This Robotic Starfish Lets Biologists Study Marine Life

MIT CSAIL roboticists have developed an effective technique for rapidly designing, fabricating, and refining soft robots.

It can be difficult to study animals without our observation affecting their behavior. At one extreme end of the spectrum, you have scientists studying animals in captivity, such as traditional lab rats and mice. Behavioral information gathered in the lab can’t be used to reliably predict behavior in nature. At the other end of the spectrum, you have animals living in their natural habitats far from any human infrastructure, equipment, or involvement. The goal for many researchers is to get as close to that side of the spectrum as possible, while still being able to observe the wildlife in question. That’s why a team of roboticists from MIT CSAIL have created a robotic starfish that could let biologists unobtrusively study marine life.

A team from MIT CSAIL (Computer Science and Artificial Intelligence Lab) previously built a soft robotic fish that was used to study real fish in Fiji back in 2018. That allowed for some interesting capabilities, but needed a relatively large amount of power to operate and required some human oversight. This new starfish design is also a soft robot, but it is much more efficient. It can stay still for long periods of time in order to covertly observe marine life and can use its unique method of propulsion to move when needed. While some species of starfish do swim, most simply crawl along on hundreds of tiny “feet.” This robotic starfish, however, is made to swim. It is the methodology used to fabricate the robot and teach it how to swim that makes this project particularly useful.

The body of the robotic starfish was made from soft, flexible silicone foam, which is naturally buoyant and that can be fabricated in just a few hours. The robot has four legs and they are all actuated by a single servo motor. That can’t provide fast or granular control, so the team turned to simulations and machine learning to train the robot to swim efficiently. An initial simulation was created using parameters that were little more than guesses made by the team after manually controlling the robot. The machine learning model then observed the real world results to see how well they matched that simulation. Over the course of four rounds of simulations, the robot’s swimming technique was refined using lessons learned from the real world observations. After going through those simulations, the robot was able to swim four times faster than it could when it was controlled manually. These techniques could help roboticists rapidly design and refine soft robots in the future, including robots like this that can be used to monitor animals without disturbing them.

Cameron Coward
Writer for Hackster News. Proud husband and dog dad. Maker and serial hobbyist.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles