MIT Researchers Use AI to Design More Efficient Autonomous Underwater Gliders

Researchers from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the University of Wisconsin at Madison have developed an AI-driven method to design more efficient and diverse autonomous underwater gliders. These gliders can collect crucial data about oceanic environments, helping oceanographers measure factors like water temperature and salinity, study currents, and monitor climate change impacts. Traditionally, underwater gliders are designed to resemble tubes or torpedoes due to their hydrodynamic properties. However, this approach limits the potential for innovation and energy efficiency. The new method uses machine learning to simulate and optimize various 3D designs, allowing for the creation of unconventional shapes that outperform traditional gliders. The process begins by collecting 3D models of common underwater vehicles and marine animals, such as submarines, whales, manta rays, and sharks. These models are then enclosed in "deformation cages," which allow researchers to manipulate them and create new forms. The AI system tests these shapes in a physics simulator, evaluating their performance at different angles of attack—how the glider tilts while moving through water. The goal is to achieve the highest lift-to-drag ratio, where lift helps the glider move upward and drag reduces its backward motion. A higher ratio means the glider can travel more efficiently, expending less energy. The team used their AI-driven approach to fabricate two novel glider designs via 3D printing. One resembled an airplane with two wings, while the other had a unique, flat shape with four fins, akin to a fish. Both designs were hollow shells with small holes for water intake, making them lightweight and easy to handle. Inside, a tube-like structure housed the necessary hardware, including a buoyancy pump, a mass shifter to control the angle of attack, and electronic components. During testing, both AI-generated gliders outperformed a manually crafted torpedo-shaped glider. They moved more efficiently through the water, using less energy, which is akin to how marine animals navigate their environment with minimal effort. This improvement in energy efficiency could extend the operational range of autonomous gliders, enhancing their utility in oceanographic research. Moving forward, the researchers aim to bridge the gap between simulation and real-world performance. They plan to develop gliders capable of adapting to sudden changes in ocean currents, making them more versatile and resilient. Additionally, they are exploring thinner designs and methods to speed up the framework, potentially enabling more customization and maneuverability. The implications of this research are significant for the field of marine robotics. By leveraging AI to innovate and optimize glider designs, researchers can gather more comprehensive and detailed data about the ocean, advancing our understanding of marine ecosystems and climate change. The semi-automated design process also reduces the time and cost associated with creating new glider models, making such projects more feasible and scalable. Peter Yichen Chen, a postdoc at MIT CSAIL and co-lead researcher on the project, notes, "Our method opens up a whole new realm of possibilities for glider design, allowing us to test unconventional shapes that would be challenging for humans to design. This level of shape diversity hasn’t been explored much, and we are excited to see how these designs perform in real-world conditions." The research, supported by a Defense Advanced Research Projects Agency (DARPA) grant and the MIT-GIST Program, is a collaborative effort involving MIT CSAIL and the University of Wisconsin at Madison. The team includes CSAIL researchers Peter Yichen Chen, David Hagemann, and Pingchuan Ma, along with University of Wisconsin assistant professor Wei Wang, and MIT professors Daniela Rus and Wojciech Matusik. Industry insiders are optimistic about the potential of this AI-driven design process. They believe it could revolutionize marine exploration by facilitating the development of more efficient and adaptive gliders. This approach not only enhances the performance of existing technologies but also paves the way for new applications in environmental monitoring and oceanography, underscoring the growing role of AI in advancing scientific research. MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) is renowned for its cutting-edge research in AI and robotics, consistently pushing the boundaries of technology to solve complex real-world problems. The University of Wisconsin at Madison, known for its strong engineering programs, contributed valuable expertise in fluid dynamics and simulation, making the collaboration a powerful synergy between two leading institutions.