Invited Talk: Agnese SeminaraPhysics-informed reinforcement learning for animal navigation: strategies to navigate turbulence*

I will introduce turbulent navigation by discussing a set of behavioral experiments with sea robins, fishes with sensory appendages akin to legs that they use to dig prey from sand. Data suggest these animals sense chemicals both from the water column as well as on sand, an "alternating" behavior well documented in rodents and dogs, suggesting that animals integrate both fluid-borne and substrate cues into a multi-modal navigation strategy. I will then discuss an algorithm showing animals may alternate tasting surfaces and smelling fluids to improve search efficiency. Far from the target, the searcher is most likely to rely on bulk odors as the search is information limited. Agents explore cross wind to search for odor and then surge upwind when cast and surge provide the same benefits, a decision boundary that can be understood theoretically. In this model, agents guess where the source is and use odor to refine their guess -- but can animals respond solely to odor, without relying on a map of space? I will propose a second algorithm for turbulent navigation, where agents use temporal memory to measure salient features of odor time series. I will show that using the physics of odor cues, memory can be adjusted based on experience with no need of prior information. I will then discuss how agents can recognize and correct mistakes based on odor alone, suggesting accurate predictions can overcome intermittency, and enable navigation in the presence of turbulence.