Path Tracking Control of Hybrid-driven Robotic Fish Based on Deep Reinforcement Learning

Abstract

Hybrid-driven robotic fish (HRF) is a new type of marine robot with long endurance. With the development of artificial intelligence and deep reinforcement learning, hybrid-driven robotic fish are becoming more intelligent and autonomous. This article is based on autonomous learning and autonomous decision-making control technology, drawing on human learning and decision-making processes, so that the aircraft can accumulate past control experience in a complex marine environment, acquire knowledge, and constantly improve its own performance and adaptability to achieve path following purpose. Firstly, the movement pattern of HRF was analyzed, and then the process of Deep Reinforcement Learning was analyzed. In addition, the Deep Reinforcement Learning method was improved based on the HRF movement pattern, and a pool experiment was performed. The experimental results show that the accuracy of the HRF path following control phase based on Deep Reinforcement Learning is improved by about 3.79%, compared with the traditional PID control method. It also indicates the Deep Reinforcement Learning control method has a better path following ability. Furthermore, it is of great significance to the swarm HRFs control and application.