Towards Skills Evaluation of Elderly for Human-Robot Interaction

For a proactive and user-centered robotic assistance and communication, an assistive robot must make decisions about the level of assistance to be provided. Therefore, the robot must be aware of the preferences and the capabilities of the elderly. At the same time, relying on a sensing setup which is totally embedded in the assistive robot would increase its usability. In the framework of the RAMCIP project, a novel skills evaluation methodology has been developed to make the robot aware of the user's perceptual, cognitive and motor skills. This paper presents such a methodology and its preliminary evaluation. Based on a task analysis of the activities for which the robot provides assistance, the user's skills are given a score which is updated at different time scales based on the source of information. Highly reliable information is gathered from caregivers at a low rate by means of a graphical interface hosted by the robot. This information refers to standard medical examinations. Based on the modules for motion tracking, object and activity recognition, specific actions of ADL are selected to update motor skills score at a higher rate, which is typically twice per day. The two sources of information are then fused in a Kalman filter. Preliminary results on the illustrative example of arm precision show that the robot's sensing and cognitive capabilities suffice to obtain a state-of-the-art evaluation of the arm precision skill.