshuman-robot interactionindustrial robotics
0
0 comments X

The pith

A robot framework lets non-experts adapt skills through touch, voice commands, and graphics by combining intention detection, safe language models, movement primitives, virtual fixtures, and ergodic control.

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

The paper presents MOMO, an interactive system that lets users modify robot behaviors using three different ways: physical corrections by hand, spoken instructions, and a visual web interface for adjusting paths and parameters. It brings together five parts that work together: detecting when a person intends to teach or correct, a language model that only picks and tunes existing safe functions instead of writing new code, kernelized movement primitives to store motions, probabilistic guides to help record demonstrations, and ergodic control for tasks that cover surfaces evenly. The authors show that the language component can extend adaptations from basic path learning to surface-finishing tasks, and they tested the full setup on a seven-degree-of-freedom robot during a trade fair to show it can handle real industrial conditions.

Core claim

The paper claims that the integration of energy-based human-intention detection, a tool-based LLM architecture for safe natural language adaptation, Kernelized Movement Primitives for motion encoding, probabilistic Virtual Fixtures for guided demonstration recording, and ergodic control for surface finishing creates a single framework that supports skill adaptation through kinesthetic touch, verbal commands, and graphical editing, and that this tool-based LLM approach generalizes skill adaptation from KMPs to ergodic control, as shown by enabling voice-commanded surface finishing on a 7-DoF torque-controlled robot in an industrial demonstration setting.

What carries the argument

The tool-based LLM architecture, where the language model selects and parameterizes predefined safe functions rather than generating code, which enables natural language commands to adapt skills first encoded in KMPs and then extends those adaptations to ergodic control for surface finishing.

If this is right

  • Kinesthetic touch allows precise spatial corrections while the system detects human intentions through energy-based signals.
  • Natural language inputs can trigger high-level semantic changes to tasks using the safe tool-selection mechanism.
  • A graphical web interface supports visualization of trajectories, parameter inspection, and direct editing of via-points by drag-and-drop.
  • The same adaptation pipeline extends from KMP-encoded motions to ergodic control, enabling voice commands for surface-finishing behaviors.
  • The full integration operates on a 7-DoF torque-controlled robot and has been shown in a live industrial trade-fair setting.

Where Pith is reading between the lines

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

  • If the multi-modal system works as described, operators without programming expertise could adjust robots on the factory floor without calling in specialists.
  • The separation of language understanding from code generation might reduce the chance of dangerous outputs when extending the framework to new control methods beyond KMPs and ergodic control.
  • Combining physical, verbal, and graphical inputs in one loop could support incremental corrections where a spoken command first sets a goal and a subsequent touch refines the path.
  • Further tests in noisy or cluttered environments would be needed to confirm whether intention detection remains accurate when multiple people are present.

Load-bearing premise

The tool-based LLM architecture will reliably select and parameterize only safe predefined functions from natural language inputs without misinterpretation or unsafe behavior in unstructured industrial environments.

What would settle it

A recorded interaction in which a user issues a natural language command specifying safe parameters for a surface-finishing task and the robot instead applies incorrect or unsafe parameters would falsify the reliability of the language component.

Figures

Figures reproduced from arXiv: 2604.20468 by Alin Albu-Sch\"affer, Edoardo Fiorini, Florian Samuel Lay, Freek Stulp, Jo\~ao Silv\'erio, Korbinian Nottensteiner, Markus Knauer, Maximilian M\"uhlbauer, Promwat Angsuratanawech, Samuel Bustamante, Stefan Schneyer, Thomas Eiband, Timo Bachmann.

