Roughly 74 percent of the ocean floor has never been properly explored. The robots built to survey it are largely the same as the ones built 40 years ago: rigid, propeller-driven machines that suppress contact with the environment rather than using it.
Meanwhile, octopuses squeeze into cracks, fish harvest energy from water swirling around them, and remoras hitchhike on sharks for hours using a suction disc made from a modified fin. The gap between what sea creatures can do and what underwater robots can do has never been wider.
A team at Peking University and the Chinese Academy of Sciences thinks the problem is that bioinspired robotics has been running in the wrong direction.
Their paper, published in npj Robotics on 23 April 2026, makes an argument that sounds simple but represents a genuine shift. The standard approach is one-way: observe an animal, extract a principle, build a robot that copies it. The team argues this leaves most of the value on the table. Robots should also run experiments on biology, probing mechanisms in living creatures that are too difficult to study directly. The flow of insight should go both ways.
This matters more than it sounds.
Marine biology has a fundamental measurement problem. When you want to understand how a fish generates thrust, how an octopus coordinates its arms, or how a remora’s suction disc holds through turbulent currents, you are trying to isolate a single variable in a system that is constantly moving and adapting in ways you cannot control. You can observe, but you cannot run a controlled experiment on a living animal the way you can in a laboratory.
A soft robot built to replicate a specific biological structure can. You change the fin stiffness, run the robot, measure the thrust, change it again. You build a robotic remora disc, vary the geometry, test whether adhesion holds under turbulent conditions.
You can even build physical models of extinct creatures and test hypotheses about how they moved through the water. Extinct animals cannot be observed. Controlled physical models of them can be built and tested.
The team calls this a bidirectional framework. Biology informs robotic design; robots inform biological understanding. Alongside it, they propose a “biouniversal design strategy” — extracting principles that appear across multiple marine organisms rather than copying any single one. A mechanism that appears in both shark skin and fish lateral line sensing is more likely to be fundamental than one found in only one species.
The practical engineering consequences fall into two areas.
The first is ocean exploration. Roughly 74 percent of the ocean floor remains unmapped at the resolution needed for mineral surveying, ecosystem monitoring, or infrastructure inspection.
Rigid vehicles are too fragile and tactically limited for complex close-range work. Soft robots that can grip, probe, sample, and navigate cluttered environments using physical intelligence built into their materials would change what is accessible. The paper surveys dozens of existing hardware demonstrations across locomotion, manipulation, sensing, and adaptive control.
The second is medicine. Soft robots that navigate narrow, irregular, compliant spaces are directly relevant to minimally invasive surgery. The same design principles that allow a soft robot to move through coral without damaging it apply to navigation through biological tissue. Distributed pressure sensing derived from fish lateral lines is already being adapted for catheter guidance and internal monitoring.
The limitations are real. This is a review paper, not a hardware demonstration of the bidirectional framework itself. The “robots inform biology” loop requires systematic collaboration between roboticists and marine biologists that is not yet standard practice. Building that infrastructure is a harder problem than building the robots.
What the paper establishes is the conceptual architecture for how that collaboration should work, backed by a survey of hardware that makes it feasible now. The ocean exploration, materials science, and medical robotics communities have a clearer design vocabulary for what comes next.
Sources
Li, L., Qin, B., Gao, W., Li, Y., Zhang, Y., Wang, B., Kong, S., Wang, J., He, D. and Yu, J. “Bioinspired underwater soft robots: from biology to robotics and back.” npj Robotics 4, Article 25 (2026). DOI: 10.1038/s44182-026-00088-x https://www.nature.com/articles/s44182-026-00088-x Published: 23 April 2026. Open access. Affiliations: Peking University (College of Engineering, Institute of Ocean Research); Chinese Academy of Sciences (Institute of Hydrobiology, Institute of Automation); National University of Singapore.
More Like This
Humanoid robots have always needed a different brain for each way of moving. That just changed.
Jane holds a BSc in Biology from the University of Regina and a Master of Science in Bioscience, Technology and Public Policy from the Univesity of Winnipeg. Her reporting interests include Life Sciences, Physical Sciences and the Cosmos.