AI Framework Converts Floor Plans into Accessible Indoor Navigation for Blind Users
A team of researchers has created an innovative framework that transforms a single image of a floor plan into a detailed knowledge base to enhance navigation for individuals with blindness or low vision. This advanced system employs a multi-agent strategy to construct a spatial knowledge graph, featuring self-correcting processes and iterative feedback. The technology includes a Path Planner for devising accessible routes and a Safety Evaluator to identify potential dangers. Testing at the UMBC Math and Psychology building yielded success rates of 92.31%, 76.92%, and 61.54% for varying route lengths, significantly improving upon existing navigation solutions.
Key facts
- Framework converts a single floor plan image into a structured knowledge base
- Multi-agent module uses self-correcting pipeline with iterative retry loops
- Path Planner generates accessible navigation instructions
- Safety Evaluator agent assesses potential hazards along routes
- Evaluated on UMBC Math and Psychology building (floors MP-1 and MP-3)
- Also tested on CVC-FP benchmark
- Success rates on MP-1: 92.31% (short), 76.92% (medium), 61.54% (long)
- System reduces need for costly per-building infrastructure
Entities
Institutions
- University of Maryland, Baltimore County (UMBC)
- CVC-FP
Locations
- UMBC Math and Psychology building
- MP-1
- MP-3