When studying aquatic marine organisms under water, it can be very helpful if you can use a device that will not scare animals by acting unnatural to them. MIT scientists have created a robotic starfish with that in mind, using a new fast-growing system.
Due to variables such as flow, salinity, and buoyancy, designing underwater robots is often much more difficult than designing terrestrial robots. It is therefore not uncommon to go through multiple prototypes, each incorporating the features its predecessor lacked. Not only is this process expensive, but it can also take weeks or longer to get the finished product.
When a team led by MIT professors Wojciech Matusik and Daniela Rus decided to build a robotic mollusk starfish that could swim, they created a simulation system based on machine learning to speed development. . Based on the scientists’ request, the system created a computer model that illustrates how such a robot can be constructed and how it swims.
Matusik, Rus and colleagues quickly built an early prototype, based on that model. When that robot is tested in a water tank, its real-world performance data is fed back to the computer model, helping to optimize it even further. As a result, the team was able to produce a functional product within hours and only a few prototypes were created during implementation.
The current version of the robotic starfish has a soft silicone body, along with a single low-power motor connected to the tendons on each of the robot’s four legs. By alternately squeezing and releasing those legs, the robot can swim in the water gently and effectively. And according to the researchers, the computer model provided important design considerations that humans may have overlooked.
Postdoctoral fellow Josephine Hughes, co-author of the research paper with graduate student Tao Du.
Finally, the robot can be powered by onboard batteries and equipped with sensors like cameras to collect oceanographic data. And it could soon be joined by robotic sea turtles, rays and sharks, all of which are planned to be developed through the new simulation system.
You can see it in action in the video below.
Optimizing underwater soft robots with machine learning