As an intelligent species, bottlenose dolphins (Tursiops truncatus) demonstrate sophisticated coordinated behavior. This behavior depends on each individual’s temporal behavior profile and the spatial locations of the individuals over time. An individual animal’s decision-making in the environment is also influenced by its own dynamics and energy cost. As a result, the study of such a coordinated system requires a framework to synthesize multiple streams of information with appropriate models for extended periods. However, tools to investigate the fine-scale dynamical behavior of multiple animals during coordinated tasks are lacking. To address this gap, we present Intelligent Analysis (IA) methods that leverage experimental data to study the coordinated behavior of bottlenose dolphins and investigate how energetic cost may influence behavior.

Specifically, this dissertation addresses the following research questions:​ Q1:​ Can we develop a data-driven analysis framework that enhances dynamical behavior characterization and classification while minimizing the need for human input? Q2:​ Can existing perception and localization tools be extended to a framework for tracking multiple uncontrollable agents within a dynamic underwater environment with quantified uncertainties? Q3:​ What are the energetic trade-offs of different swimming gaits measured using a biologging tag (wearable sensors for animals)? Q4:​ How do varying environmental conditions or social coordination result in changes in the behavior and swimming gait of an animal?

This work is at the intersection of machine learning, computer vision, sensor fusion, datadriven modeling, animal behavior, and biomechanics. The AI-driven analysis framework developed in this work was used with data from animal-borne biologging tags and cameras placed in the environment to derive interpretable information about an animal’s dynamical state (e.g. gait, pose, position in the environment, and energetics). Knowledge of these dynamics facilitated the investigation of animal behavior during interactions with conspecifics and environmental features. The resulting identity preserving hour-scale tracks for each individual, together with their dynamic profiles, demonstrate that individuals respond to the same environmental features very differently and that coordinated events significantly modify an individual’s behavior. Research outcomes from this work contribute to an improved understanding of bottlenose dolphin behavior and biomechanics and can be applied to biologging tag data from other cetaceans both in managed settings and the wild. Further, the algorithms and framework developed here can be extended for applications ranging from state estimation in vehicles to multi-target tracking in robotics.