Figure 1
Figure 1. Figure 1: Overview of the framework. The three interaction modalities [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) Bearing ring insertion task: the transparent robot shows original [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
read the original abstract

Industrial robot applications require increasingly flexible systems that non-expert users can easily adapt for varying tasks and environments. However, different adaptations benefit from different interaction modalities. We present an interactive framework that enables robot skill adaptation through three complementary modalities: kinesthetic touch for precise spatial corrections, natural language for high-level semantic modifications, and a graphical web interface for visualizing geometric relations and trajectories, inspecting and adjusting parameters, and editing via-points by drag-and-drop. The framework integrates five components: energy-based human-intention detection, a tool-based LLM architecture (where the LLM selects and parameterizes predefined functions rather than generating code) for safe natural language adaptation, Kernelized Movement Primitives (KMPs) for motion encoding, probabilistic Virtual Fixtures for guided demonstration recording, and ergodic control for surface finishing. We demonstrate that this tool-based LLM architecture generalizes skill adaptation from KMPs to ergodic control, enabling voice-commanded surface finishing. Validation on a 7-DoF torque-controlled robot at the Automatica 2025 trade fair demonstrates the practical applicability of our approach in industrial settings.

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 / 2 minor

Summary. The manuscript presents the MOMO framework integrating five components—energy-based human-intention detection, a tool-based LLM architecture for safe natural-language adaptation, Kernelized Movement Primitives (KMPs), probabilistic Virtual Fixtures, and ergodic control—to enable robot skill adaptation via kinesthetic touch, verbal commands, and a graphical web interface. It claims this architecture generalizes skill adaptation from KMPs to ergodic control for voice-commanded surface finishing and validates practical applicability through a live demonstration on a 7-DoF torque-controlled robot at the Automatica 2025 trade fair.

Significance. If the integration and tool-calling LLM design function as described, the work provides a practical systems contribution for multi-modal, non-expert robot programming in industrial settings, with the safety-oriented LLM approach and extension to ergodic control offering reusable architectural patterns. The trade-fair demonstration supports real-world relevance, though the absence of quantitative results limits evaluation of robustness and seamlessness.

major comments (2)
  1. [Validation and demonstration] Validation/demonstration (abstract and experimental results section): The central claim of 'practical applicability' and 'seamless' adaptation rests on the Automatica 2025 live demonstration, yet the manuscript supplies only qualitative footage and descriptions with no reported quantitative metrics such as success rates over repeated trials, adaptation latency, error rates, or failure-mode analysis. This leaves the strength of the applicability assertion unsubstantiated.
  2. [LLM component description] Tool-based LLM architecture (component integration section): The paper states that the LLM selects and parameterizes predefined functions rather than generating code to ensure safety and enable generalization from KMPs to ergodic control. However, no details are given on the size or content of the function library, the exact prompting or intention-detection logic that prevents unsafe selections, or any empirical checks for misinterpretation under unstructured inputs, which is load-bearing for the reliability claim.
minor comments (2)
  1. [Abstract] The abstract introduces 'energy-based human-intention detection' without referencing the specific energy formulation or its mathematical integration with KMPs and ergodic control; adding a brief equation or pointer to prior work would aid clarity.
  2. [Framework overview] A summary table listing the five components, their modalities, and interaction points would improve readability of the overall architecture.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their constructive feedback, which highlights important areas for strengthening the manuscript. We address each major comment below, indicating planned revisions where feasible.

read point-by-point responses

  1. Referee: [Validation and demonstration] Validation/demonstration (abstract and experimental results section): The central claim of 'practical applicability' and 'seamless' adaptation rests on the Automatica 2025 live demonstration, yet the manuscript supplies only qualitative footage and descriptions with no reported quantitative metrics such as success rates over repeated trials, adaptation latency, error rates, or failure-mode analysis. This leaves the strength of the applicability assertion unsubstantiated.

    Authors: We agree that quantitative metrics would strengthen the claims of practical applicability and seamlessness. The Automatica 2025 demonstration occurred as a live trade-fair event, precluding systematic repeated trials or metric collection. In revision, we will expand the experimental results section with a detailed narrative of the demonstrated scenarios, observed adaptations across modalities, and qualitative performance indicators. We will also add a limitations subsection discussing the challenges of quantitative evaluation in live settings and future plans for controlled experiments. revision: partial

  2. Referee: [LLM component description] Tool-based LLM architecture (component integration section): The paper states that the LLM selects and parameterizes predefined functions rather than generating code to ensure safety and enable generalization from KMPs to ergodic control. However, no details are given on the size or content of the function library, the exact prompting or intention-detection logic that prevents unsafe selections, or any empirical checks for misinterpretation under unstructured inputs, which is load-bearing for the reliability claim.

