Optimality, Objectives, and Trade-Offs in Motor Control under Uncertainty
Date
2023-09-22
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Biological motor control involves multiple objectives and constraints. In this thesis, I investigated the influence of uncertainty on biological sensorimotor control and decision-making, considering various objectives. In the first study, I used a simple biped walking model simulation to study the control of a rhythmic movement under uncertainty. Uncertainty necessitates a more sophisticated form of motor control involving internal model and sensing, and their effective integration. The optimality of the neural pattern generator incorporating sensory information was shown to be dependent on the relative amount of physical disturbance and sensor noise. When the controller was optimized for state estimation, other objectives of improved energy efficiency, reduced variability, and reduced number of falls were also satisfied. In the second study, human participants performed regression and classification tasks on visually presented scatterplot data. The tasks involved a trade-off between acting on small but prevalent errors and acting on big but scarce errors. We used inverse optimization to characterize the loss function used by humans in these regression and classification tasks, and found that these loss functions change systematically as the data sparsity changed. Despite being highly variable, there were overall shifts towards compensating for prevalent small errors more when the sparsity of the visual data decreased. In the third study, I extended the pattern recognition tasks to include visually mediated force tracking. When participants tracked force targets with visual noise, we observed a slight yet consistent force tracking bias. This bias, which increased with noise, was not explained by commonly hypothesized objectives such as a tendency to reduce effort while regulating error. Additional experiments revealed that a model balancing error reduction and transition reduction tendencies effectively explained and predicted experimental data. Transition reduction tendency was further separated into recency bias and central tendency bias. Notably, this bias disappeared when the task became purely visual, suggesting that such biases could be task-dependent. These findings across the three studies provide useful insights into understanding how uncertainty changes objectives and their trade-offs in biological motor control, and in turn, results in a different control strategy and behaviors.
Description
Keywords
motor control, sensorimotor control, optimization, inverse optimization, perception, decision-making
Citation
Ryu, H. (2023). Optimality, objectives, and trade-offs in motor control under uncertainty (Doctoral thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.