To ensure a consistent and reliable mathematical representation of human motor performance and learning we provide a framework based on existing modeling methodologies. The framework uses structured and unstructured linear time invariant control models, to describe human motor behavior for individual trials of a compensatory tracking experiment. Control-theoretic metrics are used to quantify performance and learning of participant behavior across trials. Unstructured models, such as subspace identification, were determined to provided excellent prediction of human performance. Feedback learning was associated with increased weighted coherence across trials, during our characterization for trial-by-trial learning. The utility of this framework provides a means for model selection by providing information performance and learning results for various commonly used methods.

Additionally we investigate the influence of input signal attributes, frequency and complexity, on human performance and learning using m-level pseudo-random signals (m-level PRS) during a compensatory tracking task with position or velocity control. Our novel experimental design reveals that there exists a set of input signals, in particular, Low height High frequency (LhHf), that produce enhanced human performance and learning. We show the benefits of creating a model for each trial instead of trial averaging, ultimately demonstrating that reliable subspace models can be created. We identify feedback model parameters that quantify performance and learning effects across trials.

Finally we present a novel method for separating human motor performance brain and body related dynamics into two separate models by combining imaging and input-output behavioral tracking data using subspace identification. Position and velocity control based human performance models were constructed from a compensatory tracking task with a learning induced feedback criteria using a set of pseudo-random signals (PRS) inputs. Standard fMRI analysis coupled with independent component analysis (ICA) revealed the activated brain regions most related to the tracking task. Imaging data from the activated regions are used as outputs and inputs with the tracking data for both the brain and body subspace models, respectively. We compare performance for conventional human motor subspace models with our novel two part brain-body model.