    Authors: We acknowledge that additional specifics on the tool-based LLM would improve clarity and support the safety claims. In the revised manuscript, we will expand the relevant section to describe the function library (including its size and example functions for KMP and ergodic adaptations), the prompting and intention-detection logic for safe selection, and any available observations from testing with unstructured inputs. revision: yes

standing simulated objections not resolved
  • We cannot provide specific quantitative metrics (e.g., success rates, latencies) from the Automatica 2025 live demonstration, as no such data was systematically recorded during the event.

Circularity Check

0 steps flagged

No significant circularity; pure integration framework

full rationale

The manuscript describes a modular systems integration of five pre-existing components (energy-based intention detection, tool-calling LLM, KMPs, probabilistic virtual fixtures, ergodic control) under a shared interface. No equations, parameter fits, or derivations appear; the claimed generalization from KMPs to ergodic control is achieved simply by routing both through the same LLM tool layer rather than by any mathematical reduction. No self-citations are invoked as uniqueness theorems or load-bearing premises. The work is therefore self-contained as an engineering architecture whose validity rests on external demonstration rather than internal self-reference.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Relies on standard robotics assumptions about the reliability of KMPs, virtual fixtures, and ergodic control; the main addition is the integration architecture and LLM tool-calling pattern for safety.

axioms (1)
  • domain assumption Tool-based LLM architecture ensures safe adaptation by restricting outputs to predefined functions
    Invoked to justify verbal modality without code generation risks
invented entities (1)
  • MOMO framework no independent evidence
    purpose: Unified multi-modal skill adaptation system
    The integrated system is the primary contribution

pith-pipeline@v0.9.0 · 5571 in / 1245 out tokens · 51603 ms · 2026-05-10T00:00:43.690129+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

38 extracted references · 38 canonical work pages · 1 internal anchor

  1. [1]

    Villani, F

    V . Villani, F. Pini, F. Leali, and C. Secchi, “Survey on human-robot collaboration in industrial settings: Safety, intuitive interfaces, and applications,”Mechatronics, vol. 55, pp. 248–266, 2018. [Online]. Available: https://doi.org/10.1016/j.mechatronics.2018.02.009

    work page doi:10.1016/j.mechatronics.2018.02.009 2018

  2. [2]

    Villani and J

    A. Billard, S. Calinon, R. Dillmann, and S. Schaal,Robot Programming by Demonstration. Springer, 2008, pp. 1371–1394. [Online]. Available: https://doi.org/10.1007/978-3-540-30301-5 60

    work page doi:10.1007/978-3-540-30301-5 2008

  3. [3]

    Recent advances in robot learning from demonstration,

    H. Ravichandar, A. S. Polydoros, S. Chernova, and A. Billard, “Recent advances in robot learning from demonstration,”Annual Review of Control, Robotics, and Autonomous Systems, vol. 3, no. 1, pp. 297–330, 2020. [Online]. Available: https://doi.org/10.1146/ annurev-control-100819-063206

    work page 2020

  4. [4]

    Interactive learning via physical human feedback using uncertainty- aware energy tanks,

    E. Fiorini, M. Knauer, T. Eiband, M. Iskandar, and J. Silv ´erio, “Interactive learning via physical human feedback using uncertainty- aware energy tanks,”IEEE Robotics and Automation Letters (RA-L), 2026, early Access. [Online]. Available: https://ieeexplore.ieee.org/ document/11425762

    work page arXiv 2026

  5. [5]

    IROSA: Interactive robot skill adaptation using natural language,

    M. Knauer, S. Bustamante, T. Eiband, A. Albu-Sch ¨affer, F. Stulp, and J. Silv ´erio, “IROSA: Interactive robot skill adaptation using natural language,”IEEE Robotics and Automation Letters (RA-L), 2026, early Access. [Online]. Available: https://ieeexplore.ieee.org/document/ 11425760

    work page 2026

  6. [6]

    Kernelized movement primitives,

    Y . Huang, L. Rozo, J. Silv ´erio, and D. G. Caldwell, “Kernelized movement primitives,”International Journal of Robotics Research (IJRR), vol. 38, no. 7, pp. 833–852, 2019. [Online]. Available: https://doi.org/10.1177/0278364919846363

    work page doi:10.1177/0278364919846363 2019

  7. [7]

