<html>
<body>
<h2><b>Research Participation Request: INVLM-Based Indoor Navigation
System for People with Visual Impairments</b></h2> <br><br>
The below research participant solicitation is being provided for
informational purposes only. The National Federation of the Blind has no
involvement in this research, but we believe that it may contribute to
our research mission.<br><br>
<br><br>
<br>
<h2><b>Research Title: InVLM: VLM-Based Indoor Navigation System
for People with Visual Impairments</b></h2> <br><br>
Being conducted by: University of Maryland, Baltimore County<br><br>
<br>
<h2><b>Purpose of the Study</b></h2>This study aims to develop and
evaluate an indoor navigation system known as InVLM (Vision-Language
Model-based Indoor Navigation System) for individuals with visual
impairments. This system integrates Multi-Modal Foundation Models with
real-time object detection and spatial awareness technologies to enhance
effective navigation in dynamic indoor settings. The study aims to
evaluate the system's effects on navigation efficiency, user
independence, and safety compared to traditional assistive methods.
In this research, we propose a novel navigation assistance system that
integrates macro and micro-navigation for goal-based navigation with the
power of multi-modal foundation models. This system can assist BLV
individuals in navigating dynamic indoor areas. To ensure safe
navigation, we integrate a real-time open-world object detection model,
YOLO-World, as an auxiliary model alongside a Large Language Model (LLM)
for more robust obstacle detection and hazard assessment. To provide
macro navigation, our system uses LLM-based floor plan analysis to
convert layouts into navigable graph representations while generating
contextual rules and trajectories. Our system combines high-level route
generation with low-level navigation to provide more efficient and
precise guidance in different indoor areas. To increase rerouting
capabilities for user reorientation, we integrate image-based
localization through image feature matching with the help of Vision
Language Models (VLMs) to detect each scene in an indoor area. Although
comprehensive experimental results are still pending, preliminary trials
demonstrate significant improvements in navigation efficiency and user
independence.<br><br>
<br>
<h2><b>Role of the Participants and Anticipated Length of the Study
</b></h2><b>Eligibility Criteria <br><br>
</b>We are looking for participants who
<ul>
<li>are blind or have low vision,
<li>Currently working in creative domains or have experience with
creative tasks. Creative domains may include different forms of art such
as photography, creative writing (e.g., fictions), drawing, fiber arts,
music, etc. or professions that include creating prototypes or design
materials, such as interaction design, product design, graphics design,
user experience (UI/UX) research and design, etc.
<li>are able to communicate in English,
<li>are 18 years old or above
</ul><br>
<br><br>
<b>By creative artifacts</b>, we refer to design prototypes, tactile arts
and crafts, images, photographs, drawings, creative writing samples
(e.g., fiction, but not academic papers at this time), and musical
compositions. These artifacts can be in any medium, as long as they were
created by the participant as part of their creative practice.<br><br>
<br><br>
Activities: Short questionnaire, brief training, and navigation tasks
with a sighted assistant<br><br>
Each participant will be asked to complete a set of 5 to 10 navigation
tasks selected from the following list. These tasks are designed to
evaluate wayfinding, navigation, and obstacle detection. A description of
each task is provided below.<br><br>
1. Point-to-Point Navigation<br><br>
A. Example for this Task: Move from the entrance of the space to a
designated room (e.g., bedroom, conference room).<br><br>
Goal: Evaluate the systems ability to guide users through multiple
turns or corridors.<br><br>
we will evaluate different functionalities of our system, including scene
descriptions, text to speech efficiency to the user<br><br>
<br><br>
Obstacle Avoidance<br><br>
Task: Follow a path that includes one or more obstacles (e.g., chairs,
boxes) and avoid them based on verbal guidance. Goal: Evaluate the
systems hazard detection and redirection capabilities.<br><br>
Safety: No face camera; safety support provided<br><br>
<br><br>
<br>
<h2><b>Proposed Participant Compensation</b></h2>Yes,
Compensation:<br><br>
o Pre-survey: $15 (10 -15 minutes, BLV participants only)
<br><br>
<a href="https://forms.gle/REBMwfqQf76xSrVE8">
https://forms.gle/REBMwfqQf76xSrVE8</a> <br><br>
You can now fill out the pre-survey, which will take approximately 10-15
minutes to complete (BLV participants only).<br><br>
In-person sessions: $40 per session (30 - 45 minutes, up to 3 sessions in
one day)<br><br>
<br>
<h2><b> <br><br>
<br>
Link or Instructions for Participants to Enter the Study </b></h2>Sign-up
form:
<a href="https://umbc.co1.qualtrics.com/jfe/form/SV_eyUrWciWyZjEg62">
https://umbc.co1.qualtrics.com/jfe/form/SV_eyUrWciWyZjEg62</a> <br><br>
Consent form :
<a href="https://drive.google.com/file/d/13K0RjJFEL8sZfJ2wHllpQCA6d0aI3BYd/view?usp=sharing">
https://drive.google.com/file/d/13K0RjJFEL8sZfJ2wHllpQCA6d0aI3BYd/view?usp=sharing</a>
<br><br>
<br>
<h2><b>Contact the following individual with any questions </b></h2>Aydin
Ayanzadeh<br><br>
3013102304<br><br>
aydina1@umbc.edu<br><br>
<br>
</body>
</html>