Advancements in Human-Robot-Computer Research

The automated experimental facility, called the Intelligent Towing Tank (ITT), conducted around 100,000 total experiments in its first year of operation. What would normally take a PhD student to complete within five years of experiments, the ITT was able to do within weeks. The development of the ITT in the MIT Sea Grant Hydrodynamics Laboratory takes us further into the area of human-robot-computer research. 

The ITT automatically and adaptively performs, analyzes, and designs experiments. The experiments are focused on exploring vortex-induced vibrations (VIVs). VIVs are important for engineering offshore ocean structures such as marine drilling risers, which are responsible for connecting underwater oil wells to the surface. With VIVs, there are a high number of parameters involved.

The ITT is guided by active learning, and it conducts a series of experiments. Within the experiments, the parameters for each next experiment are selected by a computer. The system uses an “explore-and-exploit” methodology, which helps greatly reduce the number of experiments required for mapping and exploring the complex aspects of VIVs.

PhD candidate Dixia Fan began the project while searching for a way to reduce the thousand or so experiments that needed to be conducted by hand. That led to the development of the ITT system. 

A paper was published last month in the journal Science Robotics. 

Fan is now a postdoc, and the project was worked on by a team of researchers from the MIT Sea Grant College Program and MIT’s Department of Mechanical Engineering, École Normale Supérieure de Rennes, and Brown University. The new project showcases the type of cooperation that can take place between humans, computers, and robots in order to make scientific discoveries at a faster pace.

The ITT is a 33-foot tank, and it works without interruption or suspension. The researchers would like to see the system used in a variety of different disciplines, which could lead to the creation of new models in nonlinear systems. 

The ITT allowed Fan and his collaborators to explore a wider parametric space. “If we performed traditional techniques on the problem we study, it would take 950 years to finish the experiment,” he explained. 

In order to shorten the long time it would take for the experiment, Fan and the team integrated a Gaussian process regression learning algorithm into the ITT. By doing this, the researchers were able to reduce the amount of experiments needed, down to a few thousand. 

The robotic system is capable of automatically conducting an initial sequence of experiments. It then takes partial control over the parameters of the next experiment. 

Fan was awarded an MIT Mechanical Engineering de Florez Award for “Outstanding Ingenuity and Creative Judgement” in the development of the ITT. 

According to Michael Triantafyllou, Henry L. and Grace Doherty Professor in Ocean Science and Engineering, and also Fan’s doctoral advisor, “Dixia’s design of the Intelligent Towing Tank is an outstanding example of using novel methods to reinvigorate mature fields.”

Triantafyllou was a co-author on the paper and the director of the MIT Sea Grant College Program. 

“MIT Sea Grant has committed resources and funded projects using deep-learning methods in ocean-related problems for several years that are already paying off,” he said.

MIT is funded by the National Oceanic and Atmospheric Administration and administered by the National Sea Grant Program. It is a federal-institute partnership that combines research and engineering at MIT to help tackle ocean-related issues, 

Other contributors to the paper include George Karniadakis from Brown University, affiliated with MIT Sea Grant; Gurvan Jodin from ENS Rennes; MIT PhD candidate in mechanical engineering Yu Ma; and Thomas Consi, Luca Bonfiglio, and Lily Keyes from MIT Sea Grant.