    A probabilistic approach to multi-modal adaptive virtual fixtures,

    M. M ¨uhlbauer, T. Hulin, B. Weber, S. Calinon, F. Stulp, A. Albu- Sch¨affer, and J. Silv ´erio, “A probabilistic approach to multi-modal adaptive virtual fixtures,”IEEE Robotics and Automation Letters (RA-L), vol. 9, no. 6, pp. 5298–5305, 2024. [Online]. Available: https://doi.org/10.1109/LRA.2024.3384759

    work page doi:10.1109/lra.2024.3384759 2024

  8. [9]

    A unified framework for probabilistic dynamic-, trajectory- and vision-based virtual fixtures,

    [Online]. Available: https://arxiv.org/abs/2506.10239

    work page arXiv

  9. [10]

    An ergodic approach to robotic surface finishing with learned motion preferences,

    S. Schneyer, K. Nottensteiner, A. Albu-Sch ¨affer, F. Stulp, and J. Silv ´erio, “An ergodic approach to robotic surface finishing with learned motion preferences,”IEEE Transactions on Robotics (T-RO),

    work page

  10. [11]

    Available: https://doi.org/10.1109/TRO.2025.3641752

    [Online]. Available: https://doi.org/10.1109/TRO.2025.3641752

    work page doi:10.1109/tro.2025.3641752 2025

  11. [12]

    Understanding natural language commands for robotic navigation and mobile manipulation,

    S. Tellex, T. Kollar, S. Dickerson, M. R. Walter, A. G. Banerjee, S. Teller, and N. Roy, “Understanding natural language commands for robotic navigation and mobile manipulation,” inAAAI Conference on Artificial Intelligence (AAAI), 2011. [Online]. Available: https: //ojs.aaai.org/index.php/AAAI/article/view/7979

    work page 2011

  12. [13]

    RoboFlow: A flow-based visual programming language for mobile manipulation tasks,

    S. Alexandrova, Z. Tatlock, and M. Cakmak, “RoboFlow: A flow-based visual programming language for mobile manipulation tasks,” inIEEE International Conference on Robotics and Automation (ICRA), 2015, pp. 5537–5544. [Online]. Available: https://doi.org/10.1109/ICRA.2015. 7139973

    work page doi:10.1109/icra.2015 2015

  13. [14]

    ImmFusion: Robust mmWave-RGB Fusion for 3D Human Body Reconstruction in All Weather Conditions

    J. Liang, W. Huang, F. Xia, P. Xu, K. Hausman, B. Ichter, P. Florence, and A. Zeng, “Code as policies: Language model programs for embodied control,” inIEEE International Conference on Robotics and Automation (ICRA), 2023, pp. 9493–9500. [Online]. Available: https://doi.org/10.1109/ICRA48891.2023.10160591

    work page doi:10.1109/icra48891.2023.10160591 2023

  14. [15]

    Toolformer: Language models can teach themselves to use tools,

    T. Schick, J. Dwivedi-Yu, R. Dess `ı, R. Raileanu, M. Lomeli, E. Hambro, L. Zettlemoyer, N. Cancedda, and T. Scialom, “Toolformer: Language models can teach themselves to use tools,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 36, 2023. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/2023/hash/ d842425e4bf79ba...

    work page 2023

  15. [16]

    What are tools anyway? a survey from the language model perspective,

    Z. Wang, Z. Cheng, H. Zhu, D. Fried, and G. Neubig, “What are tools anyway? a survey from the language model perspective,” in Conference on Language Modeling (COLM), 2024. [Online]. Available: https://openreview.net/forum?id=Xh1B90iBSR

    work page 2024

  16. [17]

    OVITA: Open- vocabulary interpretable trajectory adaptations,

    A. Maurya, T. Ghosh, A. Nguyen, and R. Prakash, “OVITA: Open- vocabulary interpretable trajectory adaptations,”IEEE Robotics and Automation Letters (RA-L), vol. 10, no. 11, pp. 11 054–11 061, 2025. [Online]. Available: https://doi.org/10.1109/LRA.2025.3606309

    work page doi:10.1109/lra.2025.3606309 2025

  17. [18]

