Combined state and parameter estimation for adaptive control and feedback applications for a gait rehabilitation robot

Unsupervised gait training of patients using a robotic trainer requires an individual adaption of the modelbased control for each patient as well as a feedback of the patient's performance. For these adaption and feedback tasks, estimation of the kinematic states, the friction states, and the patient-dependent parameters of the model is necessary. In this contribution, a combined estimation approach is proposed for a new gait training robot using a central difference Kalman filter, which is based on a simplified gait trainer model extended by a dynamic friction model. In the resulting overall model, the three dominating mass parameters, which depend on the patient's mass and activity, are redefined as states with integrating character. The observer approach is applied on a prototype of the gait trainer and its accuracy is evaluated with measured data using reference weights.