Adaptable force control in robotic rehabilitation

This paper presents initial work on a direct force control framework that would be used to assist stroke patients during rehabilitation therapy in the future. This framework is expected to provide an optimal time-varying assistive force to stroke patients in varying physical and environmental conditions. This control structure has two main modules. The first module is a human arm parameter estimation model. The second module is an artificial neural network (ANN)-based PI-gain scheduling controller. The ANN uses estimated human arm parameters to select the appropriate PI gains for the direct force controller. The feasibility and efficacy of the controller is demonstrated with a PUMA 560 robotic manipulator on various artificial environments.