CSIRO outlined a robotic seagrass-monitoring trial with CQUniversity on 13 October 2026, using a surface vessel that tows an underwater camera to record coastal habitats. The research update describes a way to collect seabed imagery in difficult conditions, with a final assessment of performance expected later this year.
The camera operates close to the seabed while the uncrewed vessel travels above it. Images are linked to locations, processed on the surface platform and supported by AI-assisted analysis, with results communicated by radio. The partners are evaluating whether that arrangement can make surveys faster, safer and less expensive. Those are the questions being tested, rather than benefits already quantified by a completed evaluation.
The approach could also connect different kinds of environmental observations. CQUniversity’s researchers describe using seabed imagery to check what aerial surveys show in intertidal areas, potentially reducing the need to walk through the habitat. CSIRO engineer Dr Torsten Merz said in the release that the work could “help to accelerate seagrass monitoring and assessment”. The project forms part of the wider SeaGrow restoration effort.
The ecological reason for improving those surveys goes beyond making more detailed maps. The Reef Foundation describes seagrass meadows as habitats that stabilise sediment, store carbon and support animals including turtles and dugongs. Its restoration reporting identifies murky water, currents and silty seabeds as practical obstacles. Better observations could help researchers compare where vegetation persists with where restoration is being attempted.
The AI component will still need evidence that its classifications match conditions in the water. An earlier CSIRO project, DeepSeagrass, published images and methods for classifying seagrass from nine Moreton Bay beds surveyed in 2020. That is useful background on automated image analysis, but it is a separate project and does not establish which model the current trial uses or how accurately that model performs.
The distinction matters because collecting a photograph and interpreting it are different tasks. A successful survey needs usable images, reliable location information and a defensible assessment of what those images contain. For the current trial, useful published results would therefore include coverage, image quality and checks against specialist assessments, alongside time and cost. A larger image collection alone would not demonstrate better ecological decisions.
The mapping work also sits alongside a separate effort to automate planting. CQUniversity’s restoration programme describes trials involving Ulysses’ Mako underwater robot and collaboration with Gidarjil Development Corporation. Planting seeds and surveying their eventual surroundings address different stages of restoration. Footage of Mako should therefore be identified as related background, rather than presented as the camera-towing system announced this month.
The broader research setting links technical development with coastal users. CQUniversity’s marine centre works across applied coastal research, industry and community needs in the southern Great Barrier Reef region. For restoration teams, the value of repeatable monitoring would be knowing whether a site is changing and where to investigate further. That practical purpose is more important than autonomy as a feature in itself.
The final assessment will determine how far the current system can support that work. Until the partners publish their findings, the supported conclusion is that robotic imaging is being tested as an additional monitoring tool. Evidence about repeatability, operating conditions and the effort needed to verify its output will be central to deciding where it belongs in future Australian coastal surveys.
