Algorithmic aspects of animal behavior: thermal soaring and odor trail tracking

Computational and Theoretical Neuroscience group welcomes Gautam Reddy, PhD, postdoctoral fellow at Harvard's NSF-Simons Center. For connection info, please email d.shipman@duke.edu. Abstract: In this talk, I will discuss two commonly observed yet poorly understood phenomena involving complex decision-making: 1) thermal soaring, where birds and gliders strategically use convective currents (thermals) in the atmosphere to gain height, and 2) odor trail tracking, where terrestrial animals exploit surface-bound scent trails to establish navigation routes, find mates or hunt prey. I will first introduce the natural phenomenology along with the physical challenges and constraints faced by soaring birds. Soaring strategies obtained from reinforcement learning (RL) highlight the general navigational principles and the sensory-motor cues necessary for effective soaring. I will then present our field experiments on training an autonomous glider to soar in the atmosphere using RL. For the second part of the talk, I will present recent results on odor trail tracking. Tracking behavior features zig-zagging paths with animals often staying in close contact with the trail. Upon sustained loss of contact, animals execute a sequence of "casts" - wide oscillations similar to plume-tracking insects. We propose a new hypothesis, where tracking animals continuously estimate the local heading of the trail and its uncertainty, which defines an angular sector. RL trajectories under this 'sector se