Adaptive navigation control of a bionic robotic fish in complex Karman vortex street flow fields using an LSTM-DDPG hybrid strategy

Bionic robotic fish suffer from poor navigation robustness, high energy consumption, and control lag in unsteady Karman vortex street flow fields. This study proposes a hybrid adaptive navigation strategy combining long short-term memory (LSTM) and deep deterministic policy gradient (DDPG) to achieve autonomous, efficient, and stable motion control. The method constructs a closed-loop framework of local flow field perception, temporal prediction, and continuous flexible control, eliminating dependence on pre-built flow field models. A Karman vortex street simulation platform is developed using the immersed boundary-lattice Boltzmann method (IB-LBM), and a multi-level reward function is designed to balance accuracy, stability, and energy efficiency. Numerical simulations and physical prototype experiments are conducted under Reynolds numbers Re = 500, 800, and 1000, with comparisons to PID, DQN, and MPC. Results show that the LSTM-DDPG strategy significantly improves navigation precision and anti-disturbance ability while reducing energy consumption. The average task completion rate reaches 88.3%, and average energy consumption is 5.2 J/m, which is 45.3% lower than conventional PID control. This method provides a feasible solution for robust and energy-efficient navigation of bionic robotic fish in complex ocean environments.