WanderGuide: Indoor Map-less Robotic Guide for Exploration by Blind People

Research Background and Problem Statement

• Identified Problems or Challenges:
  Existing navigation systems typically rely on predefined maps and infrastructure, making deployment in completely unknown environments costly and difficult. Additionally, many systems focus on goal-oriented navigation rather than encouraging autonomous exploration, limiting blind users' opportunities to discover new environments.

• Importance of the Problem:
  Autonomous exploration is crucial for the daily lives and social participation of blind users. It not only enhances their freedom and independence in navigating environments but also creates more enjoyable experiences, a core need that current systems fail to adequately address.

• Research Motivation and Related Work:
  Existing research primarily focuses on navigation tasks that help blind users reach specific destinations (e.g., using prebuilt maps or sensors), with little attention paid to the need for spontaneous exploration. Furthermore, only a limited number of navigation systems offer multi-purpose or perceptual capabilities. The design space for map-free navigation and exploration systems, such as WanderGuide, remains underexplored.

Solution

• Proposed Method and Solution:
  The authors developed a robotic navigation system called WanderGuide, which does not rely on predefined maps and primarily focuses on exploration functionality. The system provides environmental information to blind users through real-world scene descriptions and Q&A interactions, while also supporting user control over the robot's movement direction and navigation to specific locations.

• Innovative Features:

  • Map-Free System: WanderGuide achieves automatic detection of waypoints in the environment, overcoming the traditional reliance on maps and infrastructure.
  • Personalized Exploration Experience: The system offers three scene description modes (detailed, balanced, and concise) based on user preferences and enables in-depth interaction through Q&A.
  • Integrated Functionality: By combining automated navigation, waypoint selection, obstacle avoidance, and multimodal large language models (MLLM), the system enhances its capabilities in indoor exploration tasks.

• Implementation Steps and Key Technologies:

  1. Select a wheeled robot as the hardware platform, with automated navigation and obstacle avoidance as the foundation.
  2. Use SLAM (Simultaneous Localization and Mapping) algorithms to construct a real-time environmental cost map and select waypoints through corner detection and clustering algorithms.
  3. Integrate multimodal large language models (MLLM) to generate real-time scene descriptions and provide Q&A interaction functionality.
  4. Develop a user interface that allows users to adjust navigation speed, scene description detail, and exploration mode via buttons or voice interaction.

Research Outcomes

• Specific Achievements:

  • Successfully developed a fully functional map-free exploration system that meets the indoor exploration needs of blind users.
  • The system allows users to dynamically adjust description detail and navigation modes based on different preferences and situational requirements.
  • Features such as "Take me there" and Q&A interaction significantly enhance users' freedom to explore.

• Advantages Over Existing Solutions:

  • Does not rely on complex map construction or infrastructure, making it more adaptable to dynamic and unfamiliar environments.
  • Provides tailored scene descriptions and Q&A interactions based on user preferences, enhancing the personalized experience.
  • Supports fully independent exploration without the need for human assistance.

• Experimental or Evaluation Results:

  • Two rounds of user studies confirmed the system's effectiveness in settings such as science museums and shopping malls. Users expressed high satisfaction with the experience of goal-free exploration and personalized scene descriptions.
  • The system achieved moderate to high usability ratings (SUS scores exceeding 70) and demonstrated low workload ratings in the TLX evaluation.

• Limitations and Future Directions:

  • Limitations:
    • The system struggles with errors in recognizing and describing specific objects or location names, particularly in complex scenes where multimodal language models (MLLM) face performance bottlenecks.
    • Lacks sound source recognition capabilities, limiting its ability to enhance user exploration in audio-rich environments.
    • Experiments were primarily conducted in indoor spaces; further research is needed to evaluate the system's performance in more complex environments (e.g., areas with stairs or crowded spaces).
  • Future Directions:
    • Extend beyond visual information by integrating audio recognition and multimodal data processing capabilities.
    • Optimize hardware configurations, such as using motion-blur-resistant cameras and high-precision sensors.
    • Expand research to diverse cultural contexts and scenarios, and conduct longitudinal studies with long-term users to assess changes in usage habits.

Through this research, WanderGuide demonstrates a novel map-free robotic system that provides robust support for independent exploration by blind users, while also inspiring various possibilities for future technological and design advancements.