pith. sign in

arxiv: 2605.02983 · v1 · submitted 2026-05-04 · 💻 cs.RO · cs.SE

Human-in-the-Loop Uncertainty Analysis in Self-Adaptive Robots Using LLMs

Hassan Sartaj , Jalil Boudjadar , Mirgita Frasheri , Shaukat Ali , Peter Gorm Larsen This is my paper

Pith reviewed 2026-05-08 18:14 UTC · model grok-4.3

classification 💻 cs.RO cs.SE
keywords self-adaptive robotsuncertainty analysislarge language modelshuman-in-the-looprobot safetydesign methodologytaxonomyindustrial robotics
0
0 comments X p. Extension

The pith

RoboULM uses large language models in a human-in-the-loop process to help practitioners systematically explore uncertainties in self-adaptive robots at the design stage.

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

Self-adaptive robots must handle dynamic environments where unaddressed uncertainties risk safety violations and failures. The paper introduces RoboULM as a methodology and tool that lets practitioners guide large language models through structured prompts to identify uncertainties, their sources, impacts, and possible mitigations. It also supplies a dedicated taxonomy that catalogs these uncertainties for self-adaptive robots. When tested with 16 practitioners across four industrial use cases, the approach was rated useful and easy to understand, with particular value placed on the iterative refinement support. If the method works as intended, it offers a practical route to catching and addressing uncertainties before robots are deployed.

Core claim

RoboULM is a human-in-the-loop methodology and tool that supports practitioners in systematically exploring uncertainties at the design stage using large language models. The paper also presents an uncertainty taxonomy that catalogs uncertainties in self-adaptive robots. Evaluation with 16 practitioners from four industrial use cases shows RoboULM was perceived as useful and easy to understand, with participants especially valuing structured prompting and iterative refinement support.

What carries the argument

RoboULM methodology and tool, which combines large language models with human oversight through structured prompting and iterative refinement, together with a taxonomy that organizes uncertainties by source, impact, and mitigation in self-adaptive robots.

Load-bearing premise

Positive practitioner ratings of usefulness mean that the uncertainties surfaced by the LLM-assisted process and taxonomy actually match the ones that would cause real safety problems once the robot operates.

What would settle it

A deployment study in which robots whose uncertainties were analyzed with RoboULM still experience safety violations or failures from uncertainties that the analysis missed or misjudged.

Figures

Figures reproduced from arXiv: 2605.02983 by Hassan Sartaj, Jalil Boudjadar, Mirgita Frasheri, Peter Gorm Larsen, Shaukat Ali.

Figure 1
Figure 1. Figure 1: Overview of RoboULM and its integration into robotic development process, illustrating human-in-the-loop design-time uncertainty analysis with large language models (LLMs) based on system requirements. 3 Methodology As shown in view at source ↗
Figure 2
Figure 2. Figure 2: Results of participants’ feedback analysis across all four case studies. view at source ↗
read the original abstract

Self-adaptive robots operate in dynamic, unpredictable environments where unaddressed uncertainties can lead to safety violations and operational failures. However, systematically identifying and analyzing these uncertainties, including their sources, impacts, and mitigation strategies, remains a significant challenge given the inherent complexity of real-world environments, dynamic robotic behavior, and the rapid evolution of robotic technologies. To address this, we introduce RoboULM, a human-in-the-loop methodology and tool that supports practitioners in systematically exploring uncertainties at the design stage using large language models (LLMs). Moreover, we present an uncertainty taxonomy that provides a detailed catalog of uncertainties in self-adaptive robots. We evaluated RoboULM with 16 practitioners from four industrial use cases. The results show that RoboULM was perceived as both useful and easy to understand, with the participants particularly valuing structured prompting and iterative refinement support. These findings demonstrate the potential of RoboULM as a viable solution for systematic uncertainty analysis in complex robots.

