Estimating the Lateral Motion States of an Underwater Robot by Propeller Wake Sensing Using an Artificial Lateral Line

TL;DR

This work investigates sensing the wake of a high-speed propeller using a low-cost artificial lateral line to estimate lateral motion states for leader–follower underwater formations. It develops a data-driven pipeline combining a 1D CNN and BiLSTM to extract spatiotemporal wake features and a multi-output estimator that jointly regresses lateral displacement $x$ and classifies speed $v$ and direction $d$, with task weights optimized by the Whale Optimization Algorithm. Experimental results on a custom testbed show that the approach achieves low RMSE for $x$ (≈0.05–0.07 m) and high accuracies for $v$ (>91%) and $d$ (>99%), outperforming short-time Fourier transform and CNN baselines. The findings demonstrate the feasibility and advantages of ALL-based wake sensing for propeller-driven underwater robots in real-time estimation scenarios, while acknowledging limitations and proposing avenues for improved sensing and fusion with other sensors.

Abstract

The artificial lateral line (ALL), comprising distributed flow sensors, has been successful in sensing motion states of bioinspired underwater robots like robotic fish. However, its application to robots driven by rotating propellers remains unexplored due to the complexity of propeller wake flow. This paper investigates the feasibility of using ALL to sense propeller wake for underwater robot leader-follower formation. To estimate the lateral motion states of a leader propeller, this paper designs a multi-output deep learning network that extracts temporal and spatial features from distributed pressure measurements of propeller wake. Extensive experiments are conducted on a designed testbed, the results of which validate the effectiveness of the proposed propeller wake sensing method.

Figures (8)

• Figure 1: Illustration of ALL sensing the propeller wake of a leader underwater robot in a leader-follower formation (Top View).
• Figure 2: The design of the testing platform that comprises a water tank, a sliding guide, a leader propeller, and an ALL. The water tank measures 3.0m(L) x 2.0m(W) x 1.5m(H). The ALL consists of three pressure sensors distributively positioned with an angle of $45^\circ$ between each adjacent pair at the same horizontal level from the bottom of the cylindrical shell.
• Figure 3: The trajectories of pressure sensor data collected by the ALL in the experiment when the leader propeller moves laterally at different speeds.
• Figure 4: The schematic of the network architecture of the proposed wake flow estimator that consists of a hybrid CNN-BiLSTM network for flow feature extraction and a multi-output neural network for motion state regression and classification.
• Figure 5: The experimental results of the motion state estimation based on propeller wake sensing including the regression results on the lateral displacement $x$, and the classification results on the traveling speed $v$ and the direction of motion $d$.