Engineers from EPFL and New York University (NYU) have presented ScaFi (for ‘Scalable Fish’), a robotic fish whose design can be adapted to different sizes. The work, published in the journal npj Robotics, aims to reduce the effort required to build swimming robots suited to varied environments.
Propellers remain commonly used for exploring and monitoring aquatic environments, but they have limitations. Their blades can snag on vegetation, stir up sediment, and scare the wildlife they are meant to study, making them poorly suited to small streams, dense seagrass beds, or close-up fish observations. This is one reason why roboticists have for several years been designing machines that swim like fish, by undulating their bodies rather than spinning a propeller. The problem is that most of these robots are designed for a specific size and task, so changing scale typically means starting from scratch.
ScaFi is inspired by fish like cod and mackerel, which concentrate most of their body flexion towards the tail end. This swimming style is found in nature across species of very different sizes. The robot has a rigid front section and a flexible tail made of fiberglass rods. A single motor pulls on two tendons, a kind of cable that cross near the tail tip, producing the ‘S’ curve necessary for swimming.
A tail that adapts to size
According to the researchers, only the diameter of the rods forming the tail needs to change with the robot’s size. The larger the robot, the thicker these rods, in order to maintain a similar flexing behavior. The motor mechanism and the crossed-tendon system remain identical. Professor Nana Obayashi, first author of the study, currently an assistant professor at the NYU Center for Robotics and Embodied Intelligence, led this project during her PhD in the CREATE lab at EPFL, headed by Josie Hughes. She summarizes the challenge: ‘Today, if you want to monitor a stream and then a lake, you practically need two different robots, designed and tested from the beginning. The environments we are interested in are not all the same size, so we don’t think the tools should be either.’
The researchers believe this approach could reduce the engineering effort needed to manufacture robotic fish suited to different environments. It also offers a platform for studying how swimming performance evolves with size, a question difficult to address systematically, both in animals and in custom-built robots.
Open-water trials
The team built three robots measuring approximately 0.6, 1.1, and 2.9 meters long and evaluated their swimming. In collaboration with the Unsteady Flow Diagnostics Lab at EPFL, led by Karen Mulleners, the researchers found that the smallest robot produced vortices similar to those left by real fish. For all three sizes, the swimming motion closely matched once adjusted for body size, which, according to the authors, suggests the scaling approach preserved the fish-like gait as the robots grew almost fivefold in length.
The robots were also tested in the field: the medium model in a Swiss stream, the largest in Lake Geneva, and the smallest in streams 15 to 30 centimeters deep. During the stream trial, the robot continued swimming after a GPS signal loss.
Energy efficiency proved more difficult to scale. The two smallest robots achieved similar results, but the largest was consistently less efficient and required a different, more powerful motor. The authors mention drag and inertia, though the exact cause is not established. In disturbance tests, the smallest robot proved the most agile, but was the slowest to regain its trajectory after being pushed off course, while the larger ones were less nimble but more stable. The researchers note that the same method could apply to other soft robots, including outside water. Whether energy performance can be scaled as well as swimming motion remains, according to the study, an open question.
