Modeling animal navigation in complex environments: learning, decision-making, and fluid flows

Abstract

This thesis presents a few developments in computational tools aiding the understanding of biological systems. We first introduce modeling of animal navigation in complex environments. Chapter 1 discusses how soaring birds extract energy from turbulent flows. We design an adaptive algorithm combined with Monte Carlo tree search and statistical inference methods to capture spatiotemporal information from the dynamics of flows and assist active glider control. We also figure out the physical constraints of fluids and aerodynamic parameters of gliders for feasible energy-positive flight in turbulence. Chapter 2 discusses olfactory navigation using a Reinforcement learning framework. We model mouse navigation in continuous action and state space with partial observability. We investigate the alternation of mice’s food-searching strategies and show that our model reproduces the behaviors found in experiments. Chapter 3 presents data-driven approaches for analyzing the morphology of insect wings. We introduce the digitized dataset of grasshopper Arphia Conspersa. We present the image analysis tools to extract features of wing shapes and venation patterns from raw images. We discuss the interpopulation and intrapopulation variations of venation patterns.