Next-Gen Museum Guides: Autonomous Navigation and Visitor Interaction with an Agentic Robot

Alessandra Sciutti; Francesca Cocchella; Francesco Rea; Luca Garello; Manuel Catalano

REVIEW 3 major objections 5 minor 1 cited by

Next-Gen Museum Guides: Autonomous Navigation and Visitor Interaction with an Agentic Robot

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read An autonomous robot using a large language model for conversation and map-based navigation led 34 visitors through a maritime museum and was generally positively received.

desk verdict A genuine LLM-guided museum robot field study with honest limitations, undermined as printed by an impossible standard deviation and missing navigation metrics. read the letter →

arxiv 2507.12273 v2 pith:Q6VVUJG4 submitted 2025-07-16 cs.RO

Luca Garello , Francesca Cocchella , Alessandra Sciutti , Manuel Catalano , Francesco Rea This is my paper · ORCID

classification cs.RO

keywords autonomousmuseumguiderobothuman-robotinteractionlargelanguagemodelssimultaneouslocalizationandmappingsocialroboticsvisitorengagementspeechrecognitiondeployment

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper reports a field deployment in which the dual-arm mobile robot Alter-Ego, running the large language model (LLM) GPT-4o mini with function calling for dialogue and a simultaneous localization and mapping (SLAM) navigation stack for movement, acted as a fully autonomous guide inside a maritime museum. The authors claim that such an integrated system can conduct real tours, answer spontaneous visitor questions about exhibits, and adapt its route on request, and that the 34 participants generally received it well. The evidence combines counts from conversation logs with pre- and post-interaction surveys, and it also documents clear limitations: about one third of visitor questions fell outside the robot's knowledge base, the noisiest exhibit areas produced comprehension-failure rates above 30 percent, and perceived competence dropped slightly after the interaction. The paper's conclusion is that fully autonomous LLM-based guidance is feasible in a cultural venue, and that speech comprehension and responsiveness are the current bottlenecks.

What carries the argument

The load-bearing mechanism is the coupling of an LLM dialogue loop with navigation through two callable functions: go to(destination), which moves the robot to a museum area chosen by the LLM from a fixed list, and end tour(), which returns the robot to the entrance. The LLM is driven by a dynamic prompt rebuilt at each area transition, containing the robot's identity, current location, visited and unvisited areas, the exhibit knowledge base, and chat history, which keeps token use low while keeping answers context-aware. Around that core sit the perception and motion layers: YOLOv10-n face detection to trigger interaction, a SLAM-based navigation stack for mapping and localization, and trajectory planning with obstacle avoidance.

What would settle it

Ask two independent coders to re-label the 31 recorded conversations using a written coding scheme for correct answer, out-of-scope question, and comprehension failure; if their agreement is low, the reported means in Table II are not reproducible.

Watch

Extended reading notes

Core claim

The paper's central assertion is that a single robot can carry the whole museum-guide function end to end: it detects an approaching visitor, greets them, offers a tour, navigates to exhibit areas, answers questions through an LLM whose prompt is updated with the robot's current location and visit progress, and ends the tour when asked or after 120 seconds of silence. In the reported trial, 34 participants interacted with the robot, visiting on average 5.69 of seven areas, asking 7.29 questions, receiving satisfactory answers to 3.71 of them, hearing out-of-scope responses to 2.25, and encountering 1.26 comprehension failures. Post-interaction surveys showed higher self-other closeness and a small but significant drop in perceived competence, while agency, experience, and warmth did not change. The authors interpret these results as evidence that visitors can engage positively with an autonomous LLM-based robot guide in a real museum, with comprehension and responsiveness remaining the limiting factors.

Load-bearing premise

The quantitative description of the interaction depends on the authors' own manual labeling of each exchange as a correct answer, an out-of-scope question, or a comprehension failure, and this labeling was not checked against a written codebook or a second coder.

Editorial extensions

If this is right

Museum guides can be deployed without teleoperation: the same architecture can run a full tour as long as the exhibit areas and knowledge base are provided.
Knowledge coverage, not dialogue quality, is the main constraint: about one third of visitor questions fall outside the prepared knowledge base.
Acoustic conditions should be treated as first-order design variables: the noisiest areas produced comprehension-failure rates near 33 percent.
Fixed interaction states miss real visitor behavior: pausing, re-engaging, and ambiguous destination phrasing were not handled and sometimes ended the tour.
Announcing the robot in advance had almost no effect on perception, so positive reception is better explained by the interaction itself than by expectation framing.

Reading between the lines

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

Editorial extension: the two-function pattern of go to and end tour generalizes to any structured service environment where destinations are enumerable, so the same LLM-to-navigation glue could drive library, hospital, or fairground assistants.
Editorial extension: because the transcript classification has no reported reliability check, an independent re-coding of the same 31 logs would be a cheap, decisive test of the quantitative findings.
Editorial extension: single-visit acceptance may include a novelty effect; a longitudinal or repeated-visit study would separate genuine usefulness from first-contact curiosity.
Editorial extension: participants who compared the robot to ChatGPT or to a humanoid research robot show that expectations are set by general AI exposure, so manipulating stated capability rather than only robot presence could explain the small drop in perceived competence.

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.

Desk Editor's Note

read the letter

Read this one. It is a real field deployment of an LLM-powered museum guide robot on the Alter-Ego platform, with 34 visitors, pre/post surveys, transcripts, and an honest qualitative write-up. The headline claim that the robot was generally well-received is plausible and largely supported by the qualitative data. What is genuinely new is the integration: GPT-4o mini with function calling for tour decisions, plus SLAM navigation, tested in a museum rather than a lab. Each component is established, but the combination is a useful reference point for HRI in cultural spaces.

Credit where due: the authors document failures (comprehension errors, latency, unintended tour endings) and don't oversell the dialogue quality. The qualitative thematic analysis is transparent about its constructivist stance, and the participant quotes are illustrative. The navigation architecture is described in enough detail to replicate.

The soft spots are in the quantitative layer. The most serious is in Section VII-C: they report σpre=5.39 for the Competence scale on a 1–7 Likert scale. That is arithmetically impossible; the maximum SD for a mean of 5.66 on a 1–7 scale is about 2.5. This is likely a typo (1.39?), but as printed it invalidates the H1 competence result. The authors must recompute from raw data. Second, the title and abstract promise 'autonomous navigation' and the introduction promises 'navigation performance', but the paper reports no navigation metrics—no path tracking, no localization error, no success rate. The only behavioral data are visitor counts and error rates per area. Third, the transcript classification in Table II (correct answers, out-of-scope, comprehension failures) is manual categorization with no coding scheme or inter-coder reliability reported, so those numbers are not independently verifiable. That is a limitation, not a fatal flaw, but it needs transparency. Fourth, the conclusion overstates: it says 'measurable shifts in perceived agency, competence, and closeness,' but Agency did not change significantly; only IOS and the questionable Competence did. The H2 finding is only a marginal effect on areas visited, not on perceptions.

Overall, the qualitative core is solid and the deployment is real. The quantitative issues are fixable. This deserves a serious referee, but the authors should be asked to correct the statistics, add or explicitly scope out navigation metrics, and document the coding procedure before acceptance.

Referee Report

3 major / 5 minor

Summary. The paper presents Alter-Ego, an LLM-powered mobile robot deployed as a museum guide, and reports a field study with 34 participants. The system combines SLAM-based navigation with GPT-4o mini conversational functions and was evaluated through pre/post surveys, conversation transcriptions, and qualitative thematic analysis. The authors claim the robot was generally well-received and contributed to an engaging museum experience, with limitations in comprehension and responsiveness.

Significance. If the empirical claims are supported, this work provides valuable evidence about the deployment of an autonomous LLM-driven robot in a real cultural venue, a context where field studies remain relatively rare. The mixed-method design, the use of established HRI scales, and the attention to both quantitative and qualitative outcomes are strengths. The paper also offers concrete insights into interaction breakdowns and visitor expectations that can inform future system design. However, the credibility of the central quantitative claims is currently undermined by an impossible reported standard deviation and the absence of navigation performance metrics, so the contribution cannot be fully assessed as presented.

major comments (3)

[Section VII-C] The reported standard deviation for the Competence scale, σ_pre=5.39 with mean μ_pre=5.66 on a 1–7 Likert scale, is arithmetically impossible; the maximum possible SD given that mean is sqrt((7−5.66)(5.66−1)) ≈ 2.50. Consequently, the associated p=0.019 and the conclusion of a 'slight worsening' in perceived competence are uninterpretable as printed. The authors must re-analyze the raw survey data or correct the typo, and the H1 result cannot be evaluated until this is resolved.
[Section VII (Results) and Section IV (Experiment)] The abstract and Introduction claim 'fully autonomous navigation' and 'robust SLAM techniques', but the Results section reports no navigation performance metrics—no success rate for go_to() calls, no localization error, no path-length or time-efficiency measures, and no count of operator interventions. Without such metrics, the paper's central claim of autonomous navigation is not empirically supported; the only navigation-related data are the areas visited and interaction duration in Table II.
[Section VI-B1 and Table II] The counts of 'answers', 'out-of-scope questions', and 'comprehension failures' are derived from the authors' manual categorization of conversation transcriptions, but no coding scheme, decision criteria, or inter-coder reliability is reported. These manually coded counts underpin the quantitative description of the interaction in Table II, so the reliability of this measure is load-bearing and must be documented.

minor comments (5)

[Section VI-B and Section VII-C] The headings contain typos: 'Quantitave Analysis' (Section VI-B) and 'Quantitivative Analysis' (Section VII-C) should be 'Quantitative Analysis'.
[Section VII-C] In the H2 paragraph, 'expect for' should be 'except for'.
[Section II] The citation order in the sentence about 'RoboX, CiceRobot and Mobot' does not match the reference numbering: [1] is CiceRobot, [2] is RoboX, and [3] is Mobot, while the text lists RoboX first.
[Section I (Introduction)] The Introduction refers to 'Section 2', 'Section 3', etc., but the manuscript uses Roman numerals for sections; please harmonize the cross-references.
[Figure 3 caption] The caption states that the pre-experiment Human-like appearance score is excluded due to 'a Cronbach's alpha below acceptance rate', but the threshold and the exact value (α=0.33) should be stated in the text rather than only in the caption.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the paper is an empirical field study whose results are measured, not derived from its inputs.

full rationale

This paper does not present a derivation chain in which a prediction is algebraically or statistically forced by its inputs. The central claim that Alter-Ego was 'generally well-received and contributed to an engaging museum experience' is supported by survey data, conversation transcripts, and post-interaction feedback collected from 34 participants in a real museum. The self-citations present in the paper play a supporting, non-load-bearing role: reference [7] describes the Alter-Ego robot platform used in the study, and reference [13] is the source of scales adapted for evaluating robot movements and autonomy. Neither citation determines the reported outcomes, which depend on the participants' responses and the transcribed interactions. The authors' manual categorization of transcripts into 'correct answers,' 'out-of-scope questions,' and 'comprehension failures' is a measurement and coding choice rather than a circular derivation; even if the coding were unreliable, that would be a validity concern, not a case of the result being equivalent to the input by construction. Likewise, the self-generated survey scales for trust and interactiveness are instruments, not fitted parameters renamed as predictions. The reported standard deviation of 5.39 on a 1–7 Likert scale for Competence is arithmetically impossible and is a serious internal-consistency error that affects the reliability of the H1 result, but it is a data-reporting inconsistency rather than a circularity. Because no equation in the paper reduces a claimed result to its own assumptions or to a self-citation chain, the circularity score is minimal.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper's claims rest on standard psychometric assumptions and on the authors' manual classification of conversations; no new entities are introduced.

free parameters (2)

Interaction timeout = 120 seconds
Hard-coded timeout for assuming visitor disengagement; not justified, affects tour termination behavior and the 'end tour' counts.
Face size threshold for proximity = Not reported
YOLOv10 detection size threshold determines when interaction starts; no value or calibration given.

