image: (a) Bionic inspiration: superficial neuromast. (b) Structure of the ALL sensor. (c) Overall view and cross-section of the ALL sensor. (d) Schematic Diagram of the ALL sensor’ principle. The diagram illustrates the state changes of the sensor before and after force application. The selected cross-section corresponds to the sectional view in Fig.1(c). (e) Physical diagram of the sensor. The diagram shows some modules of the sensor, with a silicone spring inside the sensor. (f) Comprehensive overview of the ALL sensor research, including load testing and velocity estimation experiments using the proposed CLANN algorithm.
Credit: ©Science China Press
Fish can navigate murky water without relying solely on vision. Along their bodies, the lateral line system detects subtle hydrodynamic changes and helps them respond to the surrounding flow. Inspired by this biological sensing strategy, researchers at the Institute of Automation, Chinese Academy of Sciences, developed a lightweight and miniaturized flexible artificial lateral-line sensing unit that allows an underwater robot to estimate both the magnitude and direction of local flow velocity.
Underwater robots are increasingly used in marine resource exploration, environmental monitoring and ecological conservation. Yet flow sensing remains challenging in complex aquatic environments. Acoustic devices can be bulky and sensitive to operating conditions, while vision-based systems may lose reliability in turbid water. Artificial lateral lines offer a complementary route by sensing the flow field close to the robot body.
The newly developed sensor uses a dual-layer inverted cup-shaped rigid rocker, a flexible silicone spring, a light-emitting diode and four photodiodes integrated within one compact sensing unit. When water flow pushes the rocker, the silicone spring deforms and the rocker tilts. This changes how much internal light reaches each photodiode. The resulting pattern of luminous-flux variation contains information about flow velocity and direction. The integrated structure enables vector sensing without relying on a sensor array, while retaining a lightweight and miniaturized design suitable for underwater robotic platforms.
Flexible materials improve sensitivity to weak disturbances, but their large deformation, nonlinear mechanics and coupled responses make calibration difficult. To address this challenge, the researchers proposed a hybrid neural network algorithm named CLANN. The algorithm calibrates the artificial lateral line sensor and can fuse its data with measurements from an inertial measurement unit.
The team integrated the sensor into a remotely operated underwater vehicle and tested it during linear, circular and irregular motions. Rather than limiting validation to idealized conditions, the experiments considered attitude disturbances such as pitch and yaw during multiple motion states, more closely reflecting practical underwater operations. By fusing data from the artificial lateral line and an inertial measurement unit, the system estimated the robot's global velocity vector. Across the reported experiments, the velocity magnitude had a mean absolute error of 0.048 m/s, the direction had a mean absolute error of 16.49 degrees and the linearity coefficient R squared reached 0.896. The estimated trajectories had a mean absolute position-tracking error of 0.284 m.
The experiments also indicate that the sensor can operate over a velocity range of 0 to 0.4 m/s. Its combination of single-unit vector sensing, lightweight construction, miniaturized size and validation under realistic disturbance conditions may support practical flow perception for underwater robots and biomimetic robotic fish.
The researchers plan to refine the sensor structure and perception algorithm, integrate the device with a robotic fish platform and conduct validation in natural water bodies. In the longer term, artificial lateral lines could work alongside other sensors to support flow-field mapping and underwater localization when conventional velocity sensors are constrained.
The study was supported by the National Science and Technology Major Project of China, the National Natural Science Foundation of China, the Natural Science Foundation of Beijing Municipality and the Beijing Nova Program.