Virtual Coach for Wheelchair Users

Asim Smailagic, Human Computer Interaction
Dan Siewiorek, Human Computer interaction

The combination of low cost sensors and machine learning will enable a broad class of "virtual coaches" that can monitor user activity and provide interactive feedback much as a human caregiver. Our study objective is to develop an in-process, automated physical therapy trainer that assists manual wheelchair users to learn proper wheelchair use technique. The system will record wheelchair sensor data on stroke and usage patterns and use machine learning-based data analysis software to extract underlying patterns. Using a database of proper wheelchair stroke techniques, our system will be able to determine how well the user is operating the wheelchair and provide notification and guidance when the user’s technique is not correct. Our prior experiments demonstrate the ability of sensors located in various body positions to detect a wide range of activities. The stroke characteristics and activities detected by our proposed system are analogous to the activities we have detected in our prior work. Monitoring the stroke patterns of wheelchair users will aid clinicians in tracking and training proper stroke technique to reduce the incidence of repetitive stress injuries caused by improper wheelchair use technique.