    A dynamical system approach for detection and reaction to human guidance in physical human-robot interaction,

    M. Khoramshahi and A. Billard, “A dynamical system approach for detection and reaction to human guidance in physical human-robot interaction,”Autonomous Robots (AuRo), vol. 44, no. 8, pp. 1411–1429,

    work page

  18. [19]

    Available: https://doi.org/10.1007/s10514-020-09934-9

    [Online]. Available: https://doi.org/10.1007/s10514-020-09934-9

    work page doi:10.1007/s10514-020-09934-9

  19. [20]

    Virtual fixtures: Perceptual tools for telerobotic manipulation,

    L. B. Rosenberg, “Virtual fixtures: Perceptual tools for telerobotic manipulation,” inProceedings of IEEE Virtual Reality Annual International Symposium (VRAIS), 1993, pp. 76–82. [Online]. Available: https://doi.org/10.1109/VRAIS.1993.380795

    work page doi:10.1109/vrais.1993.380795 1993

  20. [21]

    Metrics for ergodicity and design of ergodic dynamics for multi-agent systems

    G. Mathew and I. Mezi ´c, “Metrics for ergodicity and design of ergodic dynamics for multi-agent systems,”Physica D: Nonlinear Phenomena, vol. 240, no. 4-5, pp. 432–442, 2011. [Online]. Available: https://doi.org/10.1016/j.physd.2010.10.010

    work page doi:10.1016/j.physd.2010.10.010 2011

  21. [22]

    Ergodicity-based cooperative multiagent area coverage via a potential field,

    S. Ivi ´c, B. Crnkovi ´c, and I. Mezi ´c, “Ergodicity-based cooperative multiagent area coverage via a potential field,”IEEE Transactions on Cybernetics (TCYB), vol. 47, no. 8, pp. 1983–1993, 2017. [Online]. Available: https://doi.org/10.1109/TCYB.2016.2634400

    work page doi:10.1109/tcyb.2016.2634400 1983

  22. [23]

    Interactive incremental learning of generalizable skills with local trajectory modulation,

    M. Knauer, A. Albu-Sch ¨affer, F. Stulp, and J. Silv ´erio, “Interactive incremental learning of generalizable skills with local trajectory modulation,”IEEE Robotics and Automation Letters (RA-L), vol. 10, no. 4, pp. 3398–3405, 2025. [Online]. Available: https://doi.org/10. 1109/LRA.2025.3542209

    work page arXiv 2025

  23. [24]

    Passive Variable Impedance For Shared Control

    M. M ¨uhlbauer, N. Werner, R. Balachandran, T. Hulin, J. Silv ´erio, F. Stulp, and A. Albu-Sch ¨affer, “Passive variable impedance for shared control,”arXiv preprint arXiv:2604.20557, 2026. [Online]. Available: https://arxiv.org/abs/2604.20557

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  24. [25]

    A study on speech recognition control for a surgical robot,

    K. Zinchenko, C.-Y . Wu, and K.-T. Song, “A study on speech recognition control for a surgical robot,”IEEE Transactions on Industrial Informatics (TII), vol. 13, no. 2, pp. 607–615, 2017. [Online]. Available: https://doi.org/10.1109/TII.2016.2625818

    work page doi:10.1109/tii.2016.2625818 2017

  25. [26]

    Links and nodes: A real-time middleware for distributed robotic systems,

    F. Schmidt, “Links and nodes: A real-time middleware for distributed robotic systems,” 2020, open-source, GPLv3. Documentation: https: //links-and-nodes.readthedocs.io. [Online]. Available: https://gitlab.com/ links and nodes/links and nodes

    work page 2020

  26. [27]

    Impedance control: An approach to manipulation,

    N. Hogan, “Impedance control: An approach to manipulation,” in1984 American Control Conference. IEEE, 07 1984

    work page 1984

  27. [28]

