Generative AI-Aided Navigation for the Visually Impaired and Blind

Navigation for the visually impaired and blind remains to be a major barrier to independence. Existing assistive tools like guide dogs or white mobility canes provide limited, immediate information within a range of about 5 feet. Alternatively, assistive applications for navigation only provide static, generalizable information about a broader area that could be a few hundred feet radius to miles. Currently, no solution effectively covers the 5 to 20 feet range, leaving users without crucial information about their surroundings in this mid-distance area. This project explores the potential of state-of-the-art vision-language models (VLMs) to provide new navigation solutions for the visually impaired and blind that bridge the aforementioned gap in information about the environment. VLMs prove capable of identifying key objects and reasoning from corresponding text and images in real time, making them the ideal candidate for assistive technology. Leveraging these capabilities, these models may be integrated into wearable or extendable devices that allow users to receive continuous support in unfamiliar environments, improving their independence and maintaining safety. This project investigates the practical application of VLMs in real-world scenarios, with an emphasis on ease of use and reliability. This work has the potential to expand the role of assistive technology in daily life and complement existing solutions for more intuitive and responsive understanding.