Functional electrical stimulation can be used to reanimate paralyzed muscles, thereby restoring movement following spinal cord injury (SCI). Developing FES controllers for complex tasks involving multiple muscles and kinematic degrees of freedom remains formidable, partly due to the challenges associated with developing robust models for control. Here we demonstrate the utility of incorporating a data-driven semi-parametric model into an FES controller. A subject-specific model was estimated from experimental data collected from a single subject with a high-level SCI and an implanted FES neuroprosthesis. The model was used as part of a feedforward controller of arm kinematics. This system performed well for regions of the workspace where sufficient muscle strength was available. It could be used to predict the additional assistance needed when FES-generated strength was not available. These results demonstrate a first step towards developing a feedforward FES controller that could be used as part of a feedback system for restoring arm control or for a cooperative system combining FES with lightweight robotics for augmentation.