부산대학교 도서관

In recent years, robotic systems have been playing an increasingly important role in physiotherapy. The aim of these platforms is to aid the recovery process from strokes or muscular damage by assisting patients to perform a number of controlled tasks, thus effectively complementing the role of the physiotherapist. In this paper, we present a novel learning from demonstration framework for cooperative control in robotic-assisted physiotherapy. Unlike other approaches, the aim of the proposed system is to guide the patients to optimally execute a task based on previously learned demonstrations. This allows the generation of patient-specific gestures under the supervision of the expert physiotherapist. The guidance is performed through stiffness control of a compliant manipulator, where the stiffness profile of the generalized trajectory is determined according to the relative importance of each section of the task. In contrast with the traditional learning approach, where the execution of the generalized trajectory by the robot is automated, this cooperative control architecture allows the patients to perform the task at their own pace, while ensuring the movements are executed correctly. Increased performance of the learning framework is accomplished through a novel fast, low-cost multi-demonstration dynamic time warping algorithm used to build the model. Experimental validation of the framework is carried out using an interactive setup designed to provide further guidance through additional visual and sensory feedback based on the task model. The results demonstrate the potential of the proposed framework, showing a significant improvement in the performance of guided tasks compared to unguided ones.