Three-Dimensional Path Following Control of an Underactuated Robotic Dolphin Using Deep Reinforcement Learning

摘要

In this paper, a novel improved deep reinforcement learning (DRL) strategy is utilized to solve the three-dimensional path-following control problem for an underactuated robotic dolphin in order to meet the complex and intractable hydrodynamic model. Firstly, a brief overview of the developed robotic dolphin consisting of two-degree-of-freedom paired pectoral fins and the slider-crank-based flapping two-joint tail is introduced. After that, path-following controller includes a novel lookahead-based guidance law, decoupling motion strategy, and a deep deterministic policy gradient algorithm based on the the prior knowledge is proposed. The novel lookahead based 3-D LOS guidance law is employed to transform the 3-D waypoints to the desired heading and pitch angles with its simplicity, intuitiveness, and small computational footprint. Finally, the effectiveness of the proposed controller is demonstrated after numerous simulations. It will offer a rich vein of insight for the real-time task execution in bioinspired underwater robots.