DeepFlow-MuJo: A hierarchical simulation platform for integrating the “Musculoskeletal-Body-Behavior” of aquatic organisms

The adaptive behavior of aquatic organisms can be understood as a form of embodied intelligence, arising from the dynamic coupling of organ-level neuromuscular actuation, body-level fluid–structure interactions (FSI), and macro-level behavioral strategies. To elucidate the mechanisms underlying high-efficiency swimming and to inform biomimetic engineering, it is essential to analyze the entire dynamic chain — neural excitation, muscle contraction, skeletal transmission, and fluid response — on the basis of first principles. However, existing research paradigms are often limited to isolated experiments on muscle actuation, kinematic simulations of the body or fins, or statistical analyses of behavioral traits, and therefore overlook the real-time reciprocal feedback across these multiple hierarchical levels. To address this, we propose DeepFlow-MuJo, a hierarchical intelligent simulation platform integrating “musculoskeletal-body-behavior” levels. This platform constructed a digital organism in a hydrodynamic environment by coupling a musculoskeletal multibody dynamics module with a FSI module based on the Immersed Boundary Lattice Boltzmann Method (IB-LBM). In addition, it incorporated a deep reinforcement learning (DRL)-based hierarchical decision-making module that autonomously controlled the entire dynamic chain, thereby mimicking the generative logic of real biological behaviors. The versatility of DeepFlow-MuJo was demonstrated by simulating diverse scenarios involving aquatic organisms and biomimetic robots: (1) the passive self-propulsion of a dead fish in a Kármán vortex street; (2) the active swimming and target tracking of a moon jellyfish; (3) the performance of a biomimetic fish under different actuation modes; and (4) the adaptive behavior of fish downstream of a hydropower station. The results show that DeepFlow-MuJo not only reproduces force generation and swimming in a physiologically realistic manner, but also provides a novel research tool for elucidating the relationship between kinematics and dynamics in aquatic organisms and for inspiring control strategies for next-generation biomimetic robots.