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 paper introduces RoboULM, a human-in-the-loop methodology and tool that uses large language models (LLMs) to support practitioners in systematically identifying, analyzing, and mitigating uncertainties (sources, impacts, and strategies) in self-adaptive robots at the design stage. It also presents a new uncertainty taxonomy for such systems. The central evaluation involves 16 practitioners across four industrial use cases, with results showing that RoboULM was perceived as useful and easy to understand, particularly valuing structured prompting and iterative refinement; the authors conclude this demonstrates RoboULM's potential as a viable solution for systematic uncertainty analysis.

Significance. If the approach were shown to produce accurate and complete uncertainty sets that demonstrably reduce safety risks, it would offer a practical, scalable aid for early-stage design in a domain where manual analysis is challenging due to environmental dynamics and system complexity. The positive practitioner feedback on usability is a strength, but the current evidence base limits significance to preliminary usability insights rather than validated improvements in uncertainty handling or safety.

major comments (2)
  1. [Abstract and Evaluation] Abstract and Evaluation section: The claim that RoboULM demonstrates 'potential as a viable solution for systematic uncertainty analysis' is load-bearing on the evaluation results, yet these results measure only perceived usefulness and ease of understanding from 16 practitioners. No objective metrics, ground-truth comparisons, or validation are reported to show that the LLM-assisted taxonomy produces complete/correct uncertainty sets or reduces actual safety violations relative to expert manual analysis.
  2. [Evaluation] Evaluation section: The study design details (e.g., exact tasks given to practitioners, metrics for 'useful' and 'easy to understand', controls for bias, or how uncertainties were cross-checked) are not provided, weakening the link between positive perception and the taxonomy's claimed comprehensiveness in cataloging sources, impacts, and mitigations.
minor comments (2)
  1. [Taxonomy presentation] Clarify the exact structure and coverage criteria of the uncertainty taxonomy (e.g., how categories were derived and whether completeness was assessed beyond the four use cases).
  2. [Use cases] Provide more detail on the four industrial use cases (domain, robot types, specific uncertainties addressed) to allow readers to assess generalizability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments correctly identify that our evaluation centers on practitioner perceptions of usability rather than objective validation of uncertainty completeness or safety outcomes. We address each point below and will revise the manuscript accordingly to align claims with the evidence and provide additional methodological details.

read point-by-point responses

  1. Referee: [Abstract and Evaluation] Abstract and Evaluation section: The claim that RoboULM demonstrates 'potential as a viable solution for systematic uncertainty analysis' is load-bearing on the evaluation results, yet these results measure only perceived usefulness and ease of understanding from 16 practitioners. No objective metrics, ground-truth comparisons, or validation are reported to show that the LLM-assisted taxonomy produces complete/correct uncertainty sets or reduces actual safety violations relative to expert manual analysis.

    Authors: We agree that the abstract claim is stronger than the supporting evidence warrants. Our evaluation was designed to assess initial usability and perceived value through practitioner feedback in industrial contexts, which we view as a necessary first step before larger-scale objective studies. We will revise the abstract to state that the results demonstrate RoboULM's potential as a usable, human-in-the-loop approach for supporting uncertainty analysis, based on positive perceptions of structured prompting and iterative refinement. We will also add a limitations paragraph noting the absence of ground-truth comparisons or safety-impact metrics and identifying these as directions for future work. revision: yes

  2. Referee: [Evaluation] Evaluation section: The study design details (e.g., exact tasks given to practitioners, metrics for 'useful' and 'easy to understand', controls for bias, or how uncertainties were cross-checked) are not provided, weakening the link between positive perception and the taxonomy's claimed comprehensiveness in cataloging sources, impacts, and mitigations.