    Hybrid force-impedance control for fast end-effector motions,

    M. Iskandar, C. Ott, A. Albu-Sch ¨affer, B. Siciliano, and A. Dietrich, “Hybrid force-impedance control for fast end-effector motions,”IEEE Robotics and Automation Letters (RA-L), vol. 8, no. 7, pp. 3931–3938, 2023

    work page 2023

  28. [29]

    Collision detection, identification, and localization on the DLR SARA robot with sensing redundancy,

    M. Iskandar, O. Eiberger, A. Albu-Sch ¨affer, A. De Luca, and A. Dietrich, “Collision detection, identification, and localization on the DLR SARA robot with sensing redundancy,” inIEEE International Conference on Robotics and Automation (ICRA), 2021, pp. 3111–3117

    work page 2021

  29. [30]

    An approach for imitation learning on Riemannian mani- folds,

    M. J. A. Zeestraten, I. Havoutis, J. a. Silv ´erio, S. Calinon, and D. G. Caldwell, “An approach for imitation learning on Riemannian mani- folds,”IEEE Robotics and Automation Letters (RA-L), vol. 2, no. 3, pp. 1240–1247, 2017

    work page 2017

  30. [31]

    Task-specific reconfiguration of variable workstations using automated planning of workcell layouts,

    T. Bachmann, O. Eiberger, T. Eiband, F. Lay, P. Angsuratanawech, I. Rodriguez, P. Lehner, F. Stulp, and K. Nottensteiner, “Task-specific reconfiguration of variable workstations using automated planning of workcell layouts,” inISR Europe 2023; 56th International Symposium on Robotics (ISR), 2023, pp. 250–257

    work page 2023

  31. [32]

    Flexible robotic assembly based on ontological representation of tasks, skills, and resources,

    P. M. Sch ¨afer, F. Steinmetz, S. Schneyer, T. Bachmann, T. Eiband, F. S. Lay, A. Padalkar, C. S ¨urig, F. Stulp, and K. Nottensteiner, “Flexible robotic assembly based on ontological representation of tasks, skills, and resources,” inProceedings of the International Conference on Principles of Knowledge Representation and Reasoning (KR), vol. 18, no. 1, ...

    work page doi:10.24963/kr.2021/73 2021

  32. [33]

    Unifying skill-based programming and programming by demonstration through ontologies,

    T. Eiband, F. Lay, K. Nottensteiner, and D. Lee, “Unifying skill-based programming and programming by demonstration through ontologies,” Procedia Computer Science, vol. 232, pp. 595–605, 2024. [Online]. Available: https://doi.org/10.1016/j.procs.2024.01.059

    work page doi:10.1016/j.procs.2024.01.059 2024

  33. [34]

    Collaborative programming of robotic task decisions and recovery behaviors,

    T. Eiband, C. Willibald, I. Tannert, B. Weber, and D. Lee, “Collaborative programming of robotic task decisions and recovery behaviors,”Autonomous Robots (AuRo), vol. 47, no. 2, pp. 229–247,

    work page

  34. [35]

    pause” / “resume

    [Online]. Available: https://doi.org/10.1007/s10514-022-10062-9 6 IEEE ROBOTICS AND AUTOMATION PRACTICE (RA-P). PREPRINT VERSION. Supplementary Material A framework for seamless physical, verbal, and graphical robot skill learning and adaptation S-I. IMPLEMENTATIONDETAILS A. Software Architecture and Technology Stack Fig. S1 shows the software architectur...

    work page doi:10.1007/s10514-022-10062-9 2000

  35. [36]

    add via-point

    The frontend sends an “add via-point” service call with the point index and new(x, y, z)position via the LN WebSocket bridge,

    work page

  36. [37]

    MOMO inserts the via-point into the KMP model with γ=10 −8,

    work page

  37. [38]

    the frontend requests the updated model mean and covariance via a second service call,

    work page

  38. [39]

    move left at the start

    the trajectory visualization refreshes, showing original (blue) and adapted (yellow) trajectories. Right-clicking a via-point opens a context menu for adapting (dragging) or deleting it, with the trajectory updating in real time after each modification. b) LLM Chat Integration.:The ChatBox compo- nent sends user text to the MOMO backend via the set_llm_in...

    work page 2000