The Assessment system consists of three components, namely 1) intention detection, 2) performance assessment and 3) behavioural monitoring.
Automatic intention detection not requiring conscious inputs from the user is an essential feature for a functional support system. We will base the arm support end-point control on gaze-based decoding of action intention using the presented high-performance low-cost eye tracking system. Reach and grasp intention detection will be additionally based on supplementary on-body sensing of muscle activivation (EMG/MMG), body movement and interface forces, and on a head-mounted environmental observation system.
Assessment of motor performance relative to capacity informs the personalised motor support intelligence about the performance reserve of the user. It will be based on information derived from the multimodal sensory interface with the person and the environment. Performance measures will be derived from sensing the task execution in relation to the level of support: to what level does the user himself contribute to the generation of movements, how frequently does the user functionally interact with the environment and how effective is the user in contributing to functional tasks. The capacity is conceived as the maximum performance the user can achieve. The first approach is to assess capacity from separate clinical tests. Subsequently, capacity will be estimated implicitly from daily-life functioning as the maximum level of user performance when the user is maximally motivated by the system.
The personalized behavioural model will predict user performance depending on arm and hand support level and supplementary motivational communication, taking into account context, including social interactions, and environment. This will be the basis for the support intelligence to decide about the minimal level of support given to the user to realise the intended movements. The behaviour model will be iteratively identified and adapted, based on the comparison between predicted and subsequently observed user performance.