Control of a wheelchair using an adaptive K-Means clustering of head poses

Abstract

Operating a wheelchair is often a difficult task for individuals with severe disabilities. Also, with the progress of the condition, the use of most current robotic assistive technologies becomes less attractive or simply not applicable anymore. In this work, we developed a system that allows a user to operate a wheelchair using only their heads. Our method utilizes an Infrared (IR) depth sensor to capture the user's head pose, while it includes an adaptive component to the detection of that pose. The adaptation, based on a type of Re-enforcement K-Means clustering, can accommodate users with limited and changing head mobility - no matter how skewed the head motion may become with the progress of the condition. We tested the system using five test subjects, who simulated `normal' an `abnormal' motions of the head. The system worked well in all cases, and all test subjects found the interface quite intuitive.