Synergy-based Gaussian Mixture Model to anticipate reaching direction identification for robotic applications

Reaching away and toward the body is one of the most im- portant upper-limb task in daily-living activities. Several robotic tech- nologies have been developed to assist neurologically impaired people with motor disabilities, such as exoskeletons and teleoperated manip- ulators. However, a high level of disability and muscle weakness could prevent an eective identication of user intention. In this paper, we present a novel approach for the classication of four reaching directions in the early phase of movement. A dimensionality-reduction algorithm based on the extraction of muscle synergies is coupled to a Gaussian Mixture Model in an evidence-accumulation framework. On average, the system identies the desired direction with 82% of accuracy at movement onset, up to 98% at 20% of reaching distance. We believe the proposed method to improve the robustness of myoelectric controlled devices, both for rehabilitation and functional assistance.