This technology enables Vision Language Models to interpret raw floor plan images and generate accurate navigation plans without requiring structured digital maps. By combining visual and text-based reasoning, it allows robots and smart devices to understand spaces directly from images. The result is flexible, scalable navigation that works wherever a map image is available.
Background: Mobile robots and smart devices have difficulty using human-oriented floor plan images because these maps are not machine-readable and require extensive preprocessing. Existing systems depend on structured digital maps or environment-specific training, and they often fail to interpret complex spatial relationships or large open spaces. As a result, robots and assistive devices cannot reliably generate navigation plans directly from raw map images, limiting their usefulness in real-world environments.
Technology Overview: The invention applies Vision Language Models to parse floor plan images and produce navigation instructions from combined visual and textual prompts. A floor plan image and a task description containing start and goal locations are provided to the VLM, which generates a multi-step action sequence for reaching the goal. The approach achieves high accuracy on navigation tasks, performing especially well on small and moderately complex maps. It leverages commercially available VLMs such as GPT-4o and Claude 3.5 to reason over unstructured map images without specialized preprocessing.
Advantages: • Enables VLMs to interpret raw floor plan images for navigation • Achieves high accuracy in multi-step navigation tasks • Integrates visual and textual reasoning for flexible task specification • Eliminates dependency on structured or preprocessed digital maps • Adapts to mobile robots and consumer devices without environment-specific training • Improves accessibility by supporting natural language interaction • Extends to robotics, urban planning, and emergency response use cases • Provides a competitive advantage in autonomous indoor navigation systems
Applications: • Mobile robotics navigation • Smartphone-based indoor navigation • Assistive navigation for visually impaired users • Indoor/outdoor navigation in dynamic environments • Emergency response planning • Urban infrastructure mapping
Intellectual Property Summary: • United States 63/884,317 Provisional Filed 09/18/2025 Status Filed
Stage of Development: Prototype
Licensing Status: This technology is available for licensing.
Licensing Potential: Strong potential for robotics developers, navigation software providers, and assistive technology companies seeking flexible, map-agnostic navigation solutions powered by AI-driven visual and language reasoning.
Additional Information: Additional technical details and evaluation results available upon request.
Inventors: Jeremy Blackburn, David DeFazio, Hrudayangam Mehta, Shiqi Zhang