Learning-Based Compensation-Corrective Control Strategy for Upper Limb Rehabilitation Robots

Trunk compensations are commonly observed when stroke patients perform reaching tasks, that negatively affect their long-term motor recovery. To restrain the compensatory patterns, this study proposes a learning-based compensation-corrective (LBCC) control strategy for upper limb rehabilitation robots. The proposed LBCC strategy comprises a learning and a reproduction phase. Specifically, a learning from demonstration framework is employed to generalize the referenced task in the learning phase. The compensatory patterns are corrected by shoulder restraint, hand assistive, and coupling force feedback, which are generated by the LBCC control strategy, in the reproduction phase. Experiments were carried out on ten healthy subjects as a feasibility study. The trunk compensations were significantly reduced in three types of reaching tasks with the force feedback. In addition, the proposed LBCC control strategy significantly enhances the upper limb motor performance, therefore, providing a user experience similar to human-assisted rehabilitation for stroke patients.