    Authors: We acknowledge the omission of these details in the current manuscript. In the revision we will expand the Evaluation section to specify: the exact tasks (practitioners applied RoboULM to uncertainty identification, impact analysis, and mitigation planning on their own industrial robot use cases); the metrics (5-point Likert scales for usefulness and ease of understanding, plus open-ended questions on valued features); bias controls (anonymous participation, no financial incentives tied to positive responses, and independent sessions); and the review process (participants iteratively refined LLM outputs and confirmed relevance to their systems, serving as domain-expert validation). These additions will clarify the evaluation's scope without overstating its reach. revision: yes

Circularity Check

0 steps flagged

No circularity detected; claims rest on new tool and external practitioner feedback

full rationale

The paper introduces RoboULM as a novel human-in-the-loop methodology and uncertainty taxonomy for self-adaptive robots, then reports results from a direct user study with 16 practitioners across four industrial use cases. The central claims concern perceived usefulness and ease of understanding, which are measured via participant feedback rather than any derivation, fitted parameter, or self-referential reduction. No equations, predictions, or load-bearing self-citations appear in the provided text that would collapse the results back to the inputs by construction. The evaluation is self-contained as an empirical assessment of a new artifact.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 2 invented entities

The central claim depends on domain assumptions about LLM effectiveness in uncertainty identification and the completeness of the new taxonomy, with no free parameters or mathematical fitting; new entities are the tool and taxonomy themselves, lacking independent falsifiable evidence beyond the user study.

axioms (2)
  • domain assumption Large language models, when guided by humans via structured prompts, can systematically surface relevant uncertainties in complex robotic systems
    Core to the RoboULM methodology; no independent verification provided beyond perceived usefulness in the study.
  • ad hoc to paper The proposed uncertainty taxonomy comprehensively catalogs sources, impacts, and mitigations for self-adaptive robots
    Introduced as a detailed catalog without external validation or comparison to existing taxonomies.
invented entities (2)
  • RoboULM no independent evidence
    purpose: Human-in-the-loop tool for LLM-supported uncertainty analysis in self-adaptive robots
    New methodology and tool introduced by the paper.
  • Uncertainty taxonomy for self-adaptive robots no independent evidence
    purpose: Catalog of uncertainties including sources, impacts, and mitigation strategies
    New taxonomy presented to support the methodology.

pith-pipeline@v0.9.0 · 5479 in / 1556 out tokens · 66093 ms · 2026-05-08T18:14:26.364409+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

15 extracted references

  1. [1]

    Out of distribution detection in self-adaptive robots with AI-powered digital twins

    Erblin Isaku, Hassan Sartaj, Shaukat Ali, Beatriz Sanguino, Tongtong Wang, Guoyuan Li, Houxiang Zhang, and Thomas Peyrucain. Out of distribution detection in self-adaptive robots with AI-powered digital twins. In2025 40th IEEE/ACM International Conference on Automated Software Engineering (ASE), pages 3403–3414. IEEE, 2025

    2025

  2. [2]

    Identifying Uncertainty in Self-Adaptive Robotics With Large Language Models.IEEE Software, 43(01):89–97, January 2026

    Hassan Sartaj, Jalil Boudjadar, Mirgita Frasheri, Shaukat Ali, and Peter Gorm Larsen. Identifying Uncertainty in Self-Adaptive Robotics With Large Language Models.IEEE Software, 43(01):89–97, January 2026

    2026

  3. [3]

    Reiya Takemura and Genya Ishigami. Uncertainty-aware trajectory planning: Using uncertainty quantification and propagation in traversability prediction of planetary rovers.IEEE Robotics & Automation Magazine, 31(2):89–99, 2024

    2024

  4. [4]

    Foundation models in robotics: Applications, challenges, and the future.The International Journal of Robotics Research, 44(5):701–739, 2025

    Roya Firoozi, Johnathan Tucker, Stephen Tian, Anirudha Majumdar, Jiankai Sun, Weiyu Liu, Yuke Zhu, Shuran Song, Ashish Kapoor, Karol Hausman, et al. Foundation models in robotics: Applications, challenges, and the future.The International Journal of Robotics Research, 44(5):701–739, 2025

    2025

  5. [5]

    The vision of autonomic computing.Computer, 36(1):41–50, 2003

    Jeffrey O Kephart and David M Chess. The vision of autonomic computing.Computer, 36(1):41–50, 2003. 8 Sartaj et al

