Azimuth Estimation of Swimming Fish by Artificial Lateral Line System

Localization and sensing in the underwater environment has been a significant but difficult problem in the field of robotics. The lateral line system (LLS) is an essential component of underwater localization and sensing for most fish. Inspired by the LLS, scientists have designed varieties of artificial lateral line systems (ALLSs) and researched their applications. This paper investigates on using ALLS for azimuth sensing in a wide range of distance and angle, even in all directions, between two robotic fish. Specifically, we consider the scenario when a robotic boxfish with ALLS stays still and uses its ALLS to obtain the flow signals, while a carangidae robotic fish swims around it freely. To estimate the azimuth of the swimming robot, first, the fast Fourier transform (FFT) and discrete wavelet transformation (DWT) are adopted to pre-process the ALLS’s data and analyze the quantity of the data. Then, two analysis model methods are proposed, that is, phase difference-based method and intensity-based method. Under the guidance of the above two methods, we adopt a recurrent neural network (RNN) to estimate the azimuth. The experimental results show that bases on ALLS and RNN, we are able to estimate the azimuth of swimming robotic fish in a wide range of around 7 times the size of the sensor array and in all directions.