A CPG-Inspired Assist-As-Needed Controller for an Upper-Limb Rehabilitation Robot

Abstract

Robot assisted rehabilitation training is promising to be an effective therapy for post stroke patients, while it's still a challenge for robot control to achieve the objective of “Assist As Needed” (AAN) This paper introduces a new method to implement AAN training for repetitive upper limb rehabilitation training in three steps: (1) the subject's motion frequency and phase are estimated adaptively using the artificial central pattern generator (CPG) algorithm, (2) the robot reference trajectory is generated based on the minimum-jerk principle and the patient's instantaneous motion phase, and (3) the robot provides assistance according to errors between the reference and actual trajectories based on impedance control method. This method takes advantage of the repetitive feature of rehabilitation training, and generates the training trajectory on-line by considering both the subject's active motion intention and human normal motion requirements. Besides, the algorithm does not need a complex human-robot interaction model and is easy to implement on multi-DOF rehabilitation robots.