    2003

  6. [6]

    Cameron, Simon Castle-Green, Muhammad Chughtai, Liz Dowthwaite, Ayse Kucukyilmaz, Horia A

    Harriet R. Cameron, Simon Castle-Green, Muhammad Chughtai, Liz Dowthwaite, Ayse Kucukyilmaz, Horia A. Maior, Victor Ngo, Eike Schneiders, and Bernd C. Stahl. A taxonomy of domestic robot failure outcomes: Understanding the impact of failure on trustworthiness of domestic robots. InInternational Symposium on Trustworthy Autonomous Systems, pages 1–14, New ...

    2024

  7. [7]

    A taxonomy of uncertainty for dynamically adaptive systems

    Andres J Ramirez, Adam C Jensen, and Betty HC Cheng. A taxonomy of uncertainty for dynamically adaptive systems. In2012 7th International Symposium on Software Engineering for Adaptive and Self-Managing Systems (SEAMS), pages 99–108. IEEE, 2012

    2012

  8. [8]

    Uncertainty in self-adaptive systems: A research community perspective.ACM Transactions on Autonomous and Adaptive Systems (TAAS), 15(4):1–36, 2021

    Sara M Hezavehi, Danny Weyns, Paris Avgeriou, Radu Calinescu, Raffaela Mirandola, and Diego Perez-Palacin. Uncertainty in self-adaptive systems: A research community perspective.ACM Transactions on Autonomous and Adaptive Systems (TAAS), 15(4):1–36, 2021

    2021

  9. [9]

    Sample size in usability studies.Communications of the ACM, 55(4):64–70, 2012

    Martin Schmettow. Sample size in usability studies.Communications of the ACM, 55(4):64–70, 2012

    2012

  10. [10]

    Understanding and resolving failures in human-robot interaction: Literature review and model development.Frontiers in Psychology, 9:861, 2018

    Shanee Honig and Tal Oron-Gilad. Understanding and resolving failures in human-robot interaction: Literature review and model development.Frontiers in Psychology, 9:861, 2018

    2018

  11. [11]

    Interaction between hotel service robots and humans: A hotel-specific service robot acceptance model (sram).Tourism Management Perspectives, 36, 2020

    Laura Fuentes-Moraleda, Patricia D ´ıaz-P´erez, Alicia Orea-Giner, Ana Mu ˜noz- Maz ´on, and Teresa Villac ´e- Molinero. Interaction between hotel service robots and humans: A hotel-specific service robot acceptance model (sram).Tourism Management Perspectives, 36, 2020

    2020

  12. [12]

    Torch-uncertainty: Deep learning uncertainty quantification

    Adrien Lafage, Olivier Laurent, Firas Gabetni, and Gianni Franchi. Torch-uncertainty: Deep learning uncertainty quantification. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2025

    2025

  13. [13]

    Uncertainty quantification for safe and reliable autonomous vehicles: A review of methods and applications.IEEE Transactions on Intelligent Transportation Systems, 2025

    Ke Wang, Chongqiang Shen, Xingcan Li, and Jianbo Lu. Uncertainty quantification for safe and reliable autonomous vehicles: A review of methods and applications.IEEE Transactions on Intelligent Transportation Systems, 2025

    2025

  14. [14]

    Safety evaluation of robot systems via uncertainty quantification

    Woo-Jeong Baek and Torsten Kr¨oger. Safety evaluation of robot systems via uncertainty quantification. InIEEE International Conference on Robotics and Automation (ICRA 2023), pages 10532–10538. IEEE, 2023

    2023

  15. [15]

    A digital twin enabled runtime analysis and mitigation for autonomous robots under uncertainties

    Jalil Boudjadar and Mirgita Frasheri. A digital twin enabled runtime analysis and mitigation for autonomous robots under uncertainties. InProceedings of the 22nd International Conference on Informatics in Control, Automation and Robotics - Volume 2: ICINCO. SciTePress, 2025. 9

    2025