Human-Centered Development of Guide-dog Robots - Project Summary In the U.S., over one million individuals live with severe visual impairments, a figure projected to double in the next 30 years. Guide dogs offer independent, natural, and safe mobility assistance, allowing handlers to walk at speeds comparable to sighted peers and navigate dynamic environments safely. However, only 2% of this population has access to guide dogs, primarily due to the limited supply of these highly trained animals. The significant cost and time required to train and deploy guide dogs (approximately $50,000 USD and 2 years) and their relatively short working span (less than 10 years) are major barriers. Additionally, guide dog handlers must provide continuous care, including medical attention, feeding, and daily exercise. Despite their compelling benefits, these challenges make guide dogs a less scalable and sustainable solution for the wider visually impaired community. Motivated by recent progress in quadruped robots and their potential for mass production and long-term sustainability, we are determined to create a practical guide-dog robot as an additional solution for navigation assistance for blind or low-vision (BLV) people. We are dedicated to aligning our development process with the authentic needs of end-users, adopting a human-centered design philosophy that involves key stakeholders throughout every stage of the process, including: 1) defining the robot’s specifications, 2) developing algorithms and integrating them into a single robotic system, and 3) conducting evaluations, identifying unforeseen issues, and iteratively refining the hardware and algorithms. Our proposed research is grounded in our comprehensive qualitative study, which included semi-structured interviews with 23 guide-dog handlers, five trainers, seven ob- servation sessions, and two blindfolded guide-dog walking experiences. Our prior work revealed the limitations of existing quadruped robots for the BLV population and pinpointed critical areas for development. Rooted in our extensive understanding of the interaction between guide dogs and their handlers, we offer technological inno- vations such as: 1) new quadruped robot hardware developed through a human-centered approach, featuring compactness, portability, extended operation, and multi-modal sensing (e.g., tactile, audio, and vision), 2) a novel co-optimization framework for robot hardware and controllers to create a robotic navigation assistant optimally designed for BLV people, and 3) a robust navigation algorithm that adapts to scene variations by creating a novel self-supervised learning framework. Our core innovation lies in introducing a human-centered development approach to creating a guide-dog robot, addressing the significant open problem spanning hardware configuration to navigation algorithm development. The developed technology will be consolidated into a single quadruped robot, which will undergo rigorous evalua- tions by guide-dog handlers and white cane users. The successful completion of this project will yield an effective navigation system for BLV individuals and trigger a paradigm shift in the field of guide-dog robot research.
