Understanding ant route navigation as a situated and embodied system: Behavioural experiments, modelling and robotics

31. März 2011

Andy Philippides and Paul Graham
Centre of Computational Neuroscience and Robotics
University of Sussex, UK

Understanding ant route navigation as a situated and embodied system: Behavioural experiments, modelling and robotics.

Abstract: It is known that ants learn long visually guided routes through complex terrain. However, the mechanisms by which visual information is first learned and then used to control a route direction are not well understood, especially in natural complex environments. For a proper understanding of this (or any) behaviour it is essential to understand the constraints placed on behaviour by morphology, sensory systems and the information available in their natural habitat. Our work has been driven by trying to understand those constraints.
    Useful information for visual navigation should be salient, easy to extract and stable over time. We have shown through behavioural studies and modelling that the 1D skyline profile between terrestrial objects and the sky is sufficient for ants to set a direction and contains enough information to guide a route. During these, and other, experiments we observed that when ants are choosing a direction they scan the world by pausing and rotating on the spot. The scanning is saccadic in nature and ants' fixations are drawn by familiar portions of the panorama.
Following these results, we propose a parsimonious algorithm for visually guided route following in which agents scan the environment and move in the direction that appears most familiar. As ants’ movement and viewing directions are coupled, a familiar view specifies a familiar movement to make. View familiarity can be implemented in a number of ways. Here we train a classifier to determine whether a given view is part of a route and use the confidence in the classification as a proxy for familiarity. We show the feasibility of our approach as a model of ant-like route acquisition by learning a series of nontrivial routes through an indoor environment using a large gantry robot equipped with a panoramic camera.