Skip to main content

Research

Safe navigation and control

Autonomous vehicles that operate close to people need navigation and control that are both precise and safe. In Marinarium’s controlled and fully observable basin, we develop and test how underwater vehicles plan and carry out manoeuvres around divers and other obstacles, and capture each run from multiple viewpoints.

An AUV manoeuvring around a diver, seen from the diver’s camera (left) and a fixed camera in the basin (right).

Diver-robot interaction

Divers and underwater robots have complementary strengths: robots can stay down longer and carry sensors and tools, while divers bring judgement, dexterity and fast decision-making. Getting them to work safely side by side requires experiments with real divers and real vehicles in a controlled, observable setting, something most robotics test tanks cannot offer, since few are built or maintained for human diving.

Marinarium was designed for diver-in-the-loop experiments, with the size, water quality, access and supervision needed for divers to work alongside vehicles in the basin. In the SHARCEX project, we use it to study diver–robot collaboration, with the aim of reducing risk and extending what people can do underwater.

A diver communicating with an underwater robot through hand signals in the Marinarium basin.

Cross-domain marine robotics

Maritime operations increasingly rely on teams of robots that work across domains: underwater, at the surface and in the air. Testing such heterogeneous fleets at sea is costly and difficult to observe. With motion capture both above and below the water, and a ground station that links the vehicles to remote operators over WiFi, 5G and acoustic communication, Marinarium lets us run complete multi-domain missions in a controlled setting.

In one such mission, an off-site operator commands an AUV and a drone to rendezvous with a surface vehicle: the drone lands on the USV while the AUV dives and docks underneath it.

Surface and underwater robots operating together in the Marinarium basin.

Underwater sim2real bridging

Simulators still fall short of capturing real hydrodynamics, so controllers and learning-based policies developed in simulation often break down when deployed on real underwater vehicles. Residual dynamics models offer a practical remedy: they learn the difference between simulated and real vehicle motion and feed it back into the simulator to make it more faithful. Learning them, however, requires accurate real-world motion data paired with the same vehicle running in simulation, something that has rarely been available underwater.

At Marinarium, the underwater motion-capture system and the facility’s digital twin provide exactly this pairing. Robots in the basin are tracked with high accuracy and mirrored in real time in the simulator, giving us the data needed to learn residual models that close the gap between underwater simulation and reality.

A BlueROV2 in the basin (centre), tracked by the motion-capture system (left) and mirrored by its digital twin in the simulator (right).

Underwater validation of spacecraft autonomy

Neutrally buoyant underwater vehicles behave much like free-flying spacecraft: they can move freely in all six degrees of freedom without feeling their weight, similar in spirit to the neutral-buoyancy pools used for astronaut training. Hydrodynamic effects are the main departure from true microgravity, which makes underwater robots a practical and accessible testbed for spacecraft guidance and control.

Through its direct connection to the co-located space robotics laboratory, Marinarium lets us run the same autonomous task, with an identical software and control stack, on underwater vehicles and on space-robot platforms, and assess how well spacecraft control strategies transfer across domains.

The same autonomous inspection task executed by a BlueROV2 in the Marinarium basin (top) and by a simulated CubeSat in orbit (bottom).