A robot fish holds position in a turbulent lab flow tank without a current sensor

A robot fish held its position in a turbulent lab flow tank using only what its own body felt, with no current sensor on board, according to a preprint from a ShanghaiTech-led team.

The team, with collaborators in China, trained a body-and-tail swimming robot in a software framework called SWiFT (Swimming With Flow Toolbox) that pairs a computer simulation of the water with a physical flow tank. The control policy was learned in simulation, then transferred to the real robot, a pipeline the field calls sim-to-real.

The team's policy closes the loop on body-only feedback, a property the authors call egocentric sensing, and was tested in turbulent flows whose velocity was unknown to the robot. According to the paper, the policy achieved "substantial improvements over state-of-the-art methods" on average positioning error (RMSE), though the abstract does not give a number, and the comparison is author-reported rather than independently validated.

The biological parallel is rheotaxis, the way real fish orient into a current. The authors frame the work as a step toward real-world deployment in natural aquatic environments, without exotic instrumentation, though the demonstrated scope is one preprint, one lab flow tank, and a peer-review status that is not yet established.