AI-Driven Detection and Control in a Bioinspired Robotic Fish

This paper presents a bioinspired robotic fish adaptable to diverse applications, integrating artificial intelligence for real-time recognition of marine objects and a biomimetic locomotion system based on a 2 Degrees of Freedom (DoFs) tail. The onboard neural network processes visual data to identify specific underwater objects, such as mussels and clams. At the same time, a real-time communication system transmits findings remotely via a Telegram-based API. Performance evaluation, including accuracy, loss, confusion matrix, and Receiver Operating Characteristic-Area Under the Curve (ROC-AUC) analysis, confirms the model’s reliability in classification tasks. Additionally, key operational metrics such as waterproof integrity, response time, manoeuvrability, and regime speed were measured, demonstrating its practicability. The proposed system offers a versatile and cost-effective solution suitable for applications in marine biology, underwater monitoring, and autonomous exploration.