Study on the Improved Water Flow Prediction Based on Classification-Regression Approach for Amphibious Spherical Robots

Inspired by the lateral line systems found in aquatic organisms, an artificial lateral line system (ALLS) for underwater robots was developed. This technology was integrated into Autonomous Underwater Vehicles (AUVs) to enhance their environmental awareness and navigation. This paper was focused on improving water flow prediction performance for a amphibious spherical robot (ASR) equipped with a ALLS comprising 12 pressure sensors. To address the challenge of the initial method relying on polynomial fitting based on data from a single sensor, which encounters difficulties when the flow direction falls between sensor angles. Two strategies were explored: increasing the quantity of sensors and implementing a Classification-Regression approach. Experimental simulations using Computational Fluid Dynamics (CFD) software were conducted to investigate the effectiveness of these strategies. Experimental results indicated that the classification-regression method significantly enhanced prediction accuracy compared to the original approach and offered a cost-effective alternative to increasing sensor quantity. This research contributed to the advancement of underwater robotics by enhancing perception capabilities, which were crucial for navigating challenging underwater environments efficiently.