Fundamental Visual Navigation Algorithms
Published in arxiv, 2024
PhD Thesis - Part One.
Talk @ ScioI Summer School 2023
Author: Patrick Govoni
Supervisor: Pawel Romanczuk
Affiliation: Institute for Theoretical Biology, Humboldt Universität zu Berlin
Group: Collective Information Processing Lab
Years: 2023-Present
Abstract:
Foraging in predictable environments requires coordinating effective movement with observable spatial context, i.e. navigation. Separate from search, navigation is controlled by two partially dissociable, concurrently developed systems in the brain. The cognitive map informs an organism of its location, bearing, and distances between environmental features, enabling shortcuts. Visual response-based navigation via routes, on the other hand, is commonly considered inflexible, ultimately subserving map-based representations. As such, navigation models widely assume the primacy of maps, constructed through predictive control and distance perception, while neglecting response- based strategies. Here we show the sufficiency of a minimal feedforward framework in a classic navigation task. Our agents, directly translating visual perception to movement, navigate to a hidden goal in an open field, an environment often assumed to require predictive map-based representations. While visual distance enables direct trajectories to the goal, two distinct algorithms develop to robustly navigate using visual angles alone. Each of the three confers unique tradeoffs as well as aligns with movement behavior observed in rodents, insects, fish, and sperm cells, suggesting the broad significance of response-based navigation throughout biology. We advocate further study of bottom-up, response-based navigation without assuming online access to computationally expensive distance perception, prediction, or maps, which may better explain behavior under energetic or attentional constraints.
Recommended citation: Govoni, P., Romanczuk, P. (2024). Fundamental Visual Navigation Algorithms: Indirect Sequential, Biased Diffusive, & Direct Pathing.
Download Paper