Abstract
In natural behavior, visual information is actively sampled from the environment by a sequence of gaze changes. The timing and choice of gaze targets, and the accompanying attentional shifts, are intimately linked with ongoing behavior. Nonetheless, modeling of the deployment of these fixations has been very difficult because they depend on characterizing the underlying task structure. Recently, advances in eye tracking during natural vision, together with the development of probabilistic modeling techniques, have provided insight into how the cognitive agenda might be included in the specification of fixations. These techniques take advantage of the decomposition of complex behaviors into modular components. A particular subset of these models casts the role of fixation as that of providing task-relevant information that is rewarding to the agent, with fixation being selected on the basis of expected reward and uncertainty about environmental state. In a previous study gaze behavior in a driving task where subjects followed a lead car and maintained a given speed, we showed that specific values for each subtasks's reward and uncertainty allowed the distribution of a human subject's fixation intervals of each of the subtasks to be approximated very closely as measured by KL divergence between human and model probability distributions (average value 0.79). The current work extends this result by showing that the ratio of measured uncertainties in the human data closely approximate the ratio used in the model, implying that both the uncertainty and reward parameters used by the model are commensurate with those used by the human subject.
Meeting abstract presented at VSS 2016