Efficient Cooperative Structured Control for a Multijoint Biomimetic Robotic Fish

In this article, we propose an efficient locomotion control method for a two-dimensional tracking task of a biomimetic four-joint robotic fish. Regarding this issue as a comprehensive optimization procedure, we propose an optimization-based cooperative structured control framework, in which the combination of evolutionary strategy and deep deterministic policy gradient is employed to optimize the same objective function. An inconsistent optimization method is presented to further enhance the effect of parameter optimization on central pattern generator model. Moreover, for the sake of a higher reward and better robustness of controllers governed by deep reinforcement learning, we propose a linear weighted controller trained with periodic method. Extensive simulation and experimental results verify the significant energy saving of the proposed method in tracking tasks. Noticeably, the cooperative structured control can save 23.97%, 22.13%, and 38.72% energy compared with sliding mode control, active disturbance rejection control, and proportional-integral-differential control in experiments, respectively, holding a great promise for the long-term intelligent work of the biomimetic robotic fish in complex aquatic environments.

Automated summary

In this article, an efficient locomotion control method for a biomimetic robotic fish in tracking tasks is proposed, utilizing an optimization-based cooperative structured control framework and a linear weighted controller trained with periodic method for deep reinforcement learning. Extensive simulation and experimental results show significant energy savings, with the cooperative structured control outperforming sliding mode control, active disturbance rejection control, and proportional-integral-differential control in terms of energy efficiency.