Visual Search and Target Selection Using a Bounded Optimal Model of State Estimation & Control

Visual attention and motor control are tightly coupled in domains requiring a human operator to interact with a visual interface. Here, we integrate a boundedly optimal visual attention model with two separate motor control models and compare the predictions made by these models against perceptual and motor data collected from human subjects engaged in a parafoveal detection task. The results indicate that humans use an optimal motor control policy limited by precision constraints – humans executed ballistic movements using near-optimal velocity (i.e., bang-bang control), but imprecision in those movements often caused participants to overshoot their targets, necessitating corrective action. Motor movements did not reflect response hedging, but rather a perceptual-motor policy permitting ballistic movements to a target only after localization confidence exceeded a threshold. We conclude that a boundedly-optimal perceptual-motor model can predict aspects of human performance visual search tasks requiring motor response