AI Designs Underwater Gliders for Ocean Exploration

Marine animals' efficient swimming has long been a source of fascination for scientists. Their bodies are optimized for hydrodynamic aquatic navigation, allowing them to travel long distances while using minimal energy.

Autonomous vehicles can also drift through the ocean, gathering data about underwater environments. However, these machines often have less diverse shapes than marine life, typically resembling tubes or torpedoes. Furthermore, testing new designs involves significant real-world experimentation.

Researchers from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) and the University of Wisconsin at Madison suggest that AI could facilitate the exploration of new glider designs. Their method utilizes machine learning to evaluate 3D designs in a physics simulator and refine them into more hydrodynamic shapes. The resulting model can be 3D-printed, using less energy than manual production.

AI-Powered Design Pipeline

MIT scientists believe this design pipeline could lead to the creation of more efficient machines that assist oceanographers in measuring water temperature and salt levels, gaining insights into currents, and monitoring climate change impacts. The team demonstrated this by creating two boogie board-sized gliders: a two-winged machine similar to an airplane and a four-winged object resembling a flat fish with fins.

Peter Yichen Chen, MIT CSAIL postdoc and co-lead researcher, emphasizes that these designs represent only a fraction of the novel shapes his team's approach can generate. He notes the development of a semi-automated process for testing unconventional designs that would be difficult for humans to create. This level of shape diversity has not been previously explored, so most of these designs have not been tested in real-world conditions.

How the AI Works

The researchers began by gathering 3D models of over 20 conventional sea exploration shapes, including submarines, whales, manta rays, and sharks. They then enclosed these models in deformation cages, mapping articulation points that could be manipulated to create new shapes. The team created a dataset of conventional and deformed shapes before simulating their performance at different angles of attack. These shapes and angles were used as inputs for a neural network that predicts and optimizes the efficiency of a glider shape at specific angles.

Optimizing Glider Performance

The team's neural network simulates how a glider would respond to underwater physics, capturing its forward movement and the drag force against it. The objective is to find the optimal lift-to-drag ratio, which indicates how well the glider is supported compared to how much it is being slowed down. A higher ratio means more efficient travel, while a lower ratio means the glider will slow down more quickly.

Niklas Hagemann, an MIT graduate student in architecture and CSAIL affiliate, notes that this ratio is useful for achieving similar gliding motion in the ocean. According to Hagemann, the AI pipeline modifies glider shapes to optimize their underwater performance by finding the best lift-to-drag ratio, which can then be exported for 3D printing.

Validating AI Predictions

To validate the AI pipeline's predictions, the researchers fabricated their two-wing design as a scaled-down model. The glider's predicted lift-to-drag ratio was only about 5 percent higher on average than the ratios recorded in wind tunnel experiments.

A digital evaluation using a physics simulator also indicated that the AI pipeline accurately predicted glider movements. To further evaluate the gliders, the team 3D-printed two designs that performed best at specific points of attack: a jet-like device at 9 degrees and the four-wing vehicle at 30 degrees. The designs were fabricated as hollow shells with small holes for flooding upon submersion, making them lightweight and requiring less material.

The researchers placed a tube-like device inside the shells, housing a pump for buoyancy control, a mass shifter for angle-of-attack control, and electronic components. Each design outperformed a handmade torpedo-shaped glider in a pool, demonstrating higher lift-to-drag ratios and reduced energy exertion, similar to the effortless navigation of marine animals.

The researchers aim to reduce the difference between simulation and real-world performance. They also intend to create machines that can adapt to changing currents, making them more versatile in oceans. Chen added that the team is exploring thinner glider designs and intends to enhance their framework for greater customization, maneuverability, and the creation of miniature vehicles.