本文由 AI 分析生成
建立時間： 2026-04-02 來源： https://arxiv.org/abs/2508.01235
Summary
NarraGuide is a prototype mobile robot that integrates location-aware LLM-based narration into a telepresence system, enabling remote users to explore and learn about unfamiliar places through dialogue. Deployed in a geology museum with 20 remote participants, the study examines how users perceived the robot’s narrative role, engaged in dialogue, and expressed preferences about bystander encounters.
NarraGuide 將具位置感知的 LLM 敘事能力整合進移動遠程呈現機器人，讓遠程用戶透過對話探索陌生地點。在地質博物館的用戶研究（N=20）中，探討用戶如何感知機器人角色及旁觀者遭遇偏好。
Prerequisites
- Robotic telepresence — the baseline system NarraGuide extends; understanding standard telepresence limitations (navigation dependency on prior knowledge) motivates the narrative addition.
- Location-aware LLM prompting — the technical core is conditioning LLM output on real-time location data; understanding RAG or context injection patterns helps.
- HRI user studies — the evaluation is a qualitative/mixed-methods study; understanding think-aloud protocols and thematic analysis matters for interpreting findings.
Core Idea
Standard telepresence robots assume users know where to go and what they’re looking at. NarraGuide inverts this: the robot proactively generates location-aware narrative guidance, providing context about exhibits and wayfinding without requiring the remote user to have prior knowledge of the space. The LLM generates dialogue-ready narration grounded in the robot’s current physical position, transforming the robot from a passive camera-on-wheels into an active narrative guide. The geology museum deployment tests whether this makes unfamiliar remote spaces genuinely explorable.
Results
- N=20 remote participants completed museum tours using NarraGuide.
- Qualitative findings: users perceived robot as having a distinct “guide” role separate from simple telepresence.
- Preferences for bystander encounter strategies were expressed (avoiding vs. engaging passersby).
- Dialogue engagement patterns observed (topic initiation, follow-up questions).
(Quantitative metrics not provided in abstract; see full paper for rating scales.)
Limitations
- Author-stated: single deployment context (geology museum); generalizability to other environments uncertain.
- Unstated: LLM narrative quality depends on location data accuracy; GPS/indoor positioning errors could degrade narrative relevance. Bystander privacy implications not addressed in abstract.
Reproducibility
- Code: not mentioned in abstract.
- Datasets: custom deployment study; proprietary museum location data.
- Compute: standard LLM inference; no training required beyond prompting.
Insights
NarraGuide represents a shift from “telepresence as remote control” to “telepresence as mediated experience.” The location-aware LLM layer solves a real gap: most telepresence studies assume the remote user has a mental map of the space, which fails for genuine exploration. The bystander preference finding is practically important — social robots in public spaces must negotiate unexpected human interactions, and user preferences about this are underexplored.
Connections
Raw Excerpt
We explore how integrating location-aware LLM-based narrative capabilities into a mobile robot can support remote exploration… Our findings reveal how users perceived the robot’s role, engaged in dialogue in the tour, and expressed preferences for bystander encountering.