AI-REEFSHIELD
Al-driven Robotic Ecosystem Exploration Framework for Securing Habitat Integrity and Life Diversity
The AI-REEFSHIELD project aims to develop an AI-supported learning and robotics system for monitoring marine restoration areas, in particular for the reintroduction of the European oyster in the North Sea. An autonomous underwater vehicle (AUV) equipped with high-performance cameras and AI algorithms will be used to map large reef areas precisely, efficiently, and with less human intervention. This will replace the previous cost- and emission-intensive monitoring methods. The innovative approach includes an explainable active vision approach for targeted autonomous monitoring, combining machine vision, reinforcement learning, explainable AI (XAI), and model-based learning methods. In addition to environmental and climate protection goals, the project also aims to improve biodiversity monitoring. The system will be validated in the pilot area of the oyster reefs in the Borkum Riffgrund nature reserve and should be transferable to other environmental applications.
Project details
The AI-REEFSHIELD project aims to develop an AI-supported robotic monitoring system for marine ecosystems, particularly within restoration efforts such as the reintroduction of the European oyster (Ostrea edulis) in the North Sea. These oyster reefs are considered highly endangered and play a vital role as CO₂ reservoirs in shallow coastal waters. Their natural filtration improves water quality, making them an important component of ecosystem resilience and climate protection.
The project consortium includes the German Research Center for Artificial Intelligence (DFKI), the Alfred Wegener Institute (AWI), and the University of Bremen. At its core is the deployment of autonomous underwater vehicles (AUVs), equipped with high-performance sensors and AI algorithms. These AUVs will autonomously collect high-resolution image data, navigate underwater terrain, and analyze environmental conditions in real time to enable continuous monitoring of reef structures.
The control system will be built on a multi-task reinforcement learning framework, enhanced by model-based learning (Model-Based RL) and explainable AI (XAI). The collected image data will be annotated using citizen science platforms and published on open-access repositories.
System validation will take place in the pilot oyster reef field within the Borkum Riffgrund nature reserve. This solution is designed to replace costly and emission-heavy ship-based monitoring missions, enabling efficient and sustainable protection of sensitive marine habitats. The system is also easily transferable to other environmental applications, such as monitoring additional endangered ecosystems or restoration sites.