Inter-trial Dynamics in Goal-oriented Tasks with Asymmetric Error and Reduced Precision

open_access
Open Access
Author:
Mahoney, Joseph Michael
Graduate Program:
Engineering Science and Mechanics
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
August 12, 2013
Committee Members:
Joseph Paul Cusumano, Dissertation Advisor/Co-Advisor
Joseph Paul Cusumano, Committee Chair/Co-Chair
Gary L Gray, Committee Member
Corina Stefania Drapaca, Committee Member
Stephen Jacob Piazza, Committee Member
Xuemei Huang, Committee Member
Keywords:
human movement
goal equivalent manifold
virtual reality
shuffleboard
skilled tasks
task manifold
GEM
UCM
dynamics
Abstract:
Humans completing successive trials of a skilled task vary their body states (e.g. joint angles and velocities) from one trial to the next. These fluctuations in repeated performance contain information about the underlying control system. The reduction of goal-level error has been hypothesized as the driver of this control system. However, previous studies have shown that the empirical trial-to-trial behavior cannot be explained by error-correction alone. In this dissertation, we create a model for updating body states from trial to trial. A stochastic optimal controller determines the input for the next trial. On every step, the controller minimizes the expected value of a cost function. This control model is applied to a simple shuffleboard task. We implement a cost function that includes costs for the goal-level error, balancing overshooting and undershooting, and the range of motion. The selection of weights for each cost term determines the mean operating point and the dynamic stability. We construct a virtual shuffleboard game to collect data for two experiments. In the first experiment, we introduce a penalty for overshooting the target. We observe identical stability properties both when the overshooting penalty is present and when absent. However, the mean operating point is set back from the target when the penalty is applied. We estimate the weights in cost function that subjects apply in both conditions and see more emphasis on balancing overshooting and undershooting with the penalty present. We are able to recreate the empirically observed behaviors in simulation using the estimated weights in the model. In the second experiment, we reduce the required precision in the shuffleboard task by making the target area larger. We expect that control effort will be reduced since small body-level fluctuations from the mean remain within the target area. A slight reduction in control is observed, but we do not see a random walk around the target, which we would expect if error-correction were the only cost. Slight differences in the weights of the cost emerge as the target thickness changes. This new control model demonstrates the importance of competing costs in explaining the behavior of human subjects. The model can be adapted for other non-shuffleboard tasks to include task-specific terms. Additionally, comparing the weights between subjects may be useful in differentiating or diagnosing pathological patients.