cs.ROOct 20, 2026

Navigation Explainability for Object Avoidance with LLMs for Assistive Robots

Authors: Courtney Young, Yi Lian, Taery Kim, Sehoon Ha

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Blind and visually impaired individuals face uncertainty when navigating unfamiliar areas. Whether a cane or a guide dog is used for navigation, this may prove to be challenging. To mitigate this challenge, we present the Context-Aware Navigation Explanation framework that explains guide dog movements for obstacle avoidance and navigation using a large language model. This system integrates perceptual context with planned future trajectories to generate natural-language explanations of robot movement. We deploy the system on a Unitree Go2 platform in indoor navigation scenarios and evaluate how different explanation context configurations affect user trust and explanation satisfaction. Results show a strong positive relationship between explanation satisfaction and trust, and suggest that combining obstacle information with future trajectory leads to the most effective explanations. These findings highlight the potential of context-aware language explanations to improve transparency and trust in assistive robot navigation.

Figures & tables

Fig. 1 : Overview of the Context-Aware Navigation Explanation system. The robot guide dog leads a BVI user while perceiving obstacles (e.g., a chair) and generating real-time natural language explanations (e.g., “Chair on left, moving right”) based on scene context and planned motion.
Fig. 2 : System overview of the Context-Aware Navigation Explanation framework. The object detection module extracts obstacle information from recent frames, while the future trajectory module provides planned waypoints and navigation intent. These inputs are combined into a structured prompt for an LLM to generate real-time natural language explanations.
ConditionFrame HistoryNavigation DataLLM Call Interval(s)Example
C11No3“Chairs detected left side, far away. Continue forward safely.”
C21Yes3“Clear path ahead. Chair left 1.7m away. Continue forward.”
C34Yes2“Person left 4.7m. Continue forward, path clear ahead.”
TABLE I : User Study Conditions
Fig. 3 : Experimental map showing two navigation routes (Path A and Path B), location of static obstacles (chairs), and dynamic agents (pedestrians) moving through the environment during evaluation of the explanation system.
Fig. 4 : Trust and Explanation Satisfaction Correlation, which shows strong correlation between TRP-HRI across participants. Left: Aggregated plot. Right: Per condition plot.
Fig. 5 : Mean scores for TPS-HRI (left) and ESS (right) . RM ANOVA shows significance across conditions.
Fig. 6 : Item-level mean scores for (a) trust (TPS-HRI) and (b) explanation satisfaction (ESS) across conditions (C1-C3). The following items have significant p values below 0.05: appropriate information, unresponive, satisfaction, sufficient detail, usefulness, and accuracy.
More in this topic

Visually Impaired Navigation Assistance

See all 24 papers

Explore similar work

CardsList

  1. Ontology-Guided Reasoning for Affordance-Based Explanations of Robot Navigation

    May 27, 2026Amar Halilovic, Vahidin Hasic, Senka KrivicOntology-Based ReasoningRobot Navigation

    Institute of Artificial Intelligence, Ulm University, 89081 Ulm, Germany · Faculty of Electrical Engineering, University of Sarajevo, 71000 Sarajevo, BiH

  2. Explaining Navigation Decisions: Stage-Aware Visual Attribution for Vision-Language Navigation

    Oct 25, 2026Yiyuan Yao, Xinpeng Ding, Minglei Chen +6VLM InterpretabilityVision-Language Navigation

    Xidian University, Xi’an, China.

  3. From Woofs to Words: Towards Intelligent Robotic Guide Dogs with Verbal Communication

    Mar 13, 2026Yohei Hayamizu, David DeFazio, Hrudayangam Mehta +7Human-Robot InteractionAssistive Robotics

    The State University of New York at Binghamton 4400 Vestal Pkwy E, Binghamton, NY 13902 USA