Can an Optimal Control Model Capture Tracking Kinematics? Insights from Healthy and Post-Stroke Robot Interactions

Kinematic deficits are an important indicator of motor and cognitive impairments in post-stroke and healthy older adults. We fit an optimal feedback control model to a robot-based pursuit tracking task to evaluate the kinematics of 6 healthy older adult and 3 post-stroke participants at two time points: baseline (pre-training) and assessment (post-training). We investigate how well the model captures healthy and stroke kinematic performance, predicts specific kinematic metrics of task performance, and reflects change due to motor learning. The preliminary optimal control model can accurately model healthy and post-stroke pursuit tracking behavior in the position and velocity domains and more accurately captured post-training behavior. Correlations in the assessment models (position median R = 0.89, velocity median R = 0.68) improved over baseline models (position median R = 0.56, velocity median R = 0.48). We also find significant increases in controller gains and decrease in positional noise between baseline and assessment models, which may reflect motor learning. Finally, we demonstrate moderate to strong linear relationships between modeled and actual results for three key kinematic metrics.