Automated robotic tracking of gelatinous animals in the deep ocean

Abstract

This dissertation introduces new technologies to enable autonomous robotic tracking of gelatinous animals in the deep ocean. The expense and challenge of accessing the deep ocean continue to limit long-duration observations of gelatinous marine animals. Behaviors of gelatinous species, collectively referred to here as jellies, are remarkably complex given their relatively simple physiology. Most jellies do not have a centralized nervous system, yet these animals demonstrate intricate foraging, migration, evasion and aggregation behaviors. New technologies are required to discover previously unobserved behaviors, to study behavioral distributions over time, and to investigate the links between behaviors and environmental stimuli. Automation addresses the two significant limitations associated with current-generation technologies for gelatinous-animal observation: human-pilot fatigue and infrastructure cost. To overcome these limitations, this dissertation develops a core automated jelly-tracking capability and demonstrates its successful deployment as a pilot-assist for a remotely-operated vehicle (ROV). Key challenges in the design of a reliable jellyfish-tracking robot include the development of visual sensing and automatic control components tailored to the application. A new vision design tool, called segmentation efficiency, is introduced to narrow the set of prospective vision-processing algorithms by analyzing a database of application-specific reference images. In the context of a jellyfish database, this performance assessment tool leads to the development of a gradient-based algorithm that reliably tracks a wide range of gelatinous animal species under varied lighting conditions. This dissertation also develops a vision-based control technique, specific to the jelly-tracking application. The control law balances the requirements to enforce the camera viewing-cone aggressively and to minimize overall vehicle thrust, which may artificially stimulate animal behavior. A series of dive experiments in the Monterey Bay validated the combined vision and control system under field conditions. These experiments resulted in repeated successful demonstrations of animal tracking for periods as long as 89 minutes. Additional theoretical work investigates the transition of demonstrated jelly-tracking technology for use with an energy-constrained autonomous underwater vehicle (AUV) platform. This work focuses on the use of a strobed lighting system to reduce the power budget and improve the data quality for tracking gelatinous animals over periods of 24 hours or more.