MBZUAI and Khalifa University Researchers Win Award for H-SURF Deep Learning Robotic Fish Swarm System

Award-Winning Breakthrough: H-SURF Ushers in a New Era of Underwater Robotics

The Sheikh Hamdan bin Zayed Award for Environmental Research has spotlighted an innovative project from Abu Dhabi's leading academic institutions: the Heterogeneous Swarm of Underwater Robotic Fish, or H-SURF. Announced on March 4, 2026, this second-place honor recognizes the collaborative efforts of researchers from Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), Khalifa University, and Sorbonne University Abu Dhabi. Led by Professor Cesare Stefanini of MBZUAI, Associate Professor Federico Renda of Khalifa University, and Dr. Giulia De Masi, the H-SURF system represents a leap forward in bio-inspired robotics, blending deep learning with swarm intelligence to tackle marine monitoring challenges.

Funded by the Technology Innovation Institute (TII) and developed at Khalifa University's Center for Autonomous Robotic Systems (KU-CARS) from November 2020 to January 2025, H-SURF addresses critical limitations in underwater operations. Traditional remotely operated vehicles (ROVs) and autonomous underwater vehicles (AUVs) struggle with poor communication, localization inaccuracies due to water's opacity, and high noise from propellers that disturbs marine life. H-SURF's fish-like robots swim silently, communicate via light pulses, and collaborate autonomously, making them ideal for sensitive ecosystems like UAE's coastal waters.

Bio-Inspired Design: Mimicking Nature for Efficiency and Stealth

H-SURF's core innovation lies in its biomimicry. Each robot resembles a real fish, featuring a streamlined body, flexible fins, and rear propellers that generate smooth, undulating propulsion. This design minimizes hydrodynamic drag and noise, allowing the swarm to blend into marine environments without alerting wildlife. The fleet comprises 30 robots: five 'leader' units with advanced sensors for remote oversight and 25 'follower' robots optimized for autonomy and inter-swarm communication.

Supported by a surface 'floater' for surface-to-land links and a submersible 'sinker' for telepresence, the system enables hybrid operations. Deployment involves releasing the swarm into target areas, where they dive, navigate obstacles, and surface as needed—all while consuming less energy than conventional thruster-based AUVs. In UAE waters, rich in mangroves and coral reefs vital for biodiversity and coastal protection, this stealthy approach is transformative.

The UAE's marine ecosystem, spanning the Arabian Gulf and Gulf of Oman, supports over 1,000 fish species and critical habitats threatened by oil activities, warming waters, and development. H-SURF's low-impact profile supports the UAE's Blue Economy goals, aligning with national strategies for sustainable ocean resource management.

Deep Learning at the Core: Real-Time Species Identification Onboard

At H-SURF's heart is embedded deep learning for computer vision. Each robot mounts front- and side-facing cameras processed by a lightweight Convolutional Neural Network (CNN) based on YOLO11n, running on a Raspberry Pi. This edge-computing setup identifies marine species in under 7 milliseconds—0.9ms preprocessing, 4.6ms inference, 1.4ms postprocessing—without cloud reliance.

Trained on datasets like Roboflow Aquarium and Google Open Images V7, augmented with UAE-specific images from aquariums and the UAE Dolphin Project, the model classifies sharks, rays, jellyfish, dolphins, belugas, and generic fish with 0.78 mean average precision (mAP). Annotation used Roboflow and Makesense tools, ensuring accuracy across 2,000+ images (82% train, 12% val, 6% test). This enables real-time biodiversity mapping, behavior tracking, anomaly detection, and population estimates, crucial for UAE's conservation efforts amid climate pressures.

Deep learning here exemplifies edge AI: models optimized for low-power hardware, reducing latency and bandwidth needs in communication-scarce underwater realms. Professor Stefanini notes, "Even with simple onboard code, the swarm carries out complex tasks like one large robot." Future expansions target more species and behaviors, enhancing precision for fishery management.

Swarm Intelligence and Multi-Agent Reinforcement Learning

H-SURF leverages swarm algorithms for decentralized control. Robots share relative positions via light signals, forming formations and executing tasks like pipeline inspection faster than solo units. Multi-Agent Reinforcement Learning (MARL) trains agents to collaborate in low-visibility, following hidden pipes by pooling neighbor data.

• Decentralized Decision-Making: Each fish acts on local info, scaling to large swarms.
• Task Allocation: Leaders coordinate, followers execute, optimizing energy.
• Resilience: Loss of one robot doesn't halt the mission; swarm adapts.

In simulations and UAE coastal tests, H-SURF demonstrated formation control with minimal boundary deviations, even as agent numbers grew. This behavior-based approach, inspired by fish schools, ensures robustness in turbulent waters.

Real-World Deployments: From Labs to UAE Coasts

Tested in Khalifa University's pools, open UAE waters, and simulations, H-SURF excelled in navigation, obstacle avoidance, and data collection. Recent UAE-Japan Mangrove Initiative saw H-SURF monitor sites pre-planting, assessing water quality, habitats, and visuals for optimal outcomes. Learn more about the collaboration.

Applications span offshore oil rig inspections (UAE's key industry), wreck searches, environmental surveys, and coral health checks amid bleaching threats. By reducing diver risks and costs, H-SURF supports Abu Dhabi's marine labs and EAD goals.

Environmental Sustainability and UAE Context

UAE's 1,300km coastline hosts vital mangroves (carbon sinks) and corals, but faces pollution, warming (Gulf temps up 1°C/decade), and development. H-SURF's silent ops preserve these, aiding net-zero ambitions. Light comms avoid acoustic harm to dolphins; bio-propulsion cuts fuel use.

The award underscores UAE's research prowess, with Khalifa and MBZUAI exemplifying Vision 2031's innovation focus. Funded by TII, it boosts national R&D, creating jobs in robotics/AI.

Challenges Overcome and Technical Hurdles

Key challenges: Energy (4-5hr limit, AI drains batteries), dataset scarcity (UAE-specific marine data), comms in turbid water. Solutions: Efficient CNNs, RL for low-data learning, hybrid floater-sinker architecture. Future: Solar charging, larger datasets via crowdsourcing.

Step-by-step CNN process: Image capture → Preprocess (resize/normalize) → YOLO inference → Postprocess (NMS bounding boxes) → Species label/output to swarm controller.

Collaborations and Path Forward

Ongoing ties with UTokyo for mangroves, potential PhD at MBZUAI. Expansions: More species (UAE turtles, dugongs), behaviors (schooling analysis), integrations (drones for aerial-aquatic swarms). Commercialization via TII for fisheries/oil sectors.

Read the original H-SURF paper for technical depth.

Implications for UAE Higher Education and Research

This accolade elevates UAE unis globally, fostering AI-robotics talent. MBZUAI's focus on deep learning, Khalifa's engineering, exemplify interdisciplinary synergy. Attracts students/faculty, aligns with scholarships/research jobs. Positions UAE as marine tech hub, aiding Blue Economy (projected $115B by 2030).

For academics, H-SURF inspires curricula in swarm AI, edge computing. Explore UAE research positions.

Photo by Harshith Raju on Unsplash

Global Ripple Effects and Future Outlook

Beyond UAE, H-SURF models scalable ocean monitoring worldwide, from Arctic to reefs. Addresses UN SDGs 14 (Life Below Water). With climate threats rising, such tech is vital. UAE leads via awards like Sheikh Hamdan, spurring innovation.

Stefanini: "To interact with life, understand it—without noise." H-SURF embodies this, promising quieter, smarter seas.