Engineers and scientists have spent two years developing AI-driven underwater vehicles capable of autonomously finding, following, and identifying deep-sea animals in real time. This Deployable AI aims to enhance our understanding of ocean life, with implications for future astrobiology missions to ocean worlds.
Overview
For the past two years, engineers and scientists have been developing a deployable artificial intelligence (AI) solution for underwater vehicles. This technology enables autonomous discovery, tracking, and identification of deep-sea animals with minimal human intervention. The goal is to accelerate scientific understanding of ocean life, a capability particularly relevant for future astrobiology missions to explore extraterrestrial ocean worlds like Enceladus and Europa.
How It Works
The Deployable AI system integrates specialized hardware, including cameras and a compact computer, with advanced software. This software incorporates 'detector' and 'supervisor' algorithms that process live video feeds to recognize predefined marine life such as fish, jellyfish, siphonophores, and comb jellies. Upon detecting an animal of interest, an 'agent' algorithm takes control, working with the vehicle's navigation system to meticulously follow the creature at a safe distance, ensuring continuous imaging without disturbance.
Initial Testing
Before real-world deployment, the AI agent underwent extensive training in a simulated environment, akin to a video game, using a Remotely Operated Vehicle (ROV) and virtual animals. Once its performance met the team's criteria, the AI was integrated into MBARI's MiniROV. Further refinement took place in a 10-meter-deep test tank, where the MiniROV was successfully taught to track an artificial jellyfish mimic, validating its foundational tracking capabilities.
Next Stop, the Ocean
In October 2024, the Deployable AI was ready for its first open-ocean trial in Monterey Bay. Operating from the Research Vessel Rachel Carson, the MiniROV, equipped with the AI agent, demonstrated considerable success in finding and following various deep-sea organisms, including siphonophores, comb jellies, and jellyfish. While showing promising results, the initial field test also highlighted areas for further improvement and refinement in the AI's autonomous navigation and identification processes.
