Modeling attention for multimodal navigation with magnetic signatures.

Many migratory animals sense Earth's magnetic field and use it for navigation, alongside stimuli in other modalities. Engineered systems could benefit from similar approaches, especially where satellite signals are unavailable. However, it is unknown how animals allocate attention between magnetic and non- magnetic stimuli. One theory is that attention arises from competition between stimuli, bottom-up biases towards salient (noticeable) stimuli, and top-down biases towards goal-relevant stimuli. We develop an attention model based on biased competition for a simulated agent that navigates between known waypoints using magnetic, olfactory, and visual cues. We characterize how attentional biases influence the agent's success rate, speed, and path efficiency during migrations in a diverse set of abstract magnetic environments, with and without fluid currents. We also explore how attention biases influence obstacle avoidance. Our work suggests that biased competition is a plausible framework for multimodal navigation in magnetoreceptive animals or engineered systems. Bottom-up biases are crucial to obstacle detection, and support navigation near waypoints. While top-down biases lead to more active navigation and improve speed, they also hinder obstacle detection. Because we did not optimize parameters, combining top-down and bottom-up biases did not lead to both efficient navigation and obstacle avoidance.