How Does Delegation in Social Interaction Evolve Over Time? Navigation with a Robot for Blind People

Paper Title

How Does Delegation in Social Interaction Evolve Over Time? Navigation with a Robot for Blind People

Publication Info

• Topic area: Assistive robotics for blind individuals, focusing on social interaction and delegation in navigation.
• Keywords: Assistive robotics, blind navigation, shared control, delegation, longitudinal study, GPT-based descriptions, social interaction, obstacle detection, user trust, adaptive systems.

Background and Problem

• Problem / challenge: Existing assistive robots for blind individuals often prioritize autonomous navigation but fail to address the dynamic and collaborative nature of social interactions in real-world environments. Prior studies have focused on short-term trials and Wizard-of-Oz setups, leaving gaps in understanding how user preferences and strategies evolve over time.
• Significance: Enhancing mobility and confidence for blind individuals is critical, as navigation challenges can limit independence. Understanding how collaboration with assistive robots evolves is essential for designing systems that support long-term adoption and adapt to user needs.
• Motivation and related work: Prior research has explored shared control paradigms and assistive navigation systems but lacks longitudinal studies examining how blind users balance independence with robot assistance in real-world contexts. This paper addresses this gap by studying delegation and collaboration over three weeks.

Solution

• Proposed approach: A navigation-assistive robot equipped with a shared control interface, GPT-based environmental descriptions, and interactive features for managing social interactions.
• Novelty:
  1. Conducted a three-week longitudinal study with six blind participants in a real-world museum setting.
  2. Analyzed evolving delegation strategies and preferences for robot-mediated social interactions versus independent action.
  3. Introduced GPT-generated environmental descriptions to support situational awareness and decision-making.
  4. Provided design implications for adaptive assistive robots that accommodate dynamic social environments and individual preferences.
• Procedure and key techniques:
  • Participants interacted with a suitcase-shaped robot equipped with LiDAR sensors, RGB-D cameras, haptic feedback, and a shared control interface.
  • The robot offered GPT-based descriptions of surroundings and obstacle-specific explanations, and participants could delegate social interactions using preset verbal prompts.
  • Scenarios included crowds, physical obstacles, and lines, with staged and natural occurrences during museum navigation.
  • Post-session interviews and Likert-scale surveys captured user perceptions and adaptation over time.

Results

• Concrete findings:
  • Delegation rates varied across participants, with some increasing reliance on the robot over time (e.g., P5 rising from 20% to 100% delegation).
  • Use of GPT-based descriptions shifted from curiosity-driven exploration to targeted, purposeful engagement.
  • Participants developed interpretive frameworks for the robot’s corrective movements, influencing their decision-making and trust.
• Advantage over baselines:
  • Demonstrated that delegation preferences evolve dynamically, contrasting prior studies that reported static preferences in short-term trials.
  • Highlighted the complementary role of GPT-generated descriptions in supporting situational awareness and shared control.
• Experiments / evaluation:
  • Six blind participants navigated museum routes across three weeks, encountering staged and natural obstacles.
  • Metrics included delegation rates, GPT usage frequencies, obstacle-related actions, and Likert-scale ratings for confidence, understanding, and appropriateness.
• Limitations and future work:
  • Challenges included robot misclassification of lines versus crowds, lack of interactive Q&A functionality in GPT descriptions, and occasional navigation failures.
  • Future work should explore longer deployments, outdoor navigation, multilingual communication, and adaptive systems that personalize support based on user behavior.

Summary

This study investigated how blind individuals interact with a navigation-assistive robot over three weeks, focusing on delegation and collaboration in social navigation. Participants shifted from exploratory to deliberate use of robot features, selectively delegating social interactions and using GPT-generated descriptions to guide decisions. Results revealed that delegation preferences are dynamic, shaped by individual differences, situational demands, and trust-building over time. The findings emphasize the need for adaptive assistive robots that support flexible, personalized collaboration and highlight the importance of longitudinal evaluation to capture evolving user strategies. Future systems should prioritize transparency, context-sensitive support, and robust recovery mechanisms to empower blind users in diverse environments.