FluidTac: A Vision-Based Arrayed Artificial Lateral Line Sensor for Underwater Ego-Motion Estimation

Abstract

Underwater ego-motion estimation is crucial for achieving autonomous navigation in Autonomous Underwater Vehicles (AUVs). Obtaining continuous and robust velocity observations in shallow, acoustically and visually constrained environments remains an open problem, especially for compact platforms where acoustic bottom-tracking and vision-based methods are prone to degradation. To this end, this letter proposes a vision-based arrayed Artificial Lateral Line (ALL) sensor named FluidTac, capable of obtaining reliable odometry in acoustically constrained shallow waters. In terms of sensing mechanism, the system utilizes a circularly distributed array of passive propellers to translate local flow into rotation. A built-in camera tracks their angular velocities to estimate the AUV's motion. We employ a neural network that regresses the body's forward velocity, which is then fused with inertial data to achieve robust ego-motion estimation. Underwater experiments demonstrate FluidTac's accuracy in perceiving flow velocity and direction. To validate the system's performance, we integrated FluidTac into an AUV for real-world lake experiments. The results indicate that the system maintains reliable state estimation in shallow, turbid, and unstructured environments, without relying on acoustic bottom-tracking or visual features. This provides a low-cost and high-efficiency solution for AUV ego-motion estimation.