Estimation of joint torques through musculoskeletal models and measurements of muscle activations can be used for real-time control of robotic devices for rehabilitation. Many works developed models for analytic one joint motion, but less are found that develop models for functional multijoint movements. In this work we develop a methodology for tuning and optimizing Hill-based EMG-driven models oriented to the force control of robotic exoskeletons for the upper limb, selecting the more suitable parameters to be optimized. The model is tuned from experimental data obtained from healthy people. The torques estimated by that model will serve as reference for force-based control of an exoskeleton for rehabilitation.