Customized Robotic Training Approaches Using the Statistics of Reaching Errors

While upper extremity training with haptic and visual feedback has been shown to assist in restoring function for individuals with stroke and related brain injuries (Teasell, Foley, Bhogal, & Speechley, 2003), outcomes vary greatly amongst individuals. Recent studies have found that manipulating error signals during training can stimulate learning. In order to improve current methods, we believe that it is necessary to customize haptic and visual interactions to address individual motor impairments. We can customize training using the statistics of errors, intervening only on the most commonly occurring errors. In addition to rehabilitation, this technique can be applied to situations where error feedback is needed, such as learning new skills. The primary goal of this thesis is to develop better training interventions to facilitate motor learning. Addressing the need for training tools for both skill learning and therapy, we explore strategies for improving upon current training paradigms. In the first study we show how force adaptation can cause participants to reach with the errors that are necessary for moving in a visually rotated scene. In the second study, we determine the best domain to represent error tendencies during learning. In the third study, we test how error statistics can enhance learning of a novel visual transformation. In the fourth study, we conduct a preliminary investigation using error statistics to customize training interventions for stroke survivors. Our results will contribute a basic understanding on how we can use error statistics to improve training environments and effect functional recovery.