Smart Robotic Inspection of Levees and Culverts: A Scoping Review of Robotic Platforms, Sensing Technologies, AI Methods, and Multi-Robot Coordination for Flood-Control Infrastructure
This scoping review synthesizes 64 empirical studies to map the current landscape of robotic platforms, sensing technologies, and AI methods for inspecting flood-control infrastructure, highlighting the high accuracy of isolated components while identifying a critical gap in integrated, multi-robot systems and proposing a six-tier reference architecture and research agenda to bridge this divide.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The earthworks that hold back rivers and protect cities from flooding are among the most critical, yet most vulnerable, pieces of modern infrastructure. Levees, dikes, and the underground pipes that carry stormwater away are constantly under siege from water pressure, soil erosion, and hidden decay. For decades, keeping these systems safe has relied on human inspectors walking along muddy banks or crawling through dark, wet tunnels with cameras on ropes. This method is dangerous, slow, and often misses the early warning signs of a disaster, such as water seeping through the soil before a breach occurs. To solve this, engineers have begun turning to machines: drones that fly overhead, robots that crawl inside pipes, and artificial intelligence that can spot a crack in a wall faster and more accurately than a tired human eye. The goal is to replace risky, guesswork-heavy patrols with a continuous, automated watch over the very ground that keeps our communities dry.
A comprehensive review of recent research reveals that while individual technologies for inspecting these flood defenses are becoming highly advanced, they are still largely working in isolation. The study, which analyzed 64 scientific papers published between 2010 and 2026, found that we have excellent tools for specific tasks. Drones equipped with thermal cameras can fly over miles of levees and detect subtle temperature differences that reveal hidden water leaks. Inside narrow, dark pipes, wheeled robots can navigate tight spaces and use high-definition cameras to find cracks or blockages. Artificial intelligence algorithms have been trained to recognize these defects automatically, identifying issues like rust, broken joints, or sediment buildup with high accuracy. However, the review highlights a significant gap: these technologies rarely work together as a single, coordinated system. Most research focuses on either the drone or the robot, but not on how they can communicate and share data to give a complete picture of a flood-control network.
The researchers discovered that the biggest hurdles are not just about building better robots, but about making them work in the harsh, real-world conditions of flood infrastructure. One major challenge is communication. When a robot enters a long, concrete pipe, the signal from a remote control or a drone often disappears within a few dozen meters because the thick walls and standing water block radio waves. Without a way to stay connected, the robot cannot send back its findings or receive new instructions. Another difficulty is navigation. Inside a smooth, featureless pipe or under a dense canopy of trees, the robots cannot rely on satellite signals to know where they are. They must instead use their own sensors to map the environment as they move, a process that can become confusing if the surroundings look the same in every direction. Furthermore, the data these robots collect is often trapped in separate systems. A drone might find a problem on the surface, while a robot finds a different issue underground, but there is no automatic way to link these two discoveries to a central map that city planners can use to decide which repairs are most urgent.
To address these fragmentation issues, the authors propose a new way of thinking about flood inspection that treats it as a unified system rather than a collection of separate tools. They suggest a six-step process where different types of robots work together in a specific sequence. First, drones would fly over a wide area to spot potential trouble spots from above. If a drone sees something suspicious, like a wet patch on a levee or a blocked pipe entrance, it would automatically dispatch a ground robot or a swimming robot to investigate that specific spot up close. As the ground robot moves deeper into a pipe where signals are lost, other robots would act as floating or stationary bridges to keep the communication line open. The data from all these different sources—images from above, 3D maps from the side, and sonar scans from below—would be combined into a single, detailed digital model of the infrastructure. This model would then be linked to city maintenance records, allowing engineers to see exactly where a problem is, how bad it is, and how quickly it needs to be fixed.
The study concludes that while the individual pieces of this puzzle are ready, the full picture is not yet complete. Only a small fraction of the research reviewed has tested these different types of robots working together in real flood-control environments. Most studies remain in laboratories or use simulations. The authors argue that the next step for the field is to move beyond testing single robots in isolation and focus on building these coordinated teams that can operate safely in wet, muddy, and signal-dead zones. By solving the problems of communication, navigation, and data integration, we can transform flood management from a reactive process of fixing broken walls after a disaster into a proactive system that prevents disasters before they happen. This shift would not only save money and time but, more importantly, would remove the need for humans to risk their lives in dangerous, unstable environments to keep our cities safe.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.