assumptions (3)

domain assumption Self-report Likert scales validly measure the intended constructs (warmth, competence, agency, etc.)
Relies on established scales [8-12], but two scales are self-generated without validation (Section V-A).
domain assumption Participants answered surveys honestly and independently
No manipulation checks beyond condition, no attention checks.
domain assumption GPT-4o mini responses were 'correct' when the authors judged them satisfactory
'Answers' parameter includes only 'correct answers' as judged by the authors, no objective rubric (Section VI-B).

how reviews work

0 comments

Cite this review

Pith. "Pith review of Next-Gen Museum Guides: Autonomous Navigation and Visitor Interaction with an Agentic Robot." pith.science (2026). https://pith.science/paper/Q6VVUJG4

@misc{pith2026250712273,
  author       = {Pith},
  title        = {Pith review of: Next-Gen Museum Guides: Autonomous Navigation and Visitor Interaction with an Agentic Robot},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q6VVUJG4}},
  note         = {Machine review of arXiv:2507.12273}
}

read the original abstract

Autonomous robots are increasingly being tested into public spaces to enhance user experiences, particularly in cultural and educational settings. This paper presents the design, implementation, and evaluation of the autonomous museum guide robot Alter-Ego equipped with advanced navigation and interactive capabilities. The robot leverages state-of-the-art Large Language Models (LLMs) to provide real-time, context aware question-and-answer (Q&A) interactions, allowing visitors to engage in conversations about exhibits. It also employs robust simultaneous localization and mapping (SLAM) techniques, enabling seamless navigation through museum spaces and route adaptation based on user requests. The system was tested in a real museum environment with 34 participants, combining qualitative analysis of visitor-robot conversations and quantitative analysis of pre and post interaction surveys. Results showed that the robot was generally well-received and contributed to an engaging museum experience, despite some limitations in comprehension and responsiveness. This study sheds light on HRI in cultural spaces, highlighting not only the potential of AI-driven robotics to support accessibility and knowledge acquisition, but also the current limitations and challenges of deploying such technologies in complex, real-world environments.

Figures

Figures reproduced from arXiv: 2507.12273 by the authors.

**Figure 2.** Alter-Ego robot in front of one area of the tour. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Mean Likert scores (1–7 scale) from pre- and post-experiment [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Mixed-Agent Museum Tour Guide Design Improves Gendered Learning Outcomes and Visitor Preferences
cs.RO 2026-07 conditional novelty 5.0 of 10

A mixed physical-and-projected robot tour team raised learning gains for female participants in the bantering condition, but engagement and experience ratings did not differ across conditions.

Reference graph

Works this paper leans on

21 extracted references · 20 canonical work pages · cited by 1 Pith paper

[1]

Cicerobot: a cognitive robot for interactive museum tours,

A. Chella, M. Liotta, and I. Macaluso, “Cicerobot: a cognitive robot for interactive museum tours,” Industrial Robot: An International Journal , vol. 34, no. 6, pp. 503–511, 2007

work page 2007
[2]

Robox at expo. 02: A large-scale installation of personal robots,

R. Siegwart, K. O. Arras, S. Bouabdallah, D. Burnier, G. Froidevaux, X. Greppin, B. Jensen, A. Lorotte, L. Mayor, M. Meisseret al., “Robox at expo. 02: A large-scale installation of personal robots,”Robotics and Autonomous Systems , vol. 42, no. 3-4, pp. 203–222, 2003

work page 2003
[3]

The mobot museum robot installations: a five year experiment,

I. Nourbakhsh, C. Kunz, and T. Willeke, “The mobot museum robot installations: a five year experiment,” in Proceedings 2003 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS

work page 2003
[4]

Interactive robots as social partners and peer tutors for children: A field trial,

T. Kanda, T. Hirano, D. Eaton, and H. Ishiguro, “Interactive robots as social partners and peer tutors for children: A field trial,” Human– Computer Interaction , vol. 19, no. 1-2, pp. 61–84, 2004

work page 2004
[5]

Tour guide robot: a 5g-enabled robot museum guide,

S. Rosa, M. Randazzo, E. Landini, S. Bernagozzi, G. Sacco, M. Pic- cinino, and L. Natale, “Tour guide robot: a 5g-enabled robot museum guide,” Frontiers in Robotics and AI , vol. 10, p. 1323675, 2024

work page 2024
[6]

Deploying social robots in museum settings: A quasi-systematic review exploring purpose and acceptability,

N. Gasteiger, M. Hellou, and H. S. Ahn, “Deploying social robots in museum settings: A quasi-systematic review exploring purpose and acceptability,” International Journal of Advanced Robotic Systems , vol. 18, no. 6, p. 17298814211066740, 2021

work page 2021
[7]

Alter-ego: a mobile robot with a functionally anthropomorphic upper body designed for physical interaction,

G. Lentini, A. Settimi, D. Caporale, M. Garabini, G. Grioli, L. Pal- lottino, M. G. Catalano, and A. Bicchi, “Alter-ego: a mobile robot with a functionally anthropomorphic upper body designed for physical interaction,” IEEE Robotics & Automation Magazine , vol. 26, no. 4, pp. 94–107, 2019

work page 2019
[8]

Blurring human–machine distinctions: Anthropomorphic appearance in social robots as a threat to human distinctiveness,

F. Ferrari, M. P. Paladino, and J. Jetten, “Blurring human–machine distinctions: Anthropomorphic appearance in social robots as a threat to human distinctiveness,” International Journal of Social Robotics , vol. 8, no. 2, pp. 287–302, 2016

work page 2016

Show all 21 references

[9]

Dimensions of mind perception,

H. M. Gray, K. Gray, and D. M. Wegner, “Dimensions of mind perception,” Science, vol. 315, no. 5812, pp. 619–619, 2007

2007
[10]

Universal dimensions of social cognition: Warmth and competence,

S. T. Fiske, A. J. Cuddy, and P. Glick, “Universal dimensions of social cognition: Warmth and competence,” Trends in cognitive sciences , vol. 11, no. 2, pp. 77–83, 2007

2007
[11]

Measuring accep- tance of an assistive social robot: a suggested toolkit,

M. Heerink, B. Krose, V . Evers, and B. Wielinga, “Measuring accep- tance of an assistive social robot: a suggested toolkit,” in RO-MAN 2009-The 18th IEEE International Symposium on Robot and Human Interactive Communication . IEEE, 2009, pp. 528–533

2009
[12]

Inclusion of other in the self scale and the structure of interpersonal closeness

A. Aron, E. N. Aron, and D. Smollan, “Inclusion of other in the self scale and the structure of interpersonal closeness.” Journal of personality and social psychology , vol. 63, no. 4, p. 596, 1992

1992
[13]

Remember me-user-centered implementation of working memory architectures on an industrial robot,

J. Bernotat, L. Landolfi, D. Pasquali, A. Nardelli, and F. Rea, “Remember me-user-centered implementation of working memory architectures on an industrial robot,” Frontiers in Robotics and AI , vol. 10, p. 1257690, 2023

2023
[14]

Using thematic analysis in psychology,

V . Braun and V . Clarke, “Using thematic analysis in psychology,” Qualitative research in psychology , vol. 3, no. 2, pp. 77–101, 2006

2006
[15]

Reflecting on reflexive thematic analysis,

——, “Reflecting on reflexive thematic analysis,” Qualitative research in sport, exercise and health , vol. 11, no. 4, pp. 589–597, 2019

2019
[16]

One size fits all? what counts as quality practice in (reflexive) thematic analysis?

——, “One size fits all? what counts as quality practice in (reflexive) thematic analysis?” Qualitative Research in Psychology, vol. 18, no. 3, pp. 328–352, 2021

2021
[17]

Thematic analysis: Striving to meet the trustworthiness criteria,

L. S. Nowell, J. M. Norris, D. E. White, and N. J. Moules, “Thematic analysis: Striving to meet the trustworthiness criteria,” International Journal of Qualitative Methods , vol. 16, no. 1, pp. 1–13, 2017

2017
[18]

The icub humanoid robot: An open-systems platform for research in cognitive development,

G. Metta, L. Natale, F. Nori, G. Sandini, D. Vernon, L. Fadiga, C. V on Hofsten, K. Rosander, M. Lopes, J. Santos-Victor et al. , “The icub humanoid robot: An open-systems platform for research in cognitive development,” Neural networks, vol. 23, no. 8-9, pp. 1125– 1134, 2010

2010
[19]

Human–robot interaction: a survey,

M. A. Goodrich and A. C. Schultz, “Human–robot interaction: a survey,” F oundations and Trends in Human–Computer Interaction , vol. 1, no. 3, pp. 203–275, 2007

2007
[20]

J. C. Nunnally, Psychometric Theory: 2d Ed . McGraw-Hill, 1978

1978
[2003]

No.03CH37453) , vol

(Cat. No.03CH37453) , vol. 4, 2003, pp. 3636–3641 vol.3

2003

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Cicerobot: a cognitive robot for interactive museum tours,

A. Chella, M. Liotta, and I. Macaluso, “Cicerobot: a cognitive robot for interactive museum tours,” Industrial Robot: An International Journal , vol. 34, no. 6, pp. 503–511, 2007

work page 2007

[2] [2]

Robox at expo. 02: A large-scale installation of personal robots,

R. Siegwart, K. O. Arras, S. Bouabdallah, D. Burnier, G. Froidevaux, X. Greppin, B. Jensen, A. Lorotte, L. Mayor, M. Meisseret al., “Robox at expo. 02: A large-scale installation of personal robots,”Robotics and Autonomous Systems , vol. 42, no. 3-4, pp. 203–222, 2003

work page 2003

[3] [3]

The mobot museum robot installations: a five year experiment,

I. Nourbakhsh, C. Kunz, and T. Willeke, “The mobot museum robot installations: a five year experiment,” in Proceedings 2003 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS

work page 2003

[4] [4]

Interactive robots as social partners and peer tutors for children: A field trial,

T. Kanda, T. Hirano, D. Eaton, and H. Ishiguro, “Interactive robots as social partners and peer tutors for children: A field trial,” Human– Computer Interaction , vol. 19, no. 1-2, pp. 61–84, 2004

work page 2004

[5] [5]

Tour guide robot: a 5g-enabled robot museum guide,

S. Rosa, M. Randazzo, E. Landini, S. Bernagozzi, G. Sacco, M. Pic- cinino, and L. Natale, “Tour guide robot: a 5g-enabled robot museum guide,” Frontiers in Robotics and AI , vol. 10, p. 1323675, 2024

work page 2024

[6] [6]

Deploying social robots in museum settings: A quasi-systematic review exploring purpose and acceptability,

N. Gasteiger, M. Hellou, and H. S. Ahn, “Deploying social robots in museum settings: A quasi-systematic review exploring purpose and acceptability,” International Journal of Advanced Robotic Systems , vol. 18, no. 6, p. 17298814211066740, 2021

work page 2021

[7] [7]

Alter-ego: a mobile robot with a functionally anthropomorphic upper body designed for physical interaction,

G. Lentini, A. Settimi, D. Caporale, M. Garabini, G. Grioli, L. Pal- lottino, M. G. Catalano, and A. Bicchi, “Alter-ego: a mobile robot with a functionally anthropomorphic upper body designed for physical interaction,” IEEE Robotics & Automation Magazine , vol. 26, no. 4, pp. 94–107, 2019

work page 2019

[8] [8]

Blurring human–machine distinctions: Anthropomorphic appearance in social robots as a threat to human distinctiveness,

F. Ferrari, M. P. Paladino, and J. Jetten, “Blurring human–machine distinctions: Anthropomorphic appearance in social robots as a threat to human distinctiveness,” International Journal of Social Robotics , vol. 8, no. 2, pp. 287–302, 2016

work page 2016