Meeting Abstract
Visual navigation in ants has been studied extensively in a variety of species. Landmark features that ants relay on to define their routes have been identified as well as how they fixate on learned cues during their return approaches. We created a computational model to study ant navigation in a procedurally generated environment. The model of foraging is based on behaviors of live ants and aims to examine a random foraging event that results in the simulated ant finding a goal and the subsequent walks to that learned goal. Using this model, we explored how memory is modified with experience, if navigation is robust regardless of the direction of the initial path, and how disruptions in the environment are predicted to affect visual navigation in live ants. Our model provides insight into processing and learning of visual information, supports that ants prioritize relevant features and do not need to constantly process information from their environment, and that routes converge on the same goal while being idiosyncratic. To study the response to changes in the environment, we tested the model in two ways: (1) remove specific portions of objects from the rendered environment after training and determine how it affects subsequent foraging; (2) add uneven terrains with random hills and valleys to the simulated environment and examine how that disrupts the model. The model predicts that features that are enroute minimally disrupt navigation when removed whereas features that are far away cause greater disruption. Finally, the model predicts the necessity of head stabilization and terrain-angle specific visual sampling to improve foraging success.
