Compensation for the passive dynamics of a five-bar neurorehabilitation robot

Summary form only given. We have designed a compensator for the passive dynamics of a five-bar neurorehabilitation robot. This model-based compensator was tuned off-line using a simplex search method to minimize the error between measured joint torques during passive movement of the robot arm and torques predicted by the model during the same movement. The simplex algorithm selected the three independent parameters of the equations of motion to minimize the squared difference between measured and estimated joint torques. These parameters are implemented in a real-time operating system (xPC; the Mathworks) controlling the robot. This system yields stable compensation of over 80% of the passive dynamics of the manipulandum. This compensation was empirically found to be stable over the entire